Lemma 3.2.1 in Baker, González-Jiménez, González, Poonen, "Finiteness theorems for modular curves of genus at least 2", Amer. J. Math. 127 (2005), 1325–1387.
enter image description here
I don't understand why "all supersingular elliptic curves over $\bar{F_p}$ are isogenous".
Could anyone help me?
This problem arises from Supersingular elliptic curves and their "functorial" structure over F_p^2.
Moreover, I found some similar question: Isogenies between elliptic curves and their endomorphism rings and Isogenies between supersingular elliptic curves.
-
1$\begingroup$ I think that probably goes back to Deuring, but it certainly follows from Tate's theorem about isogeny classes and Tate modules. Since all supersingular elliptic curves have endomorphism algebras equal to maximal orders in the unique quaternion algebra ramified only at $p$ (and $\infty$), the associated rings $\mathbb{Z}_\ell\otimes \text{End}$ are isomorphic. Now you can apply Tate's theorem. $\endgroup$– Jason StarrCommented Apr 15, 2023 at 16:36
-
2$\begingroup$ Crossposted to MSE $\endgroup$– Viktor VaughnCommented Apr 15, 2023 at 22:11
-
1$\begingroup$ You can also see Theorem 9.6 of <arxiv.org/pdf/1704.00335.pdf> (and the description of this result on p. 2). This in fact shows that for any fixed $\ell\neq p$, any pair of supersingular elliptic curves over $\bar{\mathbb{F}}_p$ are $\ell$-primarily isogenous. $\endgroup$– RajuCommented Apr 15, 2023 at 22:12
2 Answers
This can be proved in several ways (this can be found in the literature as Lemma 42.1.11 in Voight's book on Quaternion algebras, or Proposition 5.2 in https://arxiv.org/pdf/2005.01537v1.pdf). Typically the proof relies on Tate's isogeny theorem (1966) which states, among other things, that if two elliptic curves over a finite field have the same number of rational points, then they are isogenous.
$$ \newcommand{\End}{\mathrm{End}} \newcommand{\Q}{\mathbb{Q}} \newcommand{\F}{\mathbb{F}} \newcommand{\Z}{\mathbb{Z}} \newcommand{\C}{\mathbb{C}} $$
Proposition. Let $p$ be a prime. Let $E,E'$ be two supersingular elliptic curves over $\F_{p^2}$. Then $E,E'$ are isogenous over $\F_{p^{24}}$.
(Note that the $j$-invariant of supersingular elliptic curves over $\overline{\F_p}$ belongs to $\F_{p^2}$, we may assume (up to $\overline{\F_p}$-isomorphism) that $E,E'$ are defined over $\F_{p^2}$).
Proof.
Write $\alpha, \beta \in \C$ for the eigenvalues of the $p^2$-Frobenius map on the $\ell$-adic Tate module of $E$ for some $\ell \neq p$, so that $|E(\F_{p^{2r}})| = p^{2r} + 1 - (\alpha^r + \beta^r)$ for every $r \geq 1$. Since $E$ is supersingular, its trace $t_1 := \alpha + \beta$ is $\equiv 0 \pmod p$. By Hasse bound we have $|t_1| \leq 2 \sqrt{p^2} = 2p$, so we have $t_1 \in \{ -2p, -p, 0, p, 2p \}$.
In each case, $\alpha, \beta$ can be computed explicitly as roots of a quadratic polynomial (characteristic polynomial of Frobenius):
If $t_1 = -2p$, then $X^2 + 2pX + p^2 = (X+p)^2 = 0$ yields $\alpha = \beta = -p$.
If $t_1 = -p$, then $X^2 + pX + p^2 = 0$ yields $\alpha,\beta = \frac{-p + \sqrt{p^2 - 4p^2}}{2} = p \cdot \dfrac{-1 \pm i \sqrt{3}}{2}$
If $t_1 = 0$, then $X^2 + p^2 = 0$ yields $\alpha = - \beta = ip$.
If $t_1 = p$, then $X^2 - pX + p^2 = 0$ yields $\alpha,\beta = p \cdot \dfrac{1 \pm i \sqrt{3}}{2}$
If $t_1 = 2p$, then $(X - p)^2 = 0$ yields $\alpha = \beta = p$.
In all cases, we observe that $\alpha^{12} = \beta^{12} = p^{12}$ (i.e. $\alpha,\beta$ are equal to $p$ times some $12$-th root of unity [in fact either of order $1,2,3,4$ or $6$]). This proves that $$ |E(\F_{p^{24}})| = p^{24} + 1 - (\alpha^{12} + \beta^{12}) = p^{24} + 1 - 2 \cdot p^{12} = |E'(\F_{p^{24}})| $$ so that $E,E'$ are isogenous over $\F_{p^{24}}$ by Tate's theorem. This concludes the proof. $\blacksquare$
Tate's isogeny theorem certainly suffices, but one can show much more. Consider the (multi)graph whose vertices are supersingular $j$-invariants mod $p$ and whose edges are isogenies of degree $2$. This family of graphs is not only connected, but it is Ramanujan [1, Prop. 3], meaning that its eigenvalue separation is as large as possible (the graph is as well-connected as possible). Therefore $2$-isogenies alone suffice to reach any supersingular curve from any other supersingular curve. The same statement holds for any other (prime) isogeny degree $\ell$ other than $p$. For example, $37$-isogenies would work (assuming $p \neq 37$).
The proof of the Ramanujan property is too long to repeat here, but essentially the largest eigenvalue of the $\ell$-isogeny graph is the $\ell$-th coefficient of the Eisenstein series $$ \sum_{n=1}^\infty \sigma_1(n) q^n = \sum_{n=1}^\infty \left(\sum_{d \mid n} d^1\right) q^n, $$ i.e. $\ell+1$ (which is also the degree of the graph), and the remaining eigenvalues are coefficients of cusp forms of weight $2$ on the modular curve $\Gamma_0(N)$ for some suitably chosen $N$, which are upper bounded by $\sigma_0(\ell) \sqrt{\ell} = 2\sqrt{\ell}$ by the Ramanujan-Petersson conjecture.
[1] Pizer, Ramanujan Graphs and Hecke Operators, Bulletin of the AMS 23 (1) 1990, pp. 127-137, https://doi.org/10.1090/S0273-0979-1990-15918-X
-
1$\begingroup$ Nice! Is there an easy way to see that the $\ell$-isogeny superingular graph is connected? (I would need to look at Corollary 77 in D. Kohel's thesis...). $\endgroup$– WatsonCommented Apr 16, 2023 at 17:54
-
2$\begingroup$ (By the way, the Ramanujan property is nicely sketched in the document: "Enric Florit - Random walks on supersingular isogeny graphs") $\endgroup$– WatsonCommented Apr 16, 2023 at 18:13
-
2$\begingroup$ I don't think it's substantially easier to prove connectedness than Ramanujan using this approach. You need a bound on the growth of the coefficients of cusp forms in any case. One could view the Ramanujan conjecture as a sort of Riemann hypothesis, and under this view, connectedness is analogous to proving the existence of a zero-free region whereas the Ramanujan property is like the full on Riemann hypothesis. $\endgroup$– djaoCommented Apr 17, 2023 at 0:13
-
$\begingroup$ @djao Thanks for your guidance and patient to my naive question. I just found out that you mentioned this in "2.2 Isogeny graphs" of the thesis "Towards quantum-resistant cryptosystems from ssEC isogenies". I read the thesis a few months ago, but didn't connect it to this problem. In a word, I learned a lot from your answer, thank you again for inspire and help. $\endgroup$ Commented Apr 19, 2023 at 9:28