MathOverflow is a question and answer site for professional mathematicians. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Take a cusp form $f$ and let $f(q) = q + a_2q^2 + q_3q^3 + \ldots$" denote its $q$-expansion (assume that the $a_k$ are integers, and that $f$ comes from an elliptic curve $E$). Of course the series $f(1) = 1 + a_2 + a_3 + \ldots$ diverges, but I wonder whether there is any work on evaluating $f(1)$ via some regularization method (I do not even know whether $\lim_{q \to 1-0} f(q)$ exists or not). I am kind of hoping that $f(1)$ is connected with the order of the Tate-Shafarevich group of the associated elliptic curve $E$ and perhaps the order of its torsion group (the actual expression would have to be invariant under isogenies).

Does this ring a bell with anyone?

share|cite|improve this question
The limit as $q \rightarrow 1$ along the real axis, say, is certainly $0$, since $f$ is a cusp form and $q=1$ corresponds to $z=0$, which is a cusp. This is easy to see numerically for small conductor; the convergence to zero is staggeringly fast. Incidentally, Euler was already aware of this phenomenon in the related case $\sum_n (-1)^n q^{n^2}$ as $q \rightarrow 1$. – moonface Mar 26 '10 at 15:54
Maybe my answer was a bit dumb, in the case mentioned by moonface. Indeed, the value $L(f,0)$ vanishes for a cusp form $f$, due to the pole of the Gamma function at zero. So my long-winded answer is consistent with the vanishing observation of moonface. – Marty Mar 26 '10 at 16:54
up vote 2 down vote accepted

I'll give a slightly uncertain answer, based somewhat on my recollection of conversations with Zagier a month ago about similar questions.

If we were to imitate Euler, we might consider $f(1)$ as $$f(1) = \sum_{n \geq 1} a_n = \sum_{n \geq 1} a_n n^{-0} = L(f,0).$$ So the analytic continuation of the L-function suggests that $f(1)$ should be identified with the value of the L-function at zero. By the functional equation, this relates to the L-function at the right edge of the critical strip.

So, for a cusp form of weight two, arising from an elliptic curve $E$ over $Q$, the value $L(f,0)$ is related to $L(E,2)$. An interpretation of this L-value, conjectured by Zagier, was proven by Goncharov and Levin, in "Zagier's conjecture on $L(E,2)$", Invent. Math. 132 (1998).

As for the analytic question, you are considering the "value" of a cusp form $f$ on the real axis, which bounds the upper half-plane. Almost by definition, there is a Sato hyperfunction $f_{bdr}$ on the real axis, which describes this boundary behavior of the holomorphic function $f$ on the upper half-plane. I am not sure if the following is published, but I have the impression that there might be a preprint now or soon which proves the following result:

At every (positive? I don't recall) rational number $q$, the hyperfunction $f_{bdr}$ is $C^\infty$ at $q$. Its value at $1$ is $L(f,0)$ as described above.

I think that saying "a hyperfunction is $C^\infty$ at $q$" means that the hyperfunction can be expressed as the distributional derivative of a continuous function -- $f = g^{(k)}$ for some $k \geq 0$ -- and $g$ happens to be $C^\infty$ at $q$. But I'm not much of an analyst.

I think that the value $f(1)$ also exists as $\lim_{z \rightarrow 1} f(z)$ limit, if $z$ approaches $1$ via a geodesic in the upper half-plane.

I don't think you'll see Sha or the torsion directly, as these appear at the central value $L(f,1)$. On the other hand, I do think you'll find $L(f,-n)$ for all $n \geq 0$ (or equivalently, $L(f,2+n)$ ), by looking at the derivatives $f^{(n)}(1)$ of the boundary hyperfunction of $f$ at $1$.

share|cite|improve this answer
Somehow it would surprise me if $f(q)$ really did tend to $L(E,0)$ in any reasonable sense, because isn't it a result of Szpiro that if $\sum_n a_n.n^{-s}$ converges as $s$ tends to 1, then it won't converge to $L(E,1)$ but rather to something like $\sqrt{2}$ times this? I forgot what he actually proved unfortunately. Somehow if you can't get it right at 1 with 'naive' methods, what chance have you of getting it right at 0? Or do you think I've got the wrong end of the stick? – Kevin Buzzard Mar 26 '10 at 15:17
But here I'm looking at $f(z)$ as $z$ approaches $1$, via certain paths in the upper half-plane. These values have almost nothing in common with the values of the Dirichlet series as $s$ approaches zero. For cusp forms, the limits I'm talking about are in the same spirit as the convergent integral representation of the L-function. – Marty Mar 26 '10 at 15:23
Sounds like I have indeed got the wrong end of the stick then :-) Apologies. – Kevin Buzzard Mar 26 '10 at 15:52
Voloch is right: this $\sqrt{2}$ is related to partial Euler product behavior as $s$ tends to 1, not to the Dirichlet series representation for $L(s)$, which in fact converges for $s$ with real part greater than 5/6 (so at $s=1$). If you replace elliptic curve $L$-functions with Hecke $L$-functions for a quadratic character, you can turn the unexpected $\sqrt{2}$ into $1/\sqrt{2}$ with the Euler product as $s$ tends to $1/2$. Oh, and all such Euler product convergence implies GRH so you can numerically test this stuff and see it happen but you'll never unconditionally prove it. – KConrad Mar 27 '10 at 4:01
Oh, and the convergence of $L(s)$ for $s$ with real part greater than 5/6 uses the elliptic modularity theorem. – KConrad Mar 27 '10 at 4:02

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.