8
$\begingroup$

Let $F\in M_k(\Gamma_0(N),\chi)$, not necessarily an eigenform nor cuspidal, but assume that $\Bbb Q(F)$ is a number field $K$. What can one say of $\Bbb Q(F|_kW_N)$, where $W_N$ is the Fricke involution ? Experiments seem to show that the Fourier coefficients of $F|_kW_N$ divided by $\sqrt{Q}$ for some positive or negative divisor of $N$ (probably linked to the conductor of $\chi$) also lie in $K$.

More generally same question for a general Atkin--Lehner involution $W_Q$, and also for $1/2$-integral weight (in which case even a fourth root may be necessary).

It may be possible to start with a corresponding result for newforms, but I have not seen how to complete the argument.

P.S. Since I tested mainly with real characters, $\sqrt{Q}$ may of course be the Gauss sum associated to $\chi$. But the questions remain.

Improve this question
$\endgroup$
3
  • $\begingroup$ To clarify, are you saying there is a choice of $Q|N$ such that all Fourier coefficients divided by $\sqrt Q$ lie in $K$, or can $Q$ vary with the coefficients? $\endgroup$
    – Kimball
    Sep 9, 2017 at 23:21
  • $\begingroup$ Can you say that $F$ is a $K$-linear combination of eigenforms with coefficient fields in $K$? $\endgroup$
    – Kimball
    Sep 9, 2017 at 23:24
  • $\begingroup$ Yes, $Q$ is a fixed positive or negative divisor of $N$. But $\sqrt{Q}$ could instead be the Gauss sum for $\chi$ or something similar. $\endgroup$
    – Henri Cohen
    Sep 10, 2017 at 8:27

1 Answer 1

Reset to default
7
$\begingroup$

Your guess that this is related to the Gauss sum is correct. If $f$ has integer weight, level $N$ and coefficients in $K$, then $W_N(f) / \tau(\chi)$ has coefficients in $K$ also.

This follows from a lovely theorem of Shimura, which states that $M_k(\Gamma(N), \mathbf{Q}[\zeta_N])$ is preserved by the action of $SL_2(\mathbf{Z} / N)$, and that the actions of $SL_2(\mathbf{Z} / N)$ and $\operatorname{Gal}(\mathbf{Q}[\zeta_N] / \mathbf{Q})$ piece together into an action of $GL_2(\mathbf{Z} / N)$ if you identify the Galois group with the matrices of the form $\begin{pmatrix} * & 0 \\ 0 & 1\end{pmatrix}$ (or it might be $\begin{pmatrix} 1 & 0 \\ 0 & *\end{pmatrix}$, I can't remember).

To use this here, consider $f$ as an element of $M_k(\Gamma(N), K)$, and look at the action of $\begin{pmatrix} 0 & -1 \\ 1 & 0 \end{pmatrix}$; the result is $W_N(f)(z/N)$ up to a power of $N$, which has the same coefficient field as $W_N(f)$ [edit: possibly up to a factor of $\sqrt N$ in odd weights, depending on you conventions], and Shimura's result now tells you that this has coefficients in $K(\zeta_N)$ and you can read off exactly how $Gal(K(\zeta_N) / K)$ acts.

The same argument will tell you how partial Atkin--Lehner operators $W_Q$ for $Q \| N$ act, with a little more bookkeeping -- if I remember correctly, you get the Gauss sum of the Q-primary part of $\chi$ coming out.

A reference for this is: Ohta, "P-adic Eichler--Shimura isomorphisms" (Crelle #463, 1995), sections 3.5 and 3.6. Lemma 3.5.2 on page 83 is a reciprocity law describing how Galois acts on $W_Q(f)$, which is (I hope!) equivalent to the statement I gave above; and section 3.6 (pages 86-9) gives a detailed proof of the lemma using Katz's algebraic description of modular forms.

(I have no idea about the half-integer weight case.)

Improve this answer
$\endgroup$
4
  • $\begingroup$ Thanks! perfect explanation, I will look at the paper you mention. $\endgroup$
    – Henri Cohen
    Sep 10, 2017 at 17:09
  • $\begingroup$ Unfortunately, the result you mention seems to be correct (experimentally) only for even weights. In odd weights, you clearly must also divide by a Gauss sum, but also by (something like) the square root of the level. Simplest (counter)example: the Eisenstein series of weight 3 and character -3, level 3. $\endgroup$
    – Henri Cohen
    Sep 11, 2017 at 20:04
  • $\begingroup$ @HenriCohen How are you normalising the action of $W_N$? There are various conventions as to whether you have $N^k$ or $N^{k/2}$ or $N^{k-2}$ etc, which of course can throw off the normalisations when $k$ is odd. I think Ohta defines $W_N(f)(z) = N^{-1} z^k f(-1/Nz)$, which means that the operator is not an involution in weights $\ne 2$. If you are normalising so that $W_N^2 = 1$ then that probably accounts for the discrepancy. $\endgroup$
    – David Loeffler
    Sep 11, 2017 at 20:12
  • $\begingroup$ Yes, I do normalize so that $W_N^2=1$, so the discrepancy is explained. $\endgroup$
    – Henri Cohen
    Sep 12, 2017 at 8:30

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.