I'm a bit confused by the role of the Weber function in generating ray class fields of imaginary quadratic fields of class number one. More specifically, let $K$ be such a field and $E$ an elliptic curve defined over $\mathbf{Q}$ with CM by $\mathscr{O}_K$. Let $\mathfrak{m}$ be a modulus for $K$. To get the ray class field for $\mathfrak{m}$, we consider the torsion group $E[\mathfrak{m}]$. This is a free $\mathscr{O}_K/\mathfrak{m}$-module of rank $1$, and $G_K$ acts on it by $\mathscr{O}_K$-linear automorphisms. Thus we have a character $\alpha \colon \mathrm{Gal}(K(E[\mathfrak{m}])/K) \hookrightarrow \mathscr{O}_K^\times$. Moreover, the Main Theorem of CM tells us that $\alpha(\mathrm{rec}(s))$ acts by $\lambda(s) s_\mathfrak{m}^{-1}$ for $\lambda \colon \mathbf{A}_{K, f}^\times \rightarrow K^\times$ the CM Großencharacter. Note that if $s \equiv 1 \pmod {\mathfrak{m}}$, this tells us that $\alpha(\mathrm{rec}(s)) = \lambda(s) \pmod {\mathfrak{m}}$, and that if $s = (\gamma)$ is a principal idele, that $\lambda(s) = s_\mathfrak{m} \pmod{\mathfrak{m}}$.

Now, at least if $\mathfrak{m}$ is divisible by the conductor of $\lambda$ (i.e. $\mathfrak{m}$ is divisible by the primes of bad reduction of $E$), we see that $\alpha(\mathrm{rec}(s))$ only depends on the image of $s$ in $\mathrm{Cl}_\mathfrak{m}(K)$. Thus, we have a map $\alpha \circ \mathrm{rec} \colon \mathrm{Cl}_{\mathfrak{m}}(K) \rightarrow \mathrm{Gal}(K(E[\mathfrak{m}])/K) \hookrightarrow (\mathscr{O}_K/\mathfrak{m})^\times$.

If we compose $\alpha \circ \mathrm{rec}$ with the map $(\mathscr{O}_K/\mathfrak{m})^\times \rightarrow (\mathscr{O}_K/\mathfrak{m})^\times/\mathscr{O}^\times = \mathrm{Cl}_\mathfrak{m}(K)$, we certainly get an isomorphism, since we can see that $\alpha \circ \mathrm{rec}(\pi_v) = \lambda(\pi_v)$ is a generator of $v$ for all but finitely many primes $v$ of $K$ which are split over $\mathbf{Q}$.

In the sources I've read on this, it seems that one replaces $\alpha \colon \mathrm{Gal}(K(E[\mathfrak{m}])/K) \hookrightarrow \mathscr{O}_K^\times$ with $\mathrm{Gal}(K(h(E[\mathfrak{m}]))/K) \hookrightarrow (\mathscr{O}_K/\mathfrak{m})^\times/\mathscr{O}_K^\times$ where $h$ is a Weber function. This shows that $K(h(E[\mathfrak{m}]))$ is the ray class field of $K$ with conductor $\mathfrak{m}$. One sometimes comments that the field $K(E[\mathfrak{m}])$ might not be abelian over $K$ in general, since $E$ is only defined over the Hilbert class field of $K$. But in the class number one case, this problem goes away.

So what happens to $\alpha$? We've shown that when $\mathfrak{m}$ is divisible by the conductor of $\lambda$, $K(E[\mathfrak{m}])$ is an abelian extension of $K$ such that the reciprocity map kills ideals with generators which are congruent to $1$ mod $\mathfrak{m}$, so it must be contained in the ray class field with conductor $\mathfrak{m}$, i.e. the subfield $K(h(E[\mathfrak{m}])$. Thus, these are the same field. Is $\alpha$ surjective, or is the image some index $2$ (or $4$ or $6$ for the cases of extra automorphisms, I suppose) subgroup of $(\mathscr{O}_K/\mathfrak{m})^\times$ mapping isomorphically onto the ray class group?

If $\mathfrak{m}$ is not divisible by the conductor of $\lambda$, $K(h(E[\mathfrak{m}]))$ is still the ray class field of conductor $\mathfrak{m}$, but I don't think my proof shows the reciprocity map into $\mathrm{Gal}(K(E[\mathfrak{m}])/K)$ kills principal ideals with a generator which is congruent to $1$ mod $\mathfrak{m}$. Are these still the same field? If not, $K(E[\mathfrak{m}])$ is some abelian extension of $K$, so it must sit inside $K(h(E[\mathfrak{n}]))$ for some $\mathfrak{n}$ - which one?


A quick comment (which I have to post as an answer, since I was locked out of my old account): The construction of $K(j(E), E[\mathfrak{m}])$ you describe in the general gives abelian extensions of the Hilbert class field $K(j(E))$ which might not be abelian over $K$. The role of the Weber function $h$ is to select out the subfields which are abelian over $K$. Geometrically, fixing a model for $E$, this $h$ can be defined as any finite map $h: E \longrightarrow E/\operatorname{Aut}(E)\approx E/\mathcal{O}_K^{\times}$, and does not depend on the choice of model. (Analytically, it is defined as $z \in {\bf{C}}/\Lambda \approx E({\bf{C}}) \longrightarrow (\wp(z, \Lambda), \wp'(z, \Lambda))$, and does not depend on the choice of lattice $\Lambda$). Hence, the way to construct the ray class field of conductor $\mathfrak{m}$ of $K$ in general is to adjoin to the coordinates of the images in $X := E/\operatorname{Aut}(E)$ of the torsion points $E[\mathfrak{m}]$. If $O_K^{\times}$ has order 2, then the homomorphism $E \rightarrow X$ is given by the coordinate $x$; if the order is 4 it is given by $x^2,$ and if the order is 6 then it is given $x^3$. In other words, the ray class extension of conductor $\mathfrak{m}$ of $K$ is generated by $j(E)$ and the coordinates $x$, $x^2$, or $x^3$ respectively, which to be explicit is $K(j(E), h(E[\mathfrak{m}]))$ rather than $K(j(E), E[\mathfrak{m}])$. However, when the class number of $K$ is one, then it is usually $K(E[\mathfrak{m}])$ as you suggest, but could be a larger abelian extension of $K$. Sorry that this comment does not answer the original question about the exact location of $K(E[\mathfrak{m}])$. (This should come down to keeping track of action by elements of $\operatorname{Aut}(E) = \mathcal{O}_K^{\times}$ on torsion points in the proof of the main theorem). If I have time, then I will try to post an answer later.

  • $\begingroup$ Doesn't the $\mathscr O_K$-linear Galois representation of $G_K$ on $E[\mathfrak{m}]$ depend intrinsically on $E$? I thought that in the class number $1$ case, there was just one isomorphism class of elliptic curve $E$ with CM by $\mathscr O_K$. $\endgroup$ Oct 29 '18 at 19:59
  • $\begingroup$ Yes, indeed — thanks for pointing this out! You are of course right that there is only one isomorphism class when $K$ has class number one. In this case, the maximal abelian extension of $K$ is in fact generated by the torsion of an elliptic curve $E$ with CM by $O_K$. In general, adjoining the torsion of an elliptic curve $E$ with CM by $\mathcal{O}_K$ generates abelian extensions of the Hilbert class field $K(j(E))$ which are not necessarily abelian over $K$. The role of the Weber function $h$ (as defined geometrically as in my comment) is to select out subfields which are abelian over $K$. $\endgroup$ Oct 30 '18 at 16:59
  • 1
    $\begingroup$ I think the question is not asking about the general case, but the particular case of class number 1. In that case, it is known that $K(E^{tors}) = K(h(E^{tors})) = K^{ab}$. The question is whether an analogous equality holds for the $\mathfrak{m}$-torsion, i.e. what is the relationship between $K(E[\mathfrak{m}])$ and $K(h(E[\mathfrak{m}]))$. You say in your last sentence that these fields are equal -- why is this true? I have not seen this stronger statement in standard texts on this. $\endgroup$ Oct 30 '18 at 18:10
  • $\begingroup$ I don't think this is true. Even if you have an elliptic curve $E$ for which $K(E[\mathfrak{m}]) =K(h(E[\mathfrak{m}]))$, one can twist $E$ by a quadratic field $K(\sqrt{d}) \not\subseteq K(h(E[\mathfrak{m}]))$ to get a curve $E_d$. Now, $h$ doesn't depend on which twist you pick, so $K(h(E[\mathfrak{m}])) =K(h(E_d[\mathfrak{m}]))$; but $K(E_d[\mathfrak{m}]) \neq K(h(E_d[\mathfrak{m}]))$. In fact, there are infinitely many such $d$. $\endgroup$
    – Rdrr
    Jul 28 at 21:49

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.