Let $E$ be an elliptic curve over the rationals $\mathbb{Q}$. We consider the Galois representation attached to $E$ by acting on its $p$-adic Tate module $T_p(E)$, $$ \rho_E: G_{K} \rightarrow \mathrm{Aut}(T_p(E)) \cong \mathrm{GL}_2(\mathbb{Z}_p). $$ Then it cuts out a field $M := \overline{\mathbb{Q}}^{\ker(\rho_E)}$, whose Galois group over $\mathbb{Q}$ is isomorphic to the image of $\rho_E$. A theorem of Serre tells us that if $E$ has no CM, then $\mathrm{im}(\rho_E)$ is a finite index subgroup of $\mathrm{GL}_2(\mathbb{Z}_p)$ and strictly equals to $\mathrm{GL}_2(\mathbb{Z}_p)$ for all but finitely many primes $p$. So for simplicity, we take sufficiently large odd prime $p$ to guarantee $\rho_E$ has full image.

Further $\widetilde{\rho_E}: G_{K} \rightarrow \mathrm{PGL}_2(\mathbb{Z}_p)$. Then similarly consider $K := \overline{\mathbb{Q}}^{\ker(\widetilde{\rho_E})}$. We assume $E$ has no CM and $p$ sufficiently large so that $\mathrm{Gal}(K/\mathbb{Q}) \cong \mathrm{PGL}_2(\mathbb{Z}_p)$.

Ultimate Goal: The group of torsion points $E(K)_{\mathrm{tors}}$ is a finite group.

A crucial step in proving this (from my naive thought) is to ensure that

Claim 1: there exist infinitely many primes "good primes" $q$ such that the residue field for any prime of $K$ lying above $q$ is finite.

This post is asking how to prove (or disprove) this boldface claim 1?

My "attempt" is to imitate the proof of the following claim in the anticyclotomic $\mathbb{Z}_p$-extension case.

Claim 2: Let $F$ be an imaginary quadratic field and $F_{\infty}^{\mathrm{ac}}$ is its anticyclotomic $\mathbb{Z}_p$-extension. Then for any prime of $F$ which is inert in $F/\mathbb{Q}$ must split completely in $F_{\infty}^{\mathrm{ac}}/F$.

Then since there are infinitely many inert primes in $F$ (by density arguments), we have infinitely many primes that "has finite residue field in $F_{\infty}^{\mathrm{ac}}$".

The proof of Claim 2 may date back to Iwasawa in the second section of his article "On the $\mu$-invariants of $\mathbb{Z}_{\ell}$-extensions (1973)". A sketch of proof goes like this:

Let $q$ be a prime coprime to $p$ that is inert in $F$, write $\mathfrak{q}$ for the unique prime of $F$ lying above $q$, then $\mathfrak{q}$ is unramified in the $n$-th layer $F_{n}^{\mathrm{ac}}$. Let $\mathfrak{q}_n$ be a prime of $F_{n}^{\mathrm{ac}}$ lying above $\mathfrak{q}$ and $Z_n$ be the decomposition group of it for the Galois extension $F_{n}^{\mathrm{ac}}/\mathbb{Q}$. Then since $q$ is unramified in $F_{n}^{\mathrm{ac}}$ and is inert in $F$, $Z_n$ is a cyclic group of $G_n := \mathrm{Gal}(F_{n}^{\mathrm{ac}}/\mathbb{Q})$ such that $G_n = Z_n H_n$, where $H_n := \mathrm{Gal}(F_{n}^{\mathrm{ac}}/F)$. As $G_n$ is a dihedral group of order $2p^n$, it follows that $Z_n$ is a cyclic group of order two satisfying $Z_n \cap H_n = 1$. However, $Z_n \cap H_n$ is nothing but the decomposition group of $\mathfrak{q}_n$ for the extension $F_{n}^{\mathrm{ac}}/F$, hence $\mathfrak{q}$ splits completely in $F_{n}^{\mathrm{ac}}$.

Then to imitate the proof above, we need to know the parallel facts on general $\mathrm{PGL}_2(\mathbb{Z}_p)$-extensions $K/\mathbb{Q}$, for example:

  1. Specify all intermediate fields of the $\mathrm{PGL}_2(\mathbb{Z}_p)$-extension $K/\mathbb{Q}$. This is equivalent to classifying all closed subgroups of $\mathrm{PGL}_2(\mathbb{Z}_p)$ and figuring out which of them are open.

Maybe we can try some congruence subgroups? Let $\Gamma_n$ be the group of matrices in $\mathrm{GL}_2(\mathbb{Z}_p)$ that are congruence to identity matrix modulo $p^n$, and let $\overline{\Gamma_n}$ be its image of $\mathrm{PGL}_2(\mathbb{Z}_p)$. Do these groups work and are these all such subgroups? (I got hints from here.)

  1. The properties of the Galois groups of intermediate fields are good for us to run a similar argument as the anticyclotomic case. In this process, we hope to get a sufficient condition $(\star)$ for such "good primes" (that satisfying Claim 1).

Follow Greenberg's hint, maybe a common property for finite dihedral groups and $\mathrm{PGL}_2(\mathbb{Z}_p)$ is that $g$ and $g^{-1}$ are conjugate for any group element $g$. Unfortunately, I cannot see how this property is used in the above proof of anticyclotomic case either.

  1. There are infinitely many primes satisfying the sufficient condition $(\star)$ in 2.

Following Greenberg's hint, I guess that the primes $q$ such that $E$ has good supersingular reduction satisfies the abstract sufficient condition $(\star)$. Then a result of N. Elkies shows that there are infinitely many good supersingular primes. Yet I cannot see how such good supersingular properties are used in the abstract discussion of $\mathrm{PGL}_2(\mathbb{Z}_p)$-extensions. Since I am not able to do 2, figuring out 3 is impossible.

So though there is a rough three-step roadmap, I got stuck on each step. So I am here to ask if there is any way out.

Some further remarks:

  • This entire problem is motivated by solving Exercise 1.12 of Ralph Greenberg's IAS/Park City note "Introduction to Iwasawa Theory for Elliptic Curves". The hints above by Greenberg is taken from here.
  • The Claim 2 seems to work for anticyclotomic $\mathbb{Z}_p$-extension $\mathcal{F}_{\infty}^{\mathrm{ac}}$ for any CM field $\mathcal{F}/\mathcal{F}^{+}$.
  • Even if all the problems in this post are solved, we still cannot get the full proof of Greenberg's Exercise 1.12 that $E(K)_{\mathrm{tors}}$ is finite since we haven't touched the CM elliptic curves. An exercise in Silverman's book tells us that in this case, $\rho_E$ has abelian image in $\mathrm{GL}_2(\mathbb{Z}_p)$, and hence $\widetilde{\rho_E}$ has abelian image in $\mathrm{PGL}_2(\mathbb{Z}_p)$. Then in this CM elliptic curve case, it seems that we need a classification of abelian subgroups of $\mathrm{PGL}_2(\mathbb{Z}_p)$ as "Step 0" and try to run the argument above.

Or am I so stupid that missed some easy solution to this exercise? Actually, I feel like I am quite good at making things unnecessarily complicated and as a result, obtaining nothing valuable during my Ph.D. study up to now. Quite frustrated. :(

So sorry for such a long post, and thank you all for commenting and answering! :)

EDIT: Even Further Remarks: I found Greenberg's Exercise 1.16 was focusing on the CM case.

  • In Greenberg's Exercise 1.16, he considered a particular elliptic curve $y^2 = x^3 - x$ with CM $\mathbb{Z}[\sqrt{-1}]$. Let $F=\mathbb{Q}(\sqrt{-1})$. Then he asked the reader to show $K$ contains $F_{\infty}^{\mathrm{ac}}$ and $[K:F_{\infty}^{\mathrm{ac}}] < \infty$. This example (though I am still trying to prove) may inspire us to show that maybe (I am not sure at all) $K$ contains an anticyclotomic extension of finite index and then use the Claim 2 directly to conclude?
  • When searching for references, I found arXiv: 2008.04960, where the authors said in the second paragraph of the introduction that the $\mathrm{PGL}(2)$-extension does not contain the cyclotomic extension. This (though still I am trying to prove it) may further provide some evidence on that certain "anticyclotomic line" appears in $K/\mathbb{Q}$?
  • 2
    $\begingroup$ There exists no $\mathbf{Z}_p$ extension (cyclotomic or anticyclotomic or whatever) contained in a $PGL_2$ extension, just by elementary group theory: $\mathbf{Z}_p$ is not a quotient of $PGL_2(\mathbf{Z}_p)$. This is precisely why the PGL2 extensions are studied: because they are the first obvious example of p-adic Lie extensions which are "essentially" nonabelian, having no nontrivial abelian extensions inside them. $\endgroup$ Nov 22, 2023 at 14:25
  • $\begingroup$ @DavidLoeffler Thank you so much for your comment and sorry for the late reply. I managed to see the group theory fact you mentioned. Then inspired by your comment, I realized that I mixed up the CM case and the non-CM case, which seems to be completely different. In the CM case, the image of $\rho_E$ is an abelian subgroup of $\mathrm{PGL}_2(\mathbb{Z}_p)$, which could be quite small, and as Greenberg pointed out, it contains a $F_{\infty}^{\mathrm{ac}}$ "of finite index". (I have somehow managed to show this after struggling for two days, at least for $y^2=x^3-x$.) $\endgroup$
    – Hetong Xu
    Nov 24, 2023 at 7:46
  • $\begingroup$ ... Yet for the non-CM case, the image of $\rho_E$ is (except for finitely many prime $p$) the entire PGL2, which makes the story completely different. I hope I got this right this time, though still I have no idea how to get the ultimate goal proved. $\endgroup$
    – Hetong Xu
    Nov 24, 2023 at 7:49

1 Answer 1


It seems that I have obtained some positive results, but I couldn't believe my "solution" is correct since I haven't completely followed Greenberg's hints.

Throughout, $p$ is an odd prime.

Let's focus on the non-CM case only, and suppose $p$ is sufficiently large so that $\tilde{\rho}_E$ has full image $\mathrm{PGL}_2(\mathbb{Z}_p)$. For "small primes", the image is of finite index in $\mathrm{PGL}_2(\mathbb{Z}_p)$, so it doesn't harm.

Recall that we have the tower of extensions $$ \mathbb{Q} \subseteq K := \overline{\mathbb{Q}}^{\ker \tilde{\rho}_E} \subseteq M := \overline{\mathbb{Q}}^{\ker \rho_E} = \mathbb{Q}(E[p^{\infty}]). $$ Our goal is to understand $K/\mathbb{Q}$. To imitate the anticyclotomic $\mathbb{Z}_p$-case, we observe that there is a short exact sequence of abstract groups $$ 0 \rightarrow \mathrm{PSL}_2(\mathbb{Z}_p) \rightarrow \mathrm{PGL}_2(\mathbb{Z}_p) \xrightarrow{\det} \mathbb{Z}_p^{\times}/(\mathbb{Z}_p^{\times})^{2} = \{1, -1\} \rightarrow 0, $$ and $\mathrm{PGL}_2(\mathbb{Z}_p) = \mathrm{PSL}_2(\mathbb{Z}_p) \rtimes C_2$, where $C_2$ is a cyclic group of order two. So we have a quadratic extension $F/\mathbb{Q}$ such that $\mathrm{Gal}(K/F) \cong \mathrm{PSL}_2(\mathbb{Z}_p)$. Actually from the construction we see that $F$ is nothing but the fixed field of $\overline{\mathbb{Q}}$ by $\ker(\det \circ \tilde{\rho}_E)$.

In the $\mathrm{PSL}_2(\mathbb{Z}_p)$-extension $K/F$, we have many finite subextensions. We consider the quotient $\mathrm{PSL}_2(\mathbb{Z}/p^n)$ of $\mathrm{PSL}_2(\mathbb{Z}_p)$. It gives a finite subextension of $F_n$ of $K/F$ such that $$ \mathrm{Gal}(F_n/K) \cong \mathrm{PSL}_2(\mathbb{Z}/p^n), \quad \mathrm{Gal}(F_n/\mathbb{Q}) \cong \mathrm{PSL}_2(\mathbb{Z}/p^n) \rtimes C_2 \cong \mathrm{PGL}_2(\mathbb{Z}/p^n). \quad (\ast) $$

Pity 1: I still have no idea about the classification of closed/open subgroups of $\mathrm{PSL}_2(\mathbb{Z}_p)$ or $\mathrm{PGL}_2(\mathbb{Z}_p)$. But knowing some of these is sufficient to get this exercise done.

Then the story is similar to the anticyclotomic $\mathbb{Z}_p$-extension. But still, an important property need to be established:

"Property $(\dagger)$": Let $\mathfrak{q}$ be a prime of $F$ not lying above $p$, then $\mathfrak{q}$ is unramified in the $\mathrm{PSL}_2(\mathbb{Z}_p)$-extension $K/F$.

To imitate the $\mathbb{Z}_p$-extension case, it seems that we still need to know the closed subgroups of $\mathrm{PSL}_2(\mathbb{Z}_p)$. But here note that the extension $K/F$ arises from the elliptic curve $E$. We use:

Property (Neron-Ogg-Shafarevich): with notations above, $\mathfrak{q}$ is unramified in $F(E[p^{\infty}])$ if (and only if) $E$ has good reduction at $\mathfrak{q}$.

Then since $K \subseteq M=\mathbb{Q}(E[p^{\infty}]) \subseteq F(E[p^{\infty}])$, we see that at least for prime $\mathfrak{q}$ where $E/F$ has good reduction, $\mathfrak{q}$ is unramified in $K$.

Pity 2: I cannot show the "Property $(\dagger)$" for abstract $\mathrm{PSL}_2(\mathbb{Z}_p)$-extension of number fields.

Now let $q$ be a rational prime that is

  • not equal to $p$, odd
  • inert in $F/\mathbb{Q}$,
  • $E$ has good (not necessarily supersingular) reduction at $q$.

then by [Silverman, AEC, Proposition 5.4(a)], for any prime $\mathfrak{q}$ of $F$ over $q$, $E$ has good reduction at $\mathfrak{q}$, and hence $\mathfrak{q}$ is unramified in $K$. Obviously there are infinitely many such rational primes $q$.

Then we are finally ready to run the proof of Claim 2 quoted in my question, replacing the intermediate fields "$F_{n}^{\mathrm{ac}}$" there by $F_n$ here. The similar structure of Galois groups in $(\ast)$ helps us to conclude.

Edit: I was a little bit cheating here. Let me write some details here.

Let $\mathfrak{q}_n/\mathfrak{q}/q$ be the primes in the tower $F_n/F/\mathbb{Q}$. Then since $\mathfrak{q}/q$ is inert and $\mathfrak{q}_n/\mathfrak{q}$ is unramified, the decomposition group $Z(\mathfrak{q}_n/q)$ is cyclic of even order. Then by counting $\# \mathrm{PSL}_2(\mathbb{p}^n) = p^{3n-2} (p^2-1)/2$ (citing here), we see that $Z(\mathfrak{q}_n/\mathfrak{q})$ is of order at most $(p^2-1)/2$. (Here we used the assumption that $p$ is odd). Hence the ramification index $e_n$ and $f_n$ satisfies $e_n f_n \leq (p^2-1)/2$, which is bounded independent of $n$. This implies that the residue field of $q$ in $K$ is finite.

Remark: So here is still a little bit difference between the dihedral group case and the $\mathrm{PGL}_2$ case.

I hope that I haven't made mistakes in the proof. Here I would like to add a proof of Prof. David Loeffler's hint that $\mathrm{PGL}_2(\mathbb{Z}_p)$ has no quotient isomorphic to $\mathbb{Z}_p$.

We assume $p \geq 5$. For primes $p=2$ and $p=3$, actually I am not quite certain.

Proof: Suppose $\mathrm{PGL}_2(\mathbb{Z}_p) \twoheadrightarrow \mathbb{Z}_p$, then quotienting out $p\mathbb{Z}_p$, we obtain $\mathrm{PGL}_2(\mathbb{F}_p) \twoheadrightarrow \mathbb{F}_p$. This means (by counting the cardinality of the two groups $\mathrm{PGL}_2(\mathbb{F}_p)$ and $\mathrm{PGL}_2(\mathbb{F}_p)$) that we would have a normal subgroup $H$ of $\mathrm{PGL}_2(\mathbb{F}_p)$ of order $p^2-1$. We show next that this would not happen.

The key is the fact:

Fact of Galois: When $p \geq 5$, $\mathrm{PSL}_2(\mathbb{F}_p)$ is a simple group. (See K. Conrad's note here.)

This is the only place in the proof that we use the assumption $p \geq 5$. Suppose $H$ is a normal subgroup of $\mathrm{PGL}_2(\mathbb{F}_p)$, then by the simplicity result of Galois, $H \cap \mathrm{PSL}_2(\mathbb{F}_p)$ would be either trivial or the entire $\mathrm{PSL}_2(\mathbb{F}_p)$.

  • If $H \cap \mathrm{PSL}_2(\mathbb{F}_p) = \mathrm{PSL}_2(\mathbb{F}_p)$. Then $H \supseteq \mathrm{PSL}_2(\mathbb{F}_p)$. As $\mathrm{PSL}_2(\mathbb{F}_p)$ is of index two in $\mathrm{PGL}_2(\mathbb{F}_p)$, we see that either $H=\mathrm{PSL}_2(\mathbb{F}_p)$ or $H=\mathrm{PGL}_2(\mathbb{F}_p)$.
  • If $H \cap \mathrm{PSL}_2(\mathbb{F}_p) = 1$, then $H \cdot \mathrm{PSL}_2(\mathbb{F}_p) = \mathrm{PGL}_2(\mathbb{F}_p)$. By the second isomorphism theorem, $$ \mathrm{PGL}_2(\mathbb{F}_p)/H \cong H \cdot \mathrm{PSL}_2(\mathbb{F}_p) / H \cong \mathrm{PSL}_2(\mathbb{F}_p) / H \cap \mathrm{PSL}_2(\mathbb{F}_p) \cong \mathrm{PSL}_2(\mathbb{F}_p). $$ So $H$ is a normal subgroup of order $2$, strictly smaller than $p^2-1$.

So we are done.

I doubt if it is correct for $p=2$. For $p=3$, maybe it follows form a more direct computation since $\mathrm{PSL}_2(\mathbb{F}_3)$ is just the alternating group $A_4$.(Again here.)

  • $\begingroup$ It is very hard to classify open subgroups of $PGL_2$. In the final part, it should also be true for $p=2$. $\endgroup$
    – stupid boy
    Nov 25, 2023 at 19:21

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.