Let $E$ be an elliptic curve over a field $K$. I am mostly interested in the case $K$ is a number field or a local field but the following question is valid for all $K$.

Let $p$ be any prime power which is comprime to the characteristics of $K$. Galois cohomology gives a cup product map

$$H^{1}(K,E[p]) \times H^{1}(K,E[p]) \to H^{2}(K,E[p] \otimes E[p])$$

Composing this map with the map induced by the Weil pairing gives a map

$$ \phi: H^{1}(K,E[p]) \times H^{1}(K,E[p]) \to H^2(K,\mu_{p}). $$

Are there any known examples where the above map $\phi$ is non-trivial?

In theory, it is possible to compute both sides explicitly when p is small. However, computing the cup product is quite hard (for me.)


1 Answer 1


One can use the exact sequence $$ 0 \to E(K)/mE(K) \to H^1(K,E[m]) \to H^1(K,E)[m] \to 0 $$ to define a pairing $$ E(K)/mE(K) \times H^1(K,E)[m] \to H^2(K,\mu_m) $$ by taking $(Q,\xi)$ to $\phi(Q'\cup \xi')$, where $Q'$ is the image of $Q$ in $H^1(K,E[m])$ and $\xi'$ is any pullback of $\xi$ to $H^1(K,E[m])$.

For $K/\mathbb Q_p$, Tate proved that this is a perfect pairing (Tate local duality). Since it's easy to find examples with $E(K)/mE(K)\ne0$, this gives examples where your map $\phi$ is non-trivial. And indeed, if you take $m=p^k$, then you can force $E(K)/p^kE(K)$ to be quite large. (In this case, we have $H^2(K,\mu_m)\cong \frac1m\mathbb Z/\mathbb Z$, so this piece of the Brauer group makes a natural target for a pairing.)

For global fields, one does something similar to construct the Cassels-Tate pairing on the Tate-Shafarevich group, but the target space is instead $H^2(K,\mathcal{C}[m])$, where $\mathcal{C}$ is the idele class group. This is preferable to $H^2(K,\mu_m)$, since the Brauer group $H^2(K,\mathbb G_m)$ is huge.

  • $\begingroup$ Thank you for your comment. I will follow your suggestion to study the Cassels-Tate pairing. However, I have another related question about choosing the right objects. Suppose K=Q and E has conductor N and p does not divide N (that is E has good reduction at p.) Then, we can define the same pairing for $H_{et}^{1}(Z[1/Np],E[p]) \times H_{et}^{1}(Z[1/Np],E[p]) \times \to H_{et}^{2}(Z[1/Np],\mu_{p})$ (or Galois cohomology of $Q_{\Sigma}/Q$ with $\Sigma$ is the set of primes of Np). I checked that $H_{et}^{2}(Z[1/Np], \mu_{p})$ is nontrivial. What can we say about the pairing? $\endgroup$
    – user97464
    Commented Nov 15, 2016 at 21:11
  • $\begingroup$ @TungNguyen As you say, this is "another question", so you might post it as a new question, maybe with a link to this question to provide background. Offhand, I don't know the answer myself. $\endgroup$ Commented Nov 15, 2016 at 21:53
  • $\begingroup$ Thank you for your suggestion. I will post in another thread then. $\endgroup$
    – user97464
    Commented Nov 15, 2016 at 23:20

Your Answer

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