13
$\begingroup$

Let $K$ be a finite field, and $\overline{K}$ its algebraic closure. It is well known that two curves are isomorphic over $\overline{K}$ if and only if they have the same $j$-invariant. If two such curves are also $K$-isogenous, I believe we can conclude that they are $K$-isomorphic, but I cannot find any reference or elementary proof of this fact; is it easy to see, or does anyone have a reference? (it seems that this result is implicitly used in the algorithms for isogeny graphs of elliptic curves, like Kohel's, where the isogenous curves are encoded by there $j$-invariant).

As a side question, given a $j \in K$, how many curves over $K$ have this $j$-invariant, up to $K$-isomorphisms? For example, a (non-supersingular) curve and its quadratic twist both have the same $j$-invariant, but are not $K$-isogenous, so not $K$-isomorphic, so we have at least two $K$-isomorphism classes; is this it?

$\endgroup$
2
  • 3
    $\begingroup$ For the second, look up "twists" in chapter X of Silverman's book. Together with the knowledge of what the endomorphism ring is yo ushould get the answer. $\endgroup$ Oct 24, 2014 at 11:14
  • $\begingroup$ So if $K$ has characteristic $p$ then since $E$ and its Frobenius twist $E^{(p)}$ are $K$-isogenous yet have the same $j$-invariant, you are claiming that $E$ and $E^{(p)}$ are always $K$-isomorphic in such cases? $\endgroup$
    – user27920
    Oct 24, 2014 at 13:30

3 Answers 3

16
$\begingroup$

As Chris and Rene said, your second question is all about twists. See specifically The Arithmetic of Elliptic Curves (Springer), Proposition X.5.4, which says that if the characteristic of $K$ is not 2 or 3, then the twists of $E/K$ are in one-to-one correspondence with $K^*/(K^*)^n$, where $$ n=\begin{cases} 2&\text{if $j(E)\ne0,1728$},\\4&\text{if $j(E)=1728$},\\6&\text{if $j(E)=0$}.\\\end{cases}$$

So for a finite field $K=\mathbb{F}_q$ of characteristic $p\ge5$, in the first case there are always exactly two distinct twists, but for the latter two cases, the number of twists depends on whether $K$ has a fourth root, respectively cube root, of one, or equivalently, on the value of $q\bmod4$, respectively $q\bmod3$.

$\endgroup$
10
$\begingroup$

The first question seems to have been neglected. If two curves have the same $j$-invariant and are isogenous over $K$ (a finite field with $q$ elements), are they isomorphic over $K$? If they aren't, then they are twists of one another. If $j \ne 0, 1728$, the twist is a quadratic twist and if one curve has $q+1-a$ points, the other has $q+1+a$ points. They are $K$-isogenous if and only if they have the same number of points, so if and only if $a=0$. This case certainly happens. So answer to the first question is no, but only in very few cases. I am not sure what happens when $j=0, 1728$.

$\endgroup$
2
$\begingroup$

It is in general $\textit{not}$ true that $\textit{if}$ $A$ and $B$ are elliptic curves over the finite field $K$, lying in the same $K$-isogeny class, and with the same $j$-invariant, $\textit{then}$ $A$ and $B$ are $K$-isomorphic (see below). However, this is true if $A$ and $B$ are ordinary. You can find a related observation in Waterhouse's thesis (Abelian varieties over finite fields, Éc. Nor. Sup. 1969, remark page 542). Let me describe a different approach, similar to that suggested by Silverman.

I tried to be succinct in my answer, I failed as there are several details to verify. Sorry about this.

Let me slightly reformulate the question. Let $K$ be a finite field of size $q$ and characteristic $p\geq 5$. Let $E$ an elliptic curve over $K$, denote by $\pi_E:E\to E$ the Frobenius isogeny of $E$ relative to $K$. By Honda-Tate theory, $\pi_E$ defines a Weil $q$-number whose minimal polynomial over $\mathbf{Q}$ dentifies uniquely the $K$-isogeny class of $E$. Let $E'$ be a $K$-form of $E$, with associated Weil number $\pi_{E'}$. Your first question can then can be formulated as:

Can we have that $\pi_{E'}$ is conjugate to $\pi_E$ without $E'$ being $K$-isomorphic to $E$?

(Here by "conjugate" I mean that the $\pi_E$ and $\pi_{E'}$ have the same minimal polynomial over $\mathbf{Q}$.)

To answer the question let's start by describing the $K$-forms of $E$ in a slightly different (but equivalent!) way than that used by Silverman above. Set $W_K=\textrm{Aut}_K(E)$ and $W_{\bar K}=\textrm{Aut}_{\bar K}(E\times_K\bar K)$. Since $p\geq 5$, $W_{\bar K}$ is isomorphic, as a $G_K$-module, to $\mu_n$, where $n=2,4$ or $6$. Moreover, we can think of $W_{\bar K}$ as the roots of unity inside some imaginary quadratic field $F$ embedded in $\textrm{End}_{\bar K}(E\times_K\bar K)$, where $F$ is unique if $E$ is ordinary or if $n>2$.

The set of $K$-forms of $E$ is in bijection with the (group) $H:=H^1(G_K,W_{\bar K})$. Consider the two cases

i) $W_K=W_{\bar K}$;

ii) $W_K\subsetneq W_{\bar K}$.

In the first case the evaluation map $c\mapsto c(\varphi)$ of $1$-cocycles on the arithmetic Frobenius $\varphi\in G_K$ induces an isomorphism $H\simeq W_{\bar K}$.

In the second case the action of $G_K$ on $W_{\bar K}$ is by inversion, and the evaluation map $c\mapsto c(\varphi)$ as above induces an isomorphism $H\simeq W_{\bar K}/(W_{\bar K})^2$.

You can check that this description of $H$ matches the Kummer-theoretic one given above by Silverman (at least as abstract groups).

Now, if $c$ is a $1$-cocyle (for $G_K$ acting on $W_{\bar K}$), then Silverman in his book explains how to construct an elliptic curve $E^c$ over $K$ $\textit{and}$ an isomorphism $E^c\times_K\bar K\simeq E\times_K \bar K$. One can then show (I owe the omitted proof to Jakob Stix) that in the ring $\textrm{End}_{\bar K}(E\times_K\bar K)$ the following relation holds

$\pi_{E^c}=c(\varphi)\pi_E$,

where we make an implicit use of the identification $\textrm{End}_{\bar K}(E\times_K\bar K)\simeq \textrm{End}_{\bar K}(E^c\times_K\bar K)$ induced by the isomorphism $E^c\times_K\bar K\simeq E\times_K \bar K$ encoded by $c$.

In other words, the Weil numbers $\pi_{E^c}$ and $\pi_E$ differ by multiplication by a root of unity. If the cohomology class defined by $c$ is non-trivial, the above formula implies that the non-trivial twist $E^c$ and $E$ are $K$-isogenous if and only if

$\bar\pi_E=-\pi_E$,

where $\bar\pi_E=q/\pi_E$ is the complex conjugate of $\pi_E$. Notice that from what we saw above $c$ is non-trivial in $H$ iff [$c(\varphi)\neq 1$ in case i), and $c(\varphi)\not\in W_{\bar K}^2$ in case ii)].

This last condition on $\pi_E$ then amounts to require it being purely imaginary, hence conjugate to $\sqrt {-q}$. If $q=p^e$, then $\sqrt{-q}$ arises as $\pi_E$ for some elliptic curve $E$ iff [either $e$ is odd, or else $e$ is even and $p\not\equiv 1$ mod $4$]. (In the remaining cases where $e$ is even you get $K$-simple non-geometrically simple surfaces attached to $\sqrt{-q}$).

The end of the story is that if $e$ is odd or $e$ is even and $p\not\equiv 1$ mod $4$, then every elliptic curve $E$ in the $K$-isogeny class defined by $\sqrt{-q}$ admits $|H|$-many $K$-isogenous, pairwise non-isomorphic $K$-forms. All these are examples of supersingular elliptic curves ($p$ divides the trace of $\sqrt{-q}$) with not all geometric endomorphisms defined over $K$. They exhaust the instances of the phenomenon we were after.

$\endgroup$
1
  • $\begingroup$ For ordinary elliptic curves, this is Lemma 9.11.13 in "Mathematics of Public Key Cryptography. Version 2.0 Galbraith" $\endgroup$
    – Watson
    Nov 4, 2022 at 7:14

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.