On Products of Fourier Coefficients of Cusp Forms

# On Products of Fourier Coefficients of Cusp Forms

Eric Hofmann    Winfried Kohnen
###### Abstract

The purpose of this paper is to study products of Fourier coefficients of an elliptic cusp form, for a fixed positive integer , concerning both non-vanishing and non-negativity.111 2010 Mathematics Subject Classification: 11F12, 11F30

\KOMAoptions

abstract=on

## 1 Introduction

Let be an elliptic cusp form of positive integral weight with real Fourier coefficients . Let be a fixed positive integer. Then, the purpose of this paper is to study the products , regarding both non-vanishing and non-negativity. For a precise statement of our results see section 2 below.

We remark that these products have been previously investigated under different aspects, namely first by Selberg [6] and by Good [1] who studied growth properties of the sums where , and more recently by Hoffstein and Hulse [2] in connection with shifted Dirichlet convolutions.

## 2 Statement of results

Let denote the upper half-plane of complex numbers , with . The modular group acts on by fractional linear transformations, as usual.

In the following, we denote by a discrete subgroup of , which satisfies the following conditions [cf. 1, p. 98]:

1. is a finitely generated Fuchsian group of the first kind.

2. The negative identity matrix is contained in .

3. contains exactly if is an integer.

Condition (iii) means that has a cusp at , the stabilizer of which is generated by the two matrices .

We shall prove the following theorem

###### Theorem 1.

Let be a cusp form of integer weight on with real Fourier coefficients . Let and assume that is not identically zero. Then, in fact, infinitely many terms of the sequence are non-zero.

The following corollary is an immediate consequence of the theorem. Recall only that for a normalized Hecke eigenform, the Fourier coefficients are real and .

###### Corollary 1.

Let be a cusp form of integer weight and level that is a normalized Hecke eigenform with Fourier coefficients . Let and suppose that . Then, the sequence has infinitely many non-vanishing terms.

From Corollary 1, we obtain

###### Corollary 2.

Let be a normalized Hecke eigenform of integer weight on with Fourier coefficients . Then, there are infinitely many such that and are both non-zero.

###### Proof.

By results from [7] and [3], the second Fourier coefficient is non-zero modulo any prime lying above (in an appropriate finite extension of ), hence is non-zero. The assertion follows. ∎

The proof of Theorem 1 which will begin in section 3 makes use of a Dirichlet series associated to the sequence and of its analytic properties. This series was introduced by Selberg [6] and later studied by Good [1], see above.

Further, assuming is a congruence subgroup of level , we use a result of Knopp, Kohnen and Pribitkin [4], Theorem 1, on sign changes of Fourier coefficients of cusp forms to show the following:

###### Theorem 2.

Let be a cusp form of integer weight on , with real Fourier coefficients . Let . Then the sequence has infinitely many non-negative terms.

###### Remark.

In the same way, one can prove that the sequence has infinitely many non-positive terms.

The proof of this theorem is carried out in section 4. We note that, in fact a somewhat more general statement holds, as the requirements of [4], which need to be considered in addition to our conditions (i)–(iii), are minimal (see the remark on p. 4 below).

Ideally, one might hope to prove a sign change result for the sequence by combining Theorems 1 and 2 appropriately. However we have not been able to do this.

## 3 Proof of Theorem 1

We shall actually prove a more general statement than the assertion of Theorem 1.

We assume that the Fourier expansion of is given by with . Let be fixed. For , set

 cn\vcentcolon=a(n)a(n+r).

To the sequence , we attach a Dirichlet series [see 1, 6] by setting

 D(s,r)\vcentcolon=∑n≥1cn(n+r2)s(σ\vcentcolon=Re(s)>k). (1)

The theorem we shall prove is the following:

###### Theorem 1′.

Assume that for almost all and that the sum

 D(s,r)\vcentcolon=∑n≥1cn(n+r2)−s(σ>k)

in fact converges for all . Then for all .

We postpone the proof of this Theorem, and give the proof of Theorem 1 first.

###### Proof of Theorem 1.

It suffices to assume that all but a finite number of the coefficients are zero and to derive a contradiction.

Thus, let , with be those which are non-zero. Then, the Dirichlet series from (1) becomes a Dirichlet polynomial

 D(s,r)=p∑i=1cni(ni+r2)s,

and hence converges in the entire -plane. Also, by hypothesis almost all of its coefficients are . Now, by applying Theorem 1, we find that for all . This is a contradiction, and the proof is complete. ∎

The rest of this section is dedicated to the proof of Theorem 1. First, we fix some notation: In the following, denote by a complex variable with real part and imaginary part . By denote a fundamental domain for the action of on , and by the Hilbert space of complex-valued -invariant functions on that are square integrable on with respect to the invariant measure . If and are in we denote the inner product by

 ⟨f,g⟩=∫Ff(z)¯¯¯¯¯¯¯¯¯g(z)dν(z).

Even if and do not both belong to we continue to use the notation for the above integral, as long as it converges absolutely. Finally, set .

###### Proof of Theorem 1′.

We loosely follow the notation used by Good in [1]. Fix a maximal system of -inequivalent cusps, , . (Note that by our assumptions on , this system is finite and contains .)

For , the Poincaré series [see 1, p. 13] is defined as

 P0(z,s,r)=(πr)s−1/2Γ(s+12)∑γ∈Γ∞∖ΓI(γz)se(rR(γz)),

for a complex variable with . Here, stands for the stabilizer of in , consisting of the matrices , .

For each cusp , denote by an element with and for which is the stabilizer of in . Then, the non-holomorphic Eisenstein series of weight zero attached to the cusp (cf. [1, pp. 104f] or e.g. [5, Chapter II]) is defined as

 Ei(z,s)=∑γ∈Γξi∖Γ(I(g−1iγz))s(σ>1).

This sum converges absolutely in the half-plane . The Eisenstein series has meromorphic continuation to the entire -plane. Its Fourier expansion is given by

 Ei(z,s)=δi,1ys+ϕi(s)y1−s+∑m≥1ϕi(m;s)2y1/2Ks−1/2(2π|m|y)e(mx),

with coefficient functions

 ϕi(s)=√π⋅Γ(s+12)Γ(s)L(0)i(s),ϕi(m;s)=πs|m|s−12Γ(s)L(m)i(s),whereL(m)i=∑c>0c−2s∑dmodce(mdc)((∗∗cd)∈Γξi).

Note that, by the functional equation of , one has .

Now, consider for . As a function in , it has holomorphic continuation to a small neighborhood of the line and there, the functional equation holds [cf. 1, Lemma 5, p. 115f]:

 ⟨P0(⋅,s,r),F⟩ =12s−1κ∑i=1ϕi(−r;1−s)⟨Ei(⋅,s),F⟩+⟨P0(⋅,s−1,r),F⟩ (2) +(4π)12−k(n4)s−12Γ(k+s−1)Γ(s+12)∑n≥1cn(n+r2)k+s−1Δr(s,n),

where is given by [see 1, p. 119],

 Δr(s,n)\vcentcolon=1 −F(k+s−12,k+s2,s+12;(r2n+r)2) (3) −Γ(k−s)Γ(s−12)Γ(k+s−1)Γ(12−s)(4n+2rr)2s−1× ×[F(k−s2,k−s+12,32−s;(r2n+r)2)−1].

Here, is the hypergeometric function

 F(a,b,c;z)=∞∑w=0(a)w(b)w(c)ww!zw(|z|<1),(a)w=Γ(a+w)Γ(a).

From the expression in (3) one sees immediately that the series on the right hand side of (2) is convergent at least for .

Also, note that the Dirichlet series satisfies [cf. 1, Lemma 5 (i), p. 116]

 D(s+k−1,r)=(4π)k−12(4r)s−1/2Γ(s+12)Γ(k+s−1)⟨P0(⋅,s,r),F⟩(σ>1). (4)

Since by hypothesis, converges for all , we obtain from (2), (3) and (4) by continuation, that the identity

 (4π)12−k (r4)s−12Γ(k+s−1)Γ(s+12)D(k+s−1,r)= (5) 12s−1κ∑i=1ϕi(−r;1−s)⟨Ei(⋅,s),F⟩ +(4π)12−k(r4)12−sΓ(k−s)Γ(32−s)⋅D(k−s,r) +(4π)12−k(r4)s−12Γ(k+s−1)Γ(s+12)∑n≥1cn(n+r2)k+s−1⋅Δr(s,n)

holds between meromorphic functions for all .

Now, take residues at on both sides of (5). By [5, Theorems 4.3.4, 4.3.5, p. 43], the Eisenstein series are holomorphic in except for finitely many poles in the interval , which are precisely the poles of the constant coefficients . It follows that the are holomorphic at and do not contribute to the residue.

Thus, we pick up non-zero residues only from the second and third terms on the right-hand side of (5); in the third term, the only non-zero contribution comes from the second (non-constant) term of in (3). Note that for .

More precisely, we have

 0=(r4)12−k−2m1Γ(32−k−2m)1(2m)!⋅D(−2m,r)−(r4)k+2m−12Γ(2k+2m−1)Γ(k+2m+12)∑n≥1cn(n+r2)2k+2m−1(2mn)−1⋅Γ(k+2m−12)Γ(2k+2m−1)⋅Γ(12−k−2m)××(4n+2rr)2k+4m−1[F(−m,−m+12,32−k−2m,(r2n+r)2)−1].

We note that

 Γ(32−k−2m) =(12−k−2m)⋅Γ(12−k−2m) and Γ(k+2m+12) =(k+2m−12)⋅Γ(k+2m−12).

Thus, we obtain

 0= (r4)12−k−2m1(12−k−2m)D(−2m,r) −(r4)k+2m−121(k+2m−12)∑n≥1cn(n+r2)2k+2m−1× ×(4n+2rr)2k+4m−1[F(−m,−m+12,32−k−2m;(r2n+r)2)−1].

Whence, further

 0= (r4)12−k−2mD(−2m,r)+(r4)k+2m−12∑n≥1cn(n+r2)2k+2m−1× ×(4n+2rr)2k+2m−1[F(−m,−m+12,32−k−2m;(r2n+r)2)−1].

Also, we have

Therefore altogether, we obtain

 D(−2m,r)=∑n≥1cn(n+r2)2m⋅[−1+F(−m,−m+12,32−k−2m;(r2n+r)2)],

for all . It follows that for all

 0=∑n≥1cn(2n+r)2mF(−m,−m+12,32−k−2m;(r2n+r)2).

By the definition of the hypergeometric function, we have

 F(−m,−m+12,32−k−2m;z)= ∑w≥0(−m)⋯(−m+w−1)(−m+12)⋯(−m+w−1+12)(−k−2m+32)⋯(−k−2m+32+w−1)zww!.

This is a polynomial in of degree , which we denote as . All of its coefficients , are non-zero. Set and write . We find that

 ∑n≥1cnPm(2n+r)=0.

Since all the are non-zero, we find immediately that for all ,

 ∑n≥1cn(2n+r)2ν=0. (6)

Now recall that (by hypothesis) almost all the coefficients are . Trivially, if there are non negative coefficients, since the right hand side is zero, there can be no positive coefficients, either and so for all , in which case we are finished. Thus, we assume that a finite number of coefficients, , with are negative. For all , we can write the equation (6) in the form

 ∑n≥1n≠n1,…,ntcn(2n+r)2ν=−cn1(2n1+r)2ν−…−cnt(2nt+r)2ν

Whence for all ,

 ∑n≥1n≠n1,…,ntcn(2n+r2nt+n)2ν=−cn1(2n1+r2nt+r)2ν−…−cnt.

Here, as grows arbitrarily large, the right-hand side is positive and bounded, whereas,on the left-hand side, there exist with (and with ), thus, the left-hand side becomes arbitrarily large. This is a contradiction, hence for all .

Now, from (6) we get the system of linear equations with

 nt∑n=1cn(2n+r)2ν=0.

the determinant of which is easily seen to be non-zero. It follows that the remaining coefficients vanish, too. Thus, to conclude, for all . ∎

## 4 Proof of Theorem 2

Recall that for this theorem, we require to be a congruence subgroup of level (then, condition (i) from p. 2 is satisfied automatically). For the proof, we need the following lemma.

###### Lemma 1.

Let be cusp form with respect to , with Fourier coefficients . Let . Then, the Fourier expansion

 g(z)\vcentcolon=∑n≥1n≡n0mod2ra(n)e2πinz

defines a cusp form for a congruence subgroup of higher level.

###### Proof.

With the usual -operation, we can write in the form

 g(z) =1ϕ(2r)∑smod2re−πin0srzf∣k(1s2r01) =1ϕ(2r)∑n≥n0a(n)e2πinz∑smod2re2πi(n−n0)s2r

Here, denotes the totient function, and summation runs over a minimal system of representatives . Clearly, the right-hand side is modular with level a multiple of , and is cuspidal if is. ∎

Now, we are ready to prove Theorem 2.

###### Proof.

Set for all . Assume that holds for only finitely many . Then, there is an index , with the property that for all .

Since it follows that and are both non-zero and have opposite sign. By induction, it follows that all coefficients with have the same sign. Now, by Lemma 1, the resulting sequence of coefficients defines a cusp form

 g(z)=∑n≥1n≡n0mod2ra(n)e2πinz

with real Fourier coefficients, of which, by construction only a finite number are less than or equal to zero. However, by the results of Knopp, Kohnen and Pribitkin [4], Theorem 1, if all Fourier coefficients of a cusp form are real, the sequence of its coefficients has infinitely many terms of either sign. This is a contradiction. ∎

###### Remark.

Actually, for the result from [4] we use here, the only requirement is that the group mentioned in Lemma 1, besides being a Fuchsian group of the first kind with finite covolume, have and as parabolic fixed points. However, if already satisfies conditions (i)–(iii) (see p. 2), this imposes only a mild further restriction. Thus, Theorem 2 holds somewhat more generally then only for congruence subgroups.

## References

• [1] Anton Good. On various means involving the Fourier coefficients of cusp forms. Math. Z., 183(1):95–129, 1983.
• [2] Jeff Hofstein, Thomas A. Hulse, and Andre Reznikov. Multiple Dirichlet series and shifted convolutions. ArXiv e-prints, 2015, 1110.4868v3.
• [3] Winfried Hohnen. Period polynomials and congruences for Hecke algebras [eigenvalues]. Proc. Edinburgh Math. Soc. (2), 42(2):217–224, 1999.
• [4] Marvin Knopp, Winfried Kohnen, and Wladimir Pribitkin. On the signs of Fourier coefficients of cusp forms. Ramanujan J., 7(1-3):269–277, 2003. Rankin memorial issues.
• [5] Tomio Kubota. Elementary theory of Eisenstein series. Kodansha Ltd., Tokyo; Halsted Press [John Wiley & Sons], New York-London-Sydney, 1973.
• [6] Atle Selberg. On the estimation of Fourier coefficients of modular forms. In Proc. Sympos. Pure Math., Vol. VIII, pages 1–15. Amer. Math. Soc., Providence, R.I., 1965.
• [7] Jean-Pierre Serre. Valeurs propres des opérateurs de Hecke modulo . In Journées Arithmétiques de Bordeaux (Conf., Univ. Bordeaux, 1974), pages 109–117. Astérisque, Nos. 24–25. Soc. Math. France, Paris, 1975.

Mathematisches Institut, Universität Heidelberg, Im Neuenheimer Feld 205,
D-69120 Heidelberg, Germany

You are adding the first comment!
How to quickly get a good reply:
• Give credit where it’s due by listing out the positive aspects of a paper before getting into which changes should be made.
• Be specific in your critique, and provide supporting evidence with appropriate references to substantiate general statements.
• Your comment should inspire ideas to flow and help the author improves the paper.

The better we are at sharing our knowledge with each other, the faster we move forward.
The feedback must be of minimum 40 characters and the title a minimum of 5 characters