Applications Strong multiplicity one for \(\GL(n)\)

Let \pi=\otimes\pi_p and \pi'=\otimes\pi_p' be cuspidal automorphic representations of the group \GL(n,\A_\Q). For a finite prime p for which \pi_p and \pi_p' are both unramified, let A(\pi_p) (resp.A(\pi_p')) represent the semisimple conjugacy class in \GL(n,\C) corresponding to \pi_p (resp.\pi_p'). The strong multiplicity one theorem for \GL(n) states that if A(\pi_p)=A(\pi_p') for almost all p, then \pi=\pi'. The following result implies, in particular, that the equality of traces {\rm tr}(A(\pi_p))={\rm tr}(A(\pi_p')) for almost all p is sufficient to reach the same conclusion. The traces could even be different at every prime, if those differences decreased sufficiently rapidly as a function ofp.

Suppose that \pi and \pi' are (unitary) cuspidal automorphic representations of \GL(n,\A_\Q) satisfying

\sum_{p\le X} p\,\log(p)\left| \tr A(\pi_p) - \tr A(\pi_p')\right|^2 \ll X .

Assume a partial Ramanujan bound for some \theta\lt \frac16 holds for both incomplete L-functions L_{{\rm fin}}(s,\pi) and L_{{\rm fin}}(s,\pi'). Then \pi=\pi'.

We apply Theorem to L_1(s)=L_{{\rm fin}}(s,\pi) and L_2(s)=L_{{\rm fin}}(s,\pi'). The condition on the spectral parameters \Re(\mu_j),\ \Re(\nu_j)>-\frac12 is satisfied by Proposition 2.1 of. By, the partial symmetric square L-function for \GL(n) has meromorphic continuation to all of \C and only finitely many poles in \sigma \geq 1. Using the fact that the partial Rankin-Selberg L-function of a representation of \GL(n) with itself has no zeros in \sigma \geq 1 (see) and that the partial exterior square L-function of \GL(n) has only finitely many poles (see), we see that partial symmetric square L-function for \GL(n) has only finitely many zeros in \sigma \geq 1. This gives us condition) of Theorem. The conclusion of Theorem is that L_1(s)=L_2(s). By the familiar strong multiplicity one theorem for \GL(n), this implies \pi_1=\pi_2.

Siegel modular forms

In this section we prove Theorem. We start by giving some background on Siegel modular forms for \Sp(4,\Z). Let the symplectic group of similitudes of genus 2 be defined by

\GSp(4) := \{g \in \GL(4) : {}^{t}g J g = \lambda(g) J, \lambda(g) \in \GL(1) \} \\ \mbox{ where } J = \mat{}{I_2}{-I_2}{}.

Let \SSp(4) be the subgroup with \lambda(g)=1. The group \GSp^+(4,\R) := \{ g \in \GSp(4,\R) : \lambda(g) > 0 \} acts on the Siegel upper half space \HH_2 := \{ Z \in M_2(\C) : {}^{t}Z = Z,\:{\rm Im}(Z) > 0 \} by

g \langle Z \rangle := (AZ+B)(CZ+D)^{-1} \qquad \mbox{ where } g = \mat{A}{B}{C}{D} \in \GSp^+(4,\R), Z \in \HH_2.

Let us define the slash operator |_k for a positive integer k acting on holomorphic functions F on \HH_2 by

(F|_kg)(Z) := \lambda(g)^k \det(CZ+D)^{-k} F(g \langle Z \rangle), \\ g = \mat{A}{B}{C}{D} \in \GSp^+(4,\R), Z \in \HH_2.

The slash operator is defined in such a way that the center of \GSp^+(4,\R) acts trivially. Let S_k^{(2)} be the space of holomorphic Siegel cusp forms of weight k, genus 2 with respect to \Gamma^{(2)} := \SSp(4,\Z). Then F \in S_k^{(2)} satisfies F |_k \gamma = F for all \gamma \in \Gamma^{(2)}.

Let us now describe the Hecke operators acting on S_k^{(2)}. For a matrix M \in \GSp^+(4,\R) \cap M_{4}(\Z), we have a finite disjoint decomposition

\Gamma^{(2)} M \Gamma^{(2)} = \bigsqcup\limits_i \Gamma^{(2)} M_i.

For F \in S_k^{(2)}, define

T_k(\Gamma^{(2)} M \Gamma^{(2)})F := \det(M)^{\frac{k-3}2}\sum\limits_i F|_kM_i.

Note that this operator agrees with the one defined in. Let F \in S_k^{(2)} be a simultaneous eigenfunction for all the T_k(\Gamma^{(2)} M \Gamma^{(2)}), M \in \GSp^+(4,\R) \cap M_{4}(\Z), with corresponding eigenvalue \mu_F(\Gamma^{(2)} M \Gamma^{(2)}). For any prime number p, it is known that there are three complex numbers \alpha_0^F(p), \alpha_1^F(p), \alpha_2^F(p) such that, for any M with \lambda(M) = p^r, we have

\mu_F(\Gamma^{(2)} M \Gamma^{(2)}) =\alpha_0^F(p)^r \sum_i \prod\limits_{j=1}^2 (\alpha_i^F(p)p^{-j})^{d_{ij}},

where \Gamma^{(2)} M \Gamma^{(2)} = \bigsqcup_i \Gamma^{(2)} M_i, with

M_i = \mat{A_i}{B_i}{0}{D_i} \quad \mbox{ and } \quad D_i = \begin{bmatrix}p^{d_{i1}}\amp \ast\\0\amp p^{d_{i2}} \end{bmatrix} .

Henceforth, if there is no confusion, we will omit the F and p in describing the \alpha_i^F(p) to simplify the notations. The \alpha_0, \alpha_1, \alpha_2 are the classical Satake p-parameters of the eigenform F. It is known that they satisfy

\alpha_0^2 \alpha_1 \alpha_2 = p^{2k-3}.

For any n > 0, define the Hecke operators T_k(n) by

T_k(n) = \sum\limits_{\lambda(M)=n} T_k(\Gamma^{(2)} M \Gamma^{(2)}).

Let the eigenvalues of F corresponding to T_k(n) be denoted by \mu_F(n). Set \alpha_p =p^{-(k-3/2)}\alpha_0 and \beta_p = p^{-(k-3/2)}\alpha_0 \alpha_1. Then formulas for the Hecke eigenvalues \mu_F(p) and \mu_F(p^2) in terms of \alpha_p and \beta_p are

\mu_F(p)\amp = p^{k-3/2} \big(\alpha_p + \alpha_p^{-1} + \beta_p + \beta_p^{-1}\big), \mu_F(p^2)\amp =p^{2k-3}\big(\alpha_p^2+\alpha_p^{-2}+(\alpha_p+\alpha_p^{-1})(\beta_p+\beta_p^{-1})+\beta_p^2+\beta_p^{-2}+2-\frac1p\big) .

The Ramanujan bound in this context is

|\alpha_p| = |\beta_p| = 1.

This is closely related to our use of that term for L-functions, as can be seen from the spin L-function ofF:

L(s,F,\spin)=\prod_p F_p(p^{-s},\spin)^{-1},

where F_p(X,\spin)=(1 - \alpha_p X)(1 - \alpha_p^{-1} X)(1 - \beta_p X)(1 - \beta_p^{-1} X). It satisfies the functional equation

\Lambda(s,F,\spin) :=\mathstrut\amp \Gamma_\C(s+\tfrac12)\Gamma_\C(s+k-\tfrac32) L(s,F,\spin) =\mathstrut\amp \varepsilon \Lambda(s,\overline{F},\spin),

where \varepsilon=(-1)^k.

Let a(p) be the pth Dirichlet coefficient of L(s,F,\spin). We will use the fact that each F falls into one of two classes.

  1. a(p)=p^{1/2}+p^{-1/2}+\beta_p+\beta_p^{-1}, where \beta_p is the Satake p-parameter of a holomorphic cusp form on \GL(2) of weight 2k-2. In this case F is a Saito-Kurokawa lifting; for more details on Saito-Kurokawa liftings we refer to. Note that |\beta_p|=1, so that a(p)=p^{1/2}+O(1) in the Saito-Kurokawa case.
  2. a(p)=O(1). This is the Ramanujan conjecture for non-Saito-Kurokawa liftings, which has been proven in.

Theorem is now a consequence of the following stronger result.

Suppose F_j, for j=1,2, are Siegel Hecke eigenforms of weight k_j for \Sp(4,\Z), with Hecke eigenvalues \mu_j(n). If

\sum_{p\le X} p\,\log(p) \left|p^{3/2-k_1}\mu_1(p)-p^{3/2-k_2}\mu_2(p)\right|^2\ll X

as X\to\infty, then k_1=k_2 and F_1 and F_2 have the same eigenvalues for the Hecke operator T(n) for alln.

For i=1,2 let a_i(p) be the pth Dirichlet coefficient of L(s,F_i,\spin). Then a_i(p)=\alpha_{i,p}+\alpha_{i,p}^{-1}+\beta_{i,p}+\beta_{i,p}^{-1}, where \alpha_{i,p},\beta_{i,p} are the Satake p-parameters of F_i, as explained after . By ,

\mu_i(p)= p^{k_i-3/2} \big(\alpha_{i,p} + \alpha_{i,p}^{-1} + \beta_{i,p} + \beta_{i,p}^{-1}\big).

Hence, condition translates into

\sum_{p\le X} p\,\log(p) \left|a_1(p)-a_2(p)\right|^2\ll X.

From the remarks made before the theorem, we see that either F_1,F_2 are both Saito-Kurokawa lifts, or neither of them is a Saito-Kurokawa lift.

Assume first that F_1,F_2 are both Saito-Kurokawa lifts. Then, for i=1,2, there exist modular forms f_i of weight 2k_i-2 and with Satake parameters \beta_{i,p} such that a_i(p)=p^{1/2}+p^{-1/2}+\beta_{i,p}+\beta_{i,p}^{-1}. From we obtain

\sum_{p\le X} p\,\log(p) \left|b_1(p)-b_2(p)\right|^2\ll X,

where b_{i,p}=\beta_{i,p}+\beta_{i,p}^{-1}. Note that b_{i,p} is the pth Dirichlet coefficient of (the analytically normalized L-function) L(s,f_i). Since the Ramanujan conjecture is known for elliptic modular forms, Theorem applies. We conclude 2k_1-2=2k_2-2 and L(s,f_1)=L(s,f_2). Hence k_1=k_2 and L(s,F_1,\spin)=L(s,F_2,\spin). The equality of spin L-functions implies \mu_1(p)=\mu_2(p) and \mu_1(p^2)=\mu_2(p^2) for all p. Since T(p) and T(p^2) generate the p-component of the Hecke algebra, it follows that \mu_1(n)=\mu_2(n) for all n.

Now assume that F_1 and F_2 are both not Saito-Kurokawa lifts. Then, using the fact that the Ramanujan conjecture is known for F_1 and F_2, Theorem applies to L_1(s)=L(s,F_1,\spin) and L_2(s)=L(s,F_2,\spin). We conclude that k_1=k_2 and that the two spin L-functions are identical. As above, this implies \mu_1(n)=\mu_2(n) for all n.

Hyperelliptic curves

Let X/\Q be an elliptic or hyperelliptic curve,

X:\ y^2 = f(x),

where f\in \Z[x], and let N_X(p) be the number of points on X modp. In Serre's recent book, the title of Section6.3 is ``About N_X(p)-N_Y(p),'' in which he gives a description of what can happen if N_X(p)-N_Y(p) is bounded. For Serre, X and Y are much more general than hyperelliptic cuves, but we use the hyperelliptic curve case to illustrate an application of multiplicity one results for L-functions.

Recall that the Hasse-Weil L-function of X,

L(X,s) = \sum_{n=1}^\infty \frac{a_X(n)}{n^s},

has coefficients a_X(p^n)=p^n+1-N_X(p^n) if p is prime, which gives the general case by multiplicativity. The L-function (conjecturally if g_X\ge 2) satisfies the functional equation

\Lambda(X,s) = N_X^{s/2}\,\Gamma_\C(s)^{g_X} L(X,s) = \pm \Lambda(X,2-s),

where N_X is the conductor and g_X= \lfloor ({\rm deg}(f)-1)/2 \rfloor is the genus ofX.

Suppose X and Y are hyperelliptic curves and N_X(p)-N_Y(p) is bounded. If the Hasse-Weil L-functions of X and Y satisfy their conjectured functional equation, then X and Y have the same conductor and genus, and N_X(p^e)=N_Y(p^e) for all p,e.

Note that this result can be found in Serre's book without the hypothesis of functional equation. But Serre's proof involves more machinery than we use here.

To apply Theorem, we first form the analytically normalized L-function

L(s,X)=L(X,s+\tfrac12) =\sum \frac{a_X(n)/\sqrt{n}}{n^{s}} =\sum \frac{b_X(n)}{n^{s}},

say. Note that we have the functional equation

\Lambda(s,X) = N_X^{s/2} \Gamma_\C(s+\tfrac12)^{g_X} L(s,X) = \pm \Lambda(1-s,X).

The Hasse bound for a_X(n) implies the Ramanujan bound for L(s,X). The condition |N_X(p)-N_Y(p)| \ll 1 is equivalent to

|b_X(p) - b_Y(p)|\ll \frac{1}{\sqrt{\mathstrut p}},

which implies

\sum_{p\le T} p|b_X(p) - b_Y(p)|^2 \log(p) \ll \sum_{p\le T} \log(p) \sim T,

by the prime number theorem. Thus, Theorem applies and we conclude that L(X,s)=L(Y,s).

If one knew that L(s,X) and L(s,Y) were ``automorphic'', then Theorem would apply, and a much weaker bound on |N_X(p)-N_Y(p)| would allow one to conclude that N_X(p^e)=N_Y(p^e) for all p,e. For example, if E,E' are elliptic curves over \Q, then |N_{E}(p)-N_{E'}(p)|\le 1.4 \sqrt{\mathstrut p} for all but finitely many p implies N_{E}(p)=N_{E'}(p) for allp.

\appendix