Abstract
Let \(k \ge 2\) and N be positive integers and let \(\chi \) be a Dirichlet character modulo N. Let f(z) be a modular form in \(M_k(\Gamma _0(N),\chi )\). Then we have a unique decomposition \(f(z)=E_f(z)+S_f(z)\), where \(E_f(z) \in E_k(\Gamma _0(N),\chi )\) and \(S_f(z) \in S_k(\Gamma _0(N),\chi )\). In this paper, we give an explicit formula for \(E_f(z)\) in terms of Eisenstein series whose coefficients are sum of divisors function. Then we apply our result to certain families of eta quotients and to representations of positive integers by 2k–ary positive definite quadratic forms in order to give an alternative version of Siegel’s formula for the weighted average number of representations of an integer by quadratic forms in the same genus. Our formula for the latter is in terms of sum of divisors function and does not involve computation of local densities.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction and notation
Let \({\mathbb {N}}\), \({\mathbb {N}}_0\), \({\mathbb {Z}}\), \({\mathbb {Q}}\), \({\mathbb {C}}\), and \({\mathbb {H}}\) denote the sets of positive integers, non-negative integers, integers, rational numbers, complex numbers, and upper half-plane of complex numbers, respectively. Throughout the paper, z denotes a complex number in \({\mathbb {H}}\), p always denotes a prime number, all divisors considered are positive divisors, q stands for \(e^{2 \pi i z}\), and \(\chi _d(n)\) denotes the Kronecker symbol \(\displaystyle {\Bigl ({\frac{d}{n}}\Bigr )_{K} }\), where we use the subscript K to avoid confusion with fractions. Let \(N \in {\mathbb {N}}\) and \(\chi \) be a Dirichlet character modulo N. The space of modular forms of weight k for \(\Gamma _0(N)\) with character \(\chi \) is denoted by \(M_{k}(\Gamma _0(N),\chi )\); and \(E_{k}(\Gamma _0(N),\chi )\), \(S_{k}(\Gamma _0(N),\chi )\) denote its Eisenstein and cusp form subspaces, respectively. Then we have
That is, given \(f(z) \in M_k(\Gamma _0(N),\chi )\), we can write
where \(E_f(z) \in E_k(\Gamma _0(N),\chi )\) and \(S_f(z) \in S_k(\Gamma _0(N),\chi )\) are uniquely determined by f. Let \( \epsilon , \psi \) be primitive Dirichlet characters such that \(\epsilon {\psi } = \chi \) (i.e., \(\epsilon (n) {\psi }(n) = \chi (n)\) for all \(n \in {\mathbb {Z}}\) coprime to N) with conductors say L and M, respectively, and suppose \({LM} \mid N \). Let d be a positive divisor of N/LM and \(2\le k \in {\mathbb {N}}\) be such that \(\chi (-1)=(-1)^k\). Let \(\omega \) be the primitive Dirichlet character corresponding to \(\epsilon {\overline{\psi }}\) and \({\mathcal {M}}_\omega \) be its conductor. We define the Eisenstein series associated with \(\epsilon \) and \({\psi }\) by
where \({\overline{\omega }}\) is the complex conjugate of \(\omega \),
is the generalized sum of divisors function associated with \(\epsilon \) and \(\psi \),
is the Gauss sum of \(\psi \), and \(B_{k,{\overline{\omega }}}\) is the k-th generalized Bernoulli number associated with \({\overline{\omega }}\) defined by
see [13, end of pg. 94]. By [5, Corollary 8.5.5] (alternatively [13, Theorem 4.7.1, (7.1.13) and Lemma 7.2.19]), we have
and
Remark 1
Let \(L(\epsilon {\overline{\psi }}, k)\) be the Dirichlet L-function defined by
The Eisenstein series we define in (2) is equal to \(\displaystyle {E_k(Mdz;\epsilon ,{\overline{\psi }})}/({2 L(\epsilon {\overline{\psi }}, k)})\) in the notation of Theorem 7.1.3 and Theorem 7.2.12 of [13], and it is also equal to \(\displaystyle {G_k({\psi },\epsilon )(dz)}/{L(\epsilon {\overline{\psi }}, k)}\) in the notation of Corollary 8.5.5 and Definition 8.5.10 of [5]. Further treatment to obtain the form in (2) is done by using the formula for \(L(\epsilon {\overline{\psi }}, k)\) given in Theorems 3.3.4 and (3.3.14) of [13]. We have normalized the Eisenstein series this way to simplify the constant terms given in (27) and (28). This in return simplifies the notation in Sects. 4 and 5.
Letting \(D(N,{\mathbb {C}})\) to denote the group of Dirichlet characters modulo N, we define
The set
constitutes a basis for \(E_k(\Gamma _0(N),\chi )\) whenever \(k \ge 2\) and \((k,\chi ) \ne (2,\chi _1)\); the set
constitutes a basis for \(E_2(\Gamma _0(N),\chi _1)\), see [5, Theorems 8.5.17 and 8.5.22], or [22, Proposition 5]. Then we have
for some \(a_f(\epsilon ,\psi ,d) \in {\mathbb {C}}\). When \(S_f(z)=0\), it is easy to determine \(a_f(\epsilon ,\psi ,d)\) by comparing the first few Fourier coefficients of f(z) expanded at \(i \infty \) and the first few Fourier coefficients of the right-hand side of (3) expanded at \(i \infty \). However, if \(S_f(z)\ne 0\) and an explicit basis for \(S_k(\Gamma _0(N),\chi )\) is not known then this method fails. In this paper we solve this problem, in other words we obtain \(a_f(\epsilon ,\psi ,d)\) explicitly (in terms of a finite sum) for all \(f \in M_k(\Gamma _0(N),\chi )\), where \(k \ge 2\), see Theorem 1. Our treatment is general and its special cases agree with the previously known formulas. Additionally, we give a new treatment of Siegel’s formula for representation numbers of quadratic forms, see Theorem 2.
Let \(a \in {\mathbb {Z}}\) and \(c \in {\mathbb {N}}_0\) be coprime. For an \(f(z) \in M_{k}(\Gamma _0(N),\chi )\), we denote the constant term of f(z) in the Fourier expansion of f(z) at the cusp a/c by
where \(b,d \in {\mathbb {Z}}\) such that \(\begin{bmatrix} a &{} b \\ c &{} d \end{bmatrix} \in SL_2({\mathbb {Z}})\). The value of \([0]_{a/c} f \) does not depend on the choice of b, d. We denote the nth Fourier coefficient of f(z) in the expansion at the cusp \(i\infty \) by [n]f. Letting \(\phi (n)\) denote the Euler totient function, we define an average associated with \(\psi \) for the constant terms of Fourier series expansions of modular forms at cusps as follows:
We note that working with this average of constant terms at cusps is a new idea which helps studying modular form spaces with non-trivial character, see Sect. 5 for details.
Letting \(v_p(n)\) to denote the highest power of p dividing n and \(\mu (n)\) be the Möbius function we are ready to state the main theorem.
Theorem 1
(Main Theorem) Let \(f(z) \in M_k(\Gamma _0(N),\chi )\), where \(N,k \in {\mathbb {N}}\), \(k \ge 2\), \(\chi \) is a Dirichlet character modulo N that satisfies \(\chi (-1)=(-1)^k\). Let \(E_f(z)\) be defined by (1). Then
where
with
and
Remark 2
Let \(c \mid N\). By Lemma 6, if a/c and \(a'/c\) are equivalent cusps of \(\Gamma _0(N)\) and \((\epsilon ,\psi ) \in {\mathcal {E}}(k,N,\chi )\) with \(M \mid c\), then \( \psi (a)[0]_{a/c} f =\psi (a')[0]_{a'/c} f\). Therefore, in the applications of Theorem 1 computing \( \psi (a)[0]_{a/c} f\) at a set of inequivalent cusps will be sufficient, see [5, Corollary 6.3.23] for a description of such a set.
Theorem 1 agrees with and extends the previously known formulas. For example, if we let \(k \in {\mathbb {N}}\) even, N squarefree, \(\chi =\chi _1\) in Theorem 1, we obtain [3, Theorem 1.1] and if we let \(k \in {\mathbb {N}}\) odd \(N \in \{3,7,11,23 \}\), \(\chi =\chi _{-N}\) in Theorem 1, we obtain [6, (11.20)]. Theorem 1 additionally extends the latter to hold for all primes N that are congruent to 3 modulo 4.
Before we apply Theorem 1 to representation numbers of quadratic forms, we give a snapshot of interesting applications. Since \(f(z) -E_f(z)\) is a cusp form, one can use our main theorem to produce cusp forms. At the end of Sect. 3 we use this idea combined with the Modularity Theorem ( [9, Theorem 8.8.1]) and consider the elliptic curve \(E_{27A}: y^2 + y=x^3 - 7\). Then we use arithmetic properties of Eisenstein series to obtain
where
with \({\mathbb {F}}_p\) denoting the finite field of p elements, see Corollary 3.
The Fourier coefficients of special functions (expanded at \(i \infty \)) have been of huge interest. A very well studied special function is the Dedekind eta function which is defined by
Quotients of Dedekind eta functions are often referred to as eta quotients. Nathan Fine in his book [10] has given several formulas for Fourier coefficients of eta quotients (expanded at \(i \infty \)). In his work when the weight of the eta quotient is an integer, the formulas are linear combinations of Eisenstein series defined above. For instance he shows that
see [10, (32.5)], and acknowledges this equation as being very beautiful. We consider the \((2k+1)\)th power of F(z), that is we consider
Using our main theorem (Theorem 1), we obtain the following analogous formula for \(F^{2k+1}(z)\) when \(k > 0\):
see Corollary 2. Using Theorem 1 one can obtain formulas in this fashion for all holomorphic eta quotients of integral weight \(k \ge 2\).
Let \({\mathcal {F}}(x_1,\ldots ,x_{2k})\) be a positive definite quadratic form with integer coefficients and \(B({\mathcal {F}})\) be the matrix associated with \({\mathcal {F}}\) whose entries are given by
Then the generating function of the number of representations of a positive integer by the quadratic form \({\mathcal {F}}\) is
In [19], Siegel gave a formula for the weighted average for representation numbers of positive definite quadratic forms in the same genus. Siegel’s formula is in terms of local densities (for other treatments of Siegel’s formula see [23] and [15, Chapter 3]). In the realm of modular forms, Siegel’s formula corresponds to the Eisenstein part of \(\theta _{\mathcal {F}}(z)\), see [1, 17, 18, Remark on p. 110] and [19]. For clarity we note that if \({\mathcal {F}}_1\) and \({\mathcal {F}}_2\) are in the same genus then \(E_{\theta _{{\mathcal {F}}_1}}(z) = E_{\theta _{{\mathcal {F}}_2}}(z)\). Below we use Theorem 1 to give an explicit formula for \(E_{\theta _{{\mathcal {F}}}}(z)\), where \({\mathcal {F}}\) is a 2k–ary positive definite quadratic form with integer coefficients. In Sect. 2, we give several applications of our formula including a comparison of our output for the form \(\sum _{j=1}^{2k} x_j^2\) with that of Arenas [1, Proposition 1], which uses Siegel’s formula.
By [13, Corollary 4.9.5], we have
where \(\displaystyle \chi =({(-1)^k \det (B({\mathcal {F}}))}/{*})_{K}\) and N is the smallest positive integer such that the matrix \(NB({\mathcal {F}})^{-1}\) has even diagonal entries. By [21, (10.2)] we have
Putting (2), (8), and (9) in Theorem 1, we obtain the following assertion concerning the representation numbers of 2k–ary quadratic forms.
Theorem 2
Let \({\mathcal {F}}(x_1,\ldots ,x_{2k})\) be a positive definite quadratic form with \(k \ge 2\); let \(\chi \) and N be as above and \(\omega \) be as in (2). Then
where
The organization of the rest of the paper is as follows. In Sect. 2, we apply Theorem 2 to the representation numbers of diagonal quadratic forms and certain non-diagonal level 2 quadratic forms. A special case of the latter leads to an equation for Ramanujan’s \(\tau \)-function. In Sect. 3, we apply our Main Theorem to certain families of eta quotients, these applications give extensions of some well-known formulas to higher weight eta quotients. In Sects. 4–6, we prove the main theorem.
2 Applications to representation numbers of certain quadratic forms
To apply (10) to specific quadratic forms we need to compute the quadratic Gauss sum. If \({\mathcal {F}}\) is a diagonal form, say \({\mathcal {F}}=\sum _{j=1}^{2k} \alpha _j x_j^2\), then we have
where, if \(\gcd (\alpha , \beta )=1\),
see [4, Theorems 1.5.2 and 1.5.4]. Next we apply this result to the form \({\mathcal {F}}=\sum _{j=1}^{2k} x_j^2\), that is, \(\alpha _j=1\) for all \(1 \le j \le 2k\). Then we have
Thus, by Theorem 2 when k is even we have
and when k is odd we have
Formulas (12) and (13) agree with Ramanujan’s statements [16, (131)–(134)], which was first proven by Mordell in [14]. In [1, Proposition 1] Arenas uses Siegel’s formula to compute \(E_{\theta _{{\mathcal {F}}}}(z)\) and obtains (12) and (13) in the same form. Now, we turn our attention to another diagonal form. Let
In [7] Cooper, Kane, and Ye found formulas for the representation numbers of \({\mathcal {F}}(k,k;p)\), where \(p=3,7,11\), or 23. Their result relies on the existence of a Hauptmodul in the levels considered. Inspired by their results, in [2], we derived formulas for the representation numbers of \({\mathcal {F}}(2a,2b;p)\) where \(a,b \in {\mathbb {N}}_0\) and p is an odd prime. These results are considered as analogs of the Ramanujan–Mordell formula and specialized version of Theorem 2 agrees with these results. Below we give formulas in all the remaining cases, that is, we find formulas for representation numbers of \({\mathcal {F}}(a,b;p)\) where \(a,b \equiv 1 \pmod {2}\) and p an odd prime.
Corollary 1
Let \(a,b \ge 1\) be odd integers such that \(a+b \ge 4\). Set \(k=(a+b)/2\) and \({\mathbf {p}}=\chi _{-4}(p) p\). Then for any odd prime p, whenever \((-1)^k = \chi _{-4}(p)\) we have
and whenever \(p \equiv (-1)^{k+1} \pmod {4}\) we have
where
Using (11) and Theorem 2 one can obtain results similar to Corollary 1 for any diagonal form. Next we consider the non-diagonal form
We obtain
Thus, \(\theta _{{\mathcal {F}}_k} \in M_{2k}(\Gamma _0(2),\chi _1)\), hence by Theorem 2 we have
When \(k=6\) we compute the first few coefficients of the cusp part of \(\theta _{{\mathcal {F}}_6}\):
The Fourier coefficients of \(\eta ^{24}(z)\) are called Ramanujan’s \(\tau \)-function and first few terms are given as follows:
It is well known that \(\eta ^{24}(z)\) and \(\eta ^{24}(2z) \in S_{12}(\Gamma _0(2),\chi _1)\) and thus by Sturm’s Theorem [5, Corollary 5.6.14] we obtain
If we compare nth coefficient of both sides of (16) we get
Since
it is not hard to deduce the well-known congruence relation
3 Applications to eta quotients
In this section, we give further applications of Theorem 1. Recall that the Dedekind eta function is defined by
Let \(k \in {\mathbb {N}}\). We define
In Corollary 2 below, we obtain formulas concerning \(f_k(z), g_k(z)\), and \(h_k(z)\). Similar formulas can be obtained via Theorem 1 for all integer weight holomorphic eta quotients.
Corollary 2
Let \(k \ge 1\) and let \(f_k(z), g_k(z)\), and \(h_k(z)\) be defined by (17), (18), and (19), respectively. Then we have
and
where
Proof
We use [5, Proposition 5.9.2] to determine
We evaluate the constant terms of \(f_k(z),g_k(z),h_k(z)\) at the relevant cusps using [12, Proposition 2.1]. We do this with the help of some SAGE functions we have written; the code is provided in the Appendix A. From these we compute
see Appendix A for details. We determine the set of tuples of characters as
Thus, we have
Now, we compute
Additionally, we have
Combining (21) and (22) we have \(a_{f_k}(\chi _{1},\chi _{-24},1)=1\).
The rest of the coefficients are obtained similarly. \(\square \)
Now, we turn our attention to special cases of these formulas. The dimension of \(S_{2}(\Gamma _0(12),\chi _{12})\) is 0, so we obtain an exact formula for \(g_1\), i.e., we have \(g_{1}(z)=E_{g_{1}}(z)\). When \(k=1\), (20) specializes to
Clearly \(h_1(z)-E_{h_1}(z)\) is a cusp form, and if we normalize \(h_1(z)-E_{h_1}(z)\) so that the coefficient of q is 1, we obtain the newform \({\mathcal {N}}_{27}(z)\) in \(S_{2}(\Gamma _0(27),\chi _1)\), that is, we have
By [8, Table 1] this newform is associated to the elliptic curve
Recall that in Sect. 1 we defined
where \({\mathbb {F}}_p\) is the finite field of p elements. Then by the Modularity Theorem, see [9, Theorem 8.8.1], we have
Thus, for all \(p \ne 3\) we have
Since \(\displaystyle [p]\frac{\eta ^{2}(9z) \eta ^{3}(27z)}{\eta (3z)} \in {\mathbb {Z}}\) for all \(p \in {\mathbb {N}}\) and \(\displaystyle [p]\frac{\eta ^{2}(9z) \eta ^{3}(27z)}{\eta (3z)} =0\) when \(p \equiv 2 \pmod {3}\), we obtain the following statement.
Corollary 3
We have
4 Orthogonal relations
In this section, we prove some orthogonal relations involving the functions \({\mathcal {R}}_{k,\epsilon ,\psi }(d,c) \) and \( {\mathcal {S}}_{k,N,\epsilon ,\psi }(d,c)\) defined in (6) and (7), respectively. These orthogonal relations concern the constant terms of the Eisenstein series and give the means to determine \(a_f(\epsilon ,\psi ,d)\) of Theorem 1. Throughout the section we assume \(k,N \in {\mathbb {N}}\) and \(\epsilon , \psi \) are primitive Dirichlet characters with conductors L, M, respectively, such that \(LM \mid N\).
Lemma 1
Let \(p \mid N\) be a prime and let \(t \mid N/p^v\), where \(v=v_p(N)\). Then, for \(0\le i \le v\), we have
Proof
Since \(t \mid N/p^v\) we have \(\gcd (t,p)=1\). Using the multiplicative properties of the Möbius function we obtain
\(\square \)
Lemma 2
If \(\gcd (t,p^i)=1\), then we have
Proof
By elementary manipulations we obtain
Proof of the second equation is similar. \(\square \)
Theorem 3
If \(c,d \mid N\), then
Proof
Let \(p \mid N\) be prime and \(v_p(N)=v\). We use Lemmas 1 and 2 to obtain
Using this recursively we obtain
Now, we prove for all \(p \mid N\) we have
We first note that
and
The cases
-
(Case 1) \(0< v_p(c) < v_p(d) \le v\),
-
(Case 2) \(0 = v_p(c) < v_p(d) \le v\),
-
(Case 3) \(v> v_p(c) > v_p(d) \ge 0\),
-
(Case 4) \(v = v_p(c) > v_p(d) \ge 0\),
-
(Case 5) \(0< v_p(c) = v_p(d) < v\),
-
(Case 6) \(0 = v_p(c) = v_p(d)\),
-
(Case 7) \(v= v_p(c) = v_p(d)\),
need to be handled separately, which is done below.
Case 1 If \(0< v_p(c) < v_p(d) \le v\), then by employing (25) for all i such that \(\vert v_p(c) - i \vert >1\) we have
Therefore, we have
which, by (25) and (26), equals to
By using multiplicative properties of Dirichlet characters we conclude that this expression is equal to 0.
Case 2 If \(0 = v_p(c) < v_p(d) \le v\), then by employing (25) and (26) we have
which equals to 0.
Case 3 If \(v> v_p(c) > v_p(d) \ge 0\), then by employing (25) and (26) we have
By using multiplicative properties of Dirichlet characters we conclude that this expression is equal to 0.
Case 4 If \(v = v_p(c) > v_p(d) \ge 0\), then by employing (25) and (26) we have
Case 5 If \(0< v_p(c) = v_p(d) < v\), then by employing (25), (26) and multiplicative properties of Dirichlet characters we have
Case 6 If \(0=v_p(c) = v_p(d) \), then by employing (25), (26) and multiplicative properties of Dirichlet characters we have
Case 7 If \(v=v_p(c) = v_p(d)\), then by employing (25), (26) and multiplicative properties of Dirichlet characters we have
This yields (24). Finally, if \(c \ne d\), then there exists a prime \(p \mid N\) such that \(v_p(c) \ne v_p(d)\). Hence by (24) the product in (23) is 0. If \(c=d\) then for all prime divisors p of N we have \(v_p(c) = v_p(d)\). Therefore, by (23) and (24) we have the desired result. \(\square \)
5 Constant terms of expansions of Eisenstein series at the cusps
Recall that \(E_{k}(\epsilon ,\psi ;dz)\) is defined by (2) and we have
and
The constant terms of Eisenstein series in the expansion at the cusp a/c with \(\gcd (a,c)=1\) are given by
where \({\mathcal {R}}_{k,\epsilon ,\psi } (c,t) \) is defined by (6). For (27) see [3, (6.2)], [5, Proposition 8.5.6 and Ex. 8.7 (i) on pg. 308]. The formula (28) is proved later in this section.
The structure of the terms \([0]_{a/c}E_k(\epsilon ,\psi ;dz)\) is complicated and difficult to work with. We observe that taking the average \([0]_{c,{\psi }} E_k( \epsilon ,\psi ;dz)\) gives constant terms a very nice structure which is easier to work with (see (38)). Throughout the section we assume \(k,N \in {\mathbb {N}}\), \(\epsilon \) and \(\psi \) are primitive Dirichlet characters with conductors L and M, respectively, such that \(LM \mid N\) and \((k,\epsilon ,\psi ) \ne (2,\chi _1,\chi _1)\).
Lemma 3
Let \(c \mid N\) and \(LMd \mid N\). If \(M \not \mid c\), or \(M \mid c\) and \(L \not \mid N/c\), then
Proof
First, assuming \(M \not \mid c\), we see that \( M \not \mid \gcd (Md,c) \). Thus, \( \gcd \left( \frac{Md}{\gcd (Md,c)},M \right) \mid M\), which implies \( {\overline{\psi }} \left( \frac{Md}{\gcd (Md,c)} \right) =0\) (since the conductor of \(\psi \) is M). Therefore, the result follows from (27).
Second, we assume \(M \mid c\) and \(L \not \mid N/c\). Setting \(c_1=c/M\), we have \(c_1 \mid N/M\). Since \(L \not \mid N/c\), we have \(c_1 \not \mid N/LM\). Since \(\frac{(N/M)}{L} \in {\mathbb {Z}}\), \(\frac{(N/M)}{c_1} \in {\mathbb {Z}}\) and \(\frac{(N/M)/c_1}{L} \not \in {\mathbb {Z}}\), we have \(\gcd (c_1,L) \ne 1\). Additionally, there exists a prime p dividing \(c_1\) such that
Since \(c_1 \mid N/M\), for all \(p \mid c_1\) we have
Therefore,
Since \(d \mid N/LM\), we have
Inequalities (29) and (32) together imply \(v_p(c_1) > v_p(d)\). Employing (31) we have
That is,
This implies \(\displaystyle \epsilon \left( \frac{c}{\gcd (Md,c)} \right) =0\) (since the conductor of \(\epsilon \) is L). Therefore, the result follows from (27). \(\square \)
Next we need the orthogonality of characters. The following lemma is a result of standard Schur orthogonality relations for the characters on the unit group \(({\mathbb {Z}}/c{\mathbb {Z}})^\times \) (see [5, Proposition 3.4.2]).
Lemma 4
Let \(c \in {\mathbb {N}}\), and let \(\psi _1,\psi _2\) be two primitive Dirichlet characters with conductors \(M_1\) and \(M_2\), respectively. If both \(M_1\) and \(M_2\) divide c, then we have
Before we prove the main result of this section we prove (28).
Lemma 5
If \(\gcd (a,c)=1\), then we have
Proof
Since \(\gcd (a,c)=1\), there exist \(\beta , \gamma \in {\mathbb {Z}}\) such that \(A=\begin{bmatrix} a &{} \beta \\ c &{} \gamma \end{bmatrix} \in SL_2({\mathbb {Z}})\). Then by [12, (1.21)] we have
where A(z) is the usual linear fractional transformation. Let \(e= \frac{a d }{\gcd (c,a d )}\) and \(g=\frac{c}{\gcd (c,a d )}\). Since \(\gcd (e,g)=1\) there exist f, h such that \( \begin{bmatrix} e &{} f \\ g &{} h \end{bmatrix} \in SL_2(Z)\). Hence we have
where in the last line we used (33). Thus, we obtain
\(\square \)
Theorem 4
If \(c \mid N\) and \((\epsilon _1,\psi _1),(\epsilon _2,\psi _2) \in \{ (\epsilon ,\psi ) \in {\mathcal {E}}(k,N,\chi ) : M \mid c \}\), then we have
If \(c \mid N\) and \((\epsilon _2,\psi _2) \in \{ (\epsilon ,\psi ) \in {\mathcal {E}}(2,N,\chi _1) : M \mid c \}\), then we have
Proof
If \((k, \epsilon , \psi ) \ne (2,\chi _1,\chi _1)\) by (27) we have
Therefore, by Lemma 4 we obtain the first part of the statement. Proof of the second part is similar. \(\square \)
6 Proof of the main theorem
Recall that \(E_{k}(\epsilon ,\psi ;dz)\) is defined by (2) and the set
constitutes a basis for \(E_k(\Gamma _0(N),\chi )\) whenever \((k,\chi ) \ne (2,\chi _1)\) and the set
constitutes a basis for \(E_2(\Gamma _0(N),\chi _1)\), see [5, Theorems 8.5.17 and 8.5.22], or [22, Proposition 5].
Now, we prove the main theorem whenever \((k,\chi ) \ne (2,\chi _1)\). Let \(f(z) \in M_k(\Gamma _0(N),\chi )\) where \(N,k \in {\mathbb {N}}\), \(k \ge 2\) and \((k,\chi ) \ne (2,\chi _1)\). By (34) we have
for some \(a_f(\epsilon ,\psi ,d) \in {\mathbb {C}}\). Our strategy for the proof is, using the interplay between the constant terms of Eisenstein series, to create sets of linear equations (see (38)) and to solve those sets of linear equations for \(a_f(\epsilon ,\psi ,d)\) using Theorem 3.
By (1) we have \(f(z)=E_f(z)+S_f(z)\), where \(E_f(z) \in E_k(\Gamma _0(N),\chi )\) and \(S_f(z) \in S_k(\Gamma _0(N),\chi )\) are unique. Since by definition \(S_f(z)\) vanishes at all cusps, we have \([0]_{a/c}f(z)=[0]_{a/c}E_f(z)\). Therefore, by (36) for each \(c \mid N\) and \(a \in {\mathbb {Z}}\) such that \(\gcd (a,c)=1\), we obtain
Let \((\epsilon _2,\psi _2) \in {\mathcal {E}}(k,N,\chi )\), and let the conductors of \(\epsilon _2\) and \(\psi _2\) be \(L_2\) and \(M_2\), respectively. Note that for each \(\psi _2\) there is a unique \(\epsilon _2\) such that \((\epsilon _2,\psi _2) \in {\mathcal {E}}(k,N,\chi )\). If we average the constant terms with \(\psi _2\) using (4), then for all \(c \mid N\) we obtain
Our goal here is to isolate a set of linear equations from which we can determine \(a_f(\epsilon _2,\psi _2,d)\) for all \(d \mid N/L_2M_2\). By Lemma 3 we have \([0]_{c, {\psi _2}} E_k(\epsilon _2,\psi _2;dz)=0\), if \(c \mid N\) is such that \(M_2 \mid c\), or \(M_2 \not \mid c\) and \(L_2 \mid N/c\). Therefore, from now on we restrict c to be in \(C_{N}(\epsilon _2,\psi _2)\), see (5) for definition. By applying Lemma 3 one more time we have \([0]_{c,\psi _2} E_k(\epsilon ,\psi ;dz)=0\) if \(M \not \mid c\). Therefore, for all \(c \in C_{N}(\epsilon _2,\psi _2)\) we have
Recall that for each \(\psi _2\) there is a unique \((\epsilon _2,\psi _2) \in {\mathcal {E}}(k,N,\chi )\). Additionally, for all \(c \in C_{N}(\epsilon _2,\psi _2)\) we have \((\epsilon _2,\psi _2) \in \{(\epsilon ,\psi ) \in {\mathcal {E}}(k,N,\chi ) : M \mid c \}\). Therefore, for all \(c \in C_{N}(\epsilon _2,\psi _2)\) we have
From this, using Theorem 4, we obtain
Since \(M_2 \mid c\) we have
Hence for all \(c \in C_{N}(\epsilon _2,\psi _2)\) we have
Below we solve the equations coming from (38) for \(a_f(\epsilon _2,\psi _2,d)\) using Theorem 3. For \(d_2 \mid N/L_2M_2\) we consider the sum
which, by (38), equals to
Rearranging the terms of (40) we obtain
Recall that \(C_{N}(\epsilon _2,\psi _2)\) is defined by (5) and is a set equivalent to the set
i.e., \(c/M_2\) runs through all the divisors of \( N/L_2M_2\) as c runs through all the elements of \(C_{N}(\epsilon _2,\psi _2)\). In Theorem 3 we use this and we replace N by \(N/L_2M_2\), t by \(c/M_2\), c by \(d_2\) and d by d to obtain
Therefore, from (41) and (42) we obtain
Since \(p \mid L_2 M_2\) implies \(\epsilon _2(p) \overline{\psi _2}(p) =0\) we have
This completes the proof of Theorem 1 when \((k,\chi ) \ne (2, \chi _1)\).
Now, if \((k,\chi ) = (2, \chi _1)\), then a basis of \(E_2(\Gamma _0(N),\chi _1)\) is given by (35). Using Lemma 5, Theorem 4, and arguments similar to the first part of this proof we obtain
where \(a_f(\epsilon ,\psi ,d)\) is as above (with \(k=2\)) and
On the other hand, we have
since
i.e., \(\sum _{1< d \mid N} c_f(\chi _1,\chi _1,d)=a_f(\chi _1,\chi _1,1)\). This completes the proof of the Main Theorem.
At last we prove a lemma which is useful in reducing the number of constant term computations in the applications of Theorem 1.
Lemma 6
Let \(f(z) \in M_k(\Gamma _0(N),\chi )\) and \(c \mid N\). Let a/c and \(a'/c\) be equivalent cusps of \(\Gamma _0(N)\). If \((\epsilon ,\psi ) \in {\mathcal {E}}(k,N,\chi )\) with \(M \mid c\), then we have
Proof
If a/c and \(a'/c\) are equivalent cusps of \(\Gamma _0(N)\), then there exists a matrix \(\begin{bmatrix} \alpha &{} \beta \\ \gamma &{} \delta \end{bmatrix} \in \Gamma _0(N)\) such that
Then using transformation properties of modular forms we have
We have \(M\mid c\) and by (43) we have \(a'=\alpha a + \beta c\), thus \(\psi (a')=\psi (\alpha )\psi ( a)\). Since \(M \mid c\), \(c \mid N\), and \(N \mid \gamma \), we have \(M \mid \gamma \), therefore, we have \( 1= \psi (1)= \psi ( \alpha \delta - \gamma \beta )\) which implies \(\psi (\alpha ) = {\overline{\psi }}(\delta )\). Putting these together, we obtain
Since \(\gcd (\delta ,N)=1\) we have \({\overline{\psi }}(\delta ) \chi (\delta )=\epsilon (\delta )\). Now, we prove \(\epsilon (\delta )=1\) which finishes the proof. Recall that \(LM \mid N\), therefore, \(c \mid M\) implies \(L \mid N/c\), i.e., \(L \mid \gamma /c\). From (43) we have \(\delta = 1 - a \gamma /c\), thus, since \(\gcd (a,c)=1\) and \(L \mid \gamma /c\), we have \(\epsilon (\delta )=\epsilon (1 - a \gamma /c)=\epsilon (1)=1\). \(\square \)
References
Arenas, A.: Quantitative aspects of the representations of integers by quadratic forms. In: Number Theory, Alemania, ISBN 3-11-011791-6, pp. 7–14 (1989)
Aygin, Z.S.: Extensions of Ramanujan–Mordell formula with coefficients \(1\) and \(p\). J. Math. Anal. Appl. 465, 690–702 (2018)
Aygin, Z.S.: On Eisenstein series in \(M_{2k}(\Gamma _0(N))\) and their applications. J. Number Theory 195, 358–375 (2019)
Berndt, B.C., Evans, R.J., Williams, K.S.: Gauss and Jacobi Sums. Wiley-Interscience, New York (1998)
Cohen, H., Strömberg, F.: Modular Forms A Classical Approach. Graduate studies in Mathematics. American Mathematical Society, Providence, RI (2017)
Cooper, S.: Ramanujan’s Theta Functions. Springer International Publishing AG, Switzerland (2017)
Cooper, S., Kane, B., Ye, D.: Analogues of the Ramanujan–Mordell theorem. J. Math. Anal. Appl. 446, 568–579 (2017)
Cremona, J.E.: Algorithms for Modular Elliptic Curves. Cambridge University Press, Cambridge (1992)
Diamond, F., Shurman, J.: A First Course in Modular Forms. Graduate Texts in Mathematics, vol. 228. Springer-Verlag, New York (2004)
Fine, N.: Basic Hypergeometric Series and Applications. American Mathematical Society, Providence, RI (1988)
Iwaniec, H.: Topics in Classical Automorphic Forms. Grad. Stud. Math., vol. 17. American Mathematical Society, Providence, RI (1997)
Köhler, G.: Eta Products and Theta Series Identities. Springer Monographs in Mathematics, Springer, New York (2011)
Miyake, T.: Modular Forms, Springer-Verlag, Berlin (1989) (translated from the Japanese by Yoshitaka Maeda)
Mordell, L.J.: On the representations of numbers as a sum of \(2r\) squares. Quart. J. Pure Appl. Math. 48, 93–104 (1917)
Opitz, S.: Computation of Eisenstein series associated with discriminant forms. Doctoral thesis. https://tuprints.ulb.tu-darmstadt.de/8261/1/20181203_Dissertation_Sebastian_Opitz.pdf
Ramanujan, S.: On certain arithmetical functions. Trans. Cambridge Philos. Soc. 22, 159–184 (1916)
Schulze-Pillot, R.: Representation of quadratic forms by integral quadratic forms. In: Alladi, K., Bhargava, M., Savitt, D., Tiep, P. (eds.) Quadratic and Higher Degree Forms. Developments in Mathematics, vol. 31. Springer, New York (2013)
Serre, J.-P.: A Course in Arithmetic. Graduate Texts in Mathematics, Springer-Verlag, New York (1973)
Siegel, C.L.: Uber die analytische theorie der quadratischen formen. Ann. Math. 36, 527–606 (1935)
The Sage Developers: SageMath. The Sage Mathematics Software System (Version 9.1). https://www.sagemath.org (2021)
Wang, X., Pei, D.: Modular Forms with Integral and Half-integral Weights. Science Press Beijing and Springer-Verlag, Berlin, Heidelberg (2012)
Weisinger, J.: Some results on classical Eisenstein series and modular forms over function fields. Thesis (Ph.D.), Harvard University, ProQuest LLC, Ann Arbor, MI (1977)
Yang, T.: An explicit formula for local densities of quadratic forms. J. Number Theory 72, 309–356 (1998)
Acknowledgements
I would like to thank the referee for pointing out the missing cases in the proof of Theorem 3, for suggesting a less ambiguous way to write the quadratic form \({\mathcal {F}}_k\) of Sect. 2 and for numerous suggestions that improved wording. I would like to thank Professor Amir Akbary for helpful discussions throughout the course of this research. I am also grateful to Professor Shaun Cooper, who gave the vision which initiated this research. Words cannot adequately express my gratitude towards Professor Emeritus Kenneth S. Williams, who has given many useful suggestions on an earlier version of this manuscript.
Author information
Authors and Affiliations
Additional information
Dedicated to Professor Emeritus Kenneth S. Williams on the occasion of his 80th birthday.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Research of the author is supported by a Pacific Institute for the Mathematical Sciences postdoctoral fellowship.
Appendix A The SAGE functions for computing the constant terms of eta quotients at a given cusp
Appendix A The SAGE functions for computing the constant terms of eta quotients at a given cusp
Let \(r_d \in {\mathbb {Z}}\), not all zeros, \(N \in {\mathbb {N}}\) and define
Assuming f(z) to be a modular form the following SAGE functions (written using version 9.1 of the software [20]) help computing \([0]_{a/c} f\), the constant term of f(z) at the cusp a/c.
By Lemma 6 it will be sufficient to compute the constant terms of the eta quotient \(f_k(z)\) defined by (17) at a set of inequivalent cusps of \(\Gamma _0(24)\), which is done below with the help of this code. The set
gives a complete set of inequivalent cusps of \(\Gamma _0(24)\), see [5, Corollary 6.3.23]. Note that if k is fixed then the code can handle the vanishing order analysis. For instance the output for the code
will be 0. However, here we are working with a general k, and therefore, the order analysis has to be done manually. When \(k \ge 1\), the vanishing order of \(f_k(z)\) is greater than 0 at cusps \(\{1/2,1/3,1/4,1/6,1/8,1/12\}\). Thus, we have
To compute \([0]_{1/1} f_k\) and \([0]_{1/24} f_k\), we run the following code:
The output will be
Simplifying these we obtain
Putting everything together, for all \(k \ge 1\) we have
Rights and permissions
About this article
Cite this article
Aygin, Z.S. Projections of modular forms on Eisenstein series and its application to Siegel’s formula. Ramanujan J 57, 1223–1252 (2022). https://doi.org/10.1007/s11139-021-00537-1
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11139-021-00537-1