Open Access

Congruences for Taylor expansions of quantum modular forms

Research in the Mathematical Sciences20141:17

DOI: 10.1186/s40687-014-0017-2

Received: 4 November 2014

Accepted: 10 November 2014

Published: 4 December 2014

Abstract

Recently, a beautiful paper of Andrews and Sellers has established linear congruences for the Fishburn numbers modulo an infinite set of primes. Since then, a number of authors have proven refined results, for example, extending all of these congruences to arbitrary powers of the primes involved. Here, we take a different perspective and explain the general theory of such congruences in the context of an important class of quantum modular forms. As one example, we obtain an infinite series of combinatorial sequences connected to the ‘half-derivatives’ of the Andrews-Gordon functions and with Kashaev’s invariant on (2m+1,2) torus knots, and we prove conditions under which the sequences satisfy linear congruences modulo at least 50% of primes.

Introduction and statement of results

In his seminal 2010 Clay lecture, Zagier defined a new class of functions with certain automorphic properties called ‘quantum modular forms’ [39]. Roughly speaking, these are complex-valued functions defined on the rational numbers which have modular transformations modulo ‘nice’ functions. Although the definition is (intentionally) vague, Zagier gave a handful of motivating examples to serve as prototypes. For example, he defined quantum modular forms related to Maass cusp forms attached to Hecke characters of real quadratic fields, as studied by Andrews, Dyson, and Hickerson [2] and Cohen [14], and he gave examples related to sums over quadratic polynomials and non-holomorphic Eichler integrals. More precisely, Zagier made the following definition.

Definition.

A quantum modular form is a function f : P 1 ( Q ) C for which f(x)−f| k γ(x) is ‘suitably nice’.

Here | k is the usual Petersson slash operator, and ‘suitably nice’ means that the obstruction to modularity satisfies an appropriate analyticity condition, e.g., C k , C , etc. One of the most striking examples of a quantum modular form is described in Zagier’s paper on Vassiliev invariants [40], in which he studies Kontsevich’s function F(q) given by
F ( q ) : = n = 0 ( q ) n ,
(1)

where ( a ) n : = ( a , q ) n : = j = 0 n 1 1 a q j .

This function does not converge on any open subset of , but converges as a finite sum for q any root of unity. Zagier’s study of F depends on the ‘sum of tails’ identity
n 0 η ( q ) q 1 24 q n = η ( τ ) D q 1 2 n 1 n χ 12 ( n ) q n 2 1 24 ,
(2)
where η ( q ) : = q 1 24 ( q ) , D ( q ) : = 1 2 + n 1 q n 1 q n , and χ 12 ( n ) : = 12 n . Recalling the identity ( q ) = n 1 χ 12 ( n ) q n 2 1 24 , we find that the last term on the right-hand side of (2) is essentially a ‘half-derivative’ of η. Zagier further observed that η(q) and η(q)D(q) vanish to infinite order as q approaches a root of unity. Thus, we have what Zagier terms a ‘strange identity’ of the shape
F ( q ) = 1 2 n 1 n χ 12 ( n ) q n 2 1 24 .
(3)

Although the left- and right-hand sides of (3) do not ever converge simultaneously, as we will see below, this identity can be interpreted as saying that there is an equality between asymptotic expansions when q=et as t→0+ (a similar statement holds as q approaches other roots of unity as well).

Since [39], there has been an explosion of research aimed at constructing examples of quantum modular forms related to Eichler integrals, extending the initial applications to knot invariants and quantum invariants of three-manifolds in [28] and [40]. For instance, such quantum modular forms are closely tied to surprising identities relating generating functions of ranks, cranks, and unimodal sequences [20]; are related to probabilities of integer partition statistics [35]; and arise in the study of negative index Jacobi forms and Kac-Wakimoto characters [7]. For further examples, see also [5],[10],[11],[16],[24],[26],[31],[36].

The general theory of this class of quantum modular forms was further elucidated by Choi, Lim, and Rhoades in [18], and from a different perspective by Bringmann and the third author in [9], where the space of ‘Eichler quantum modular forms’ was defined. In particular, for each half-integral weight cusp form, there is an associated Eichler integral with quantum modular properties. At the end of [9], a program was laid out to study the general properties of these quantum modular forms. In particular, one of the fundamental problems in the theory was identified to be the determination of the arithmetic properties of such forms. This problem was inspired by recent work of Andrews and Sellers [4] in which they studied the congruence properties of the Fishburn numbers defined by
n 0 ξ ( n ) q n : = n 0 1 q ; 1 q n .
These numbers are important in combinatorics and knot theory, and in particular, ξ(n) enumerates the number of linearly independent Vassiliev invariants of degree n[40]. For related results on the conbinatorics and asymptotic analysis of the Fishburn numbers and related sequences, see also [3],[6],[8],[12],[13],[19],[27],[30],[37]. Utilizing beautiful and clever manipulations, Andrews and Sellers proved an infinite family of congruences for the Fishburn numbers, and these results were extended in several directions by Garvan [21], Straub [38], and Ahlgren and Kim [1]. For example, the work of Andrews and Sellers and Straub implies that for all A , n N , we have
ξ 5 A n 1 ξ 5 A n 2 0 mod 5 A ,
ξ 7 A n 1 0 mod 7 A ,
ξ 1 1 A n 1 ξ 1 1 A n 2 ξ 1 1 A n 3 0 mod 1 1 A .

In the context of quantum modular forms, the Fishburn numbers are the coefficients of the power series expansion at the root of unity q=1 of the Eichler quantum modular form associated to η. In light of this connection, Garvan and Straub posed the following natural question.

Question(Garvan, Straub).

What is the general theory of such congruences for more general quantum modular forms?

In this paper, we answer this question for a class of quantum modular forms. In particular, we consider quantum modular forms which arise as Eichler integrals of unary theta series. That is, we consider the quantum modular forms associated to theta series of the form
n Z χ ( n ) n ν q n 2 a 2 b ,

where ν{0,1}, χ is a periodic sequence of mean value zero, and a , b Z . For the sake of definiteness, we will focus our attention on the case when ν=0, i.e., when the unary theta series has weight 1/2. We will shortly see how to define a ‘Fishburn-type’ sequence associated to any such unary theta series via a series expansion as q→1. We remark in passing that the analogous study of congruences for general modular forms of half-integral weight does not make sense as we do not know any algebraicity results about the associated power series coefficients.

Before stating our main result, we define a condition on the periodic sequences χ under consideration. This condition arises very naturally from the perspective of p-adic measures below and is also satisfied by our canonical family of examples arising from knot invariants (see the ‘Fishburn numbers and Hikami’s functions’ section). Specifically, our condition on χ is as follows, where
ψ u ( x ) : = 1 if x u ( mod M ) , 0 otherwise ,

and ψu,v(x):=ψ u (x)−ψ v (x).

Definition.

Let χ : Z { 0 , 1 , 1 } be a periodic function with period M and mean value zero. We say that χ is a good function if it is a sum of functions of the form ψu,v such that there exist natural numbers C:=C(u,v) with (C,M)=1 and u Cv (mod M).

Remark.

The condition for χ to be good is trivially satisfied whenever χ is supported on integers coprime to M. In particular, any quadratic Dirichlet character is good.

To describe our main result, we first consider for any a , b Z and any periodic sequence χ the following partial theta function:
P a , b , χ ( q ) : = n 0 ( n ) q n 2 a 2 b .
Note that this is essentially a half-derivative of the series n χ ( n ) q n 2 a 2 b . From the general theory outlined in [9], χ has mean value zero in particular whenever Pa,b,χ is a half-derivative of a modular form which is cuspidal at q=1. From now on, we will always assume that P a , b , χ Z [ [ q ] ] . As we will see in the proof of Theorem 4, the mean value zero property of χ implies that there is an asymptotic expansion
P a , b , χ e t n 0 α a , b , χ ( n ) t n
for some α a , b , χ ( n ) C as t→0+. Thus, we may define coefficients Ha,b,χ(n) by the relation
P a , b , χ e t n 0 H a , b , χ ( n ) 1 e t n .

It is easily seen that the coefficients Ha,b,χ(n) are defined by finite recursions in terms of αa,b,χ(n). Moreover, the explicit formulas for the coefficients in Theorem 4 show that H a , b , χ ( n ) Q for all n. We will see below that in important examples arising from knot theory, they may also be defined combinatorially in a manner similar to the definition of the Fishburn numbers above.

Our main result is then the following, where we define β p =β to be the coefficient of p1 in the base p expansion of a 2 b .

Theorem 1.

Let χ be a good function and suppose that a , b Z are chosen so that P a , b , χ Z [ [ q ] ] . Then, for any prime p not dividing b, the following are true.
  1. (1)
    If B is a positive integer such that
    a 2 b p = 1 , a 2 2 b p = 1 , , a 2 Bb p = 1 ,
     
then for all A , n N , we have
H a , b , χ p A n B 0 mod p A .
  1. (2)
    If β≠(p−1) and B is a positive integer such that
    a 2 b p 1 , a 2 2 b p 1 , , a 2 Bb p 1 ,
     
then for all A , n N , we have
H a , b , χ p A n B 0 mod p A .

Three remarks. 1) We will see in the ‘Fishburn numbers and Hikami’s functions’ section that Theorem 1 implies all known linear congruences for ξ(n). 2) We will give a stronger version of Theorem 1 in Theorem 7 below. 3) As asked by Andrews and Sellers and Straub in the case of ξ(n), it is an interesting question to determine a converse of Theorem 1 classifying linear congruences for the sequences Ha,b,χ(n). This seems difficult to prove using our techniques. From the perspective of p-adic measures outlined in the ‘Preliminaries’ section, Theorem 3 aligns with the intuitive notion that the ‘integral’ of a function which is p-adically small over a compact set is small. However, the difficulty in proving a converse result lies in the fact that the integral of a large function can still be small.

In particular, we have the following corollary, as will be shown in the ‘Fishburn numbers and Hikami’s functions’ section, which implies that infinitely many of the sequences arising in our family of examples from knot theory satisfy linear congruences modulo at least 50% of primes.

Corollary 2.

Let χ be a good function and suppose that a , b Z are chosen so that P a , b , χ Z [ [ q ] ] , and that a2b3 (mod 4) and such that a2b is not a square. Then, there is a linear congruence for Ha,b,χ(n) modulo at least 50% of primes.

Remark.

We will see in the ‘Fishburn numbers and Hikami’s functions’ section that the condition that a2b is not a square is necessary, where we will give an example in which our theorem does not guarantee any congruences.

We now highlight an important situation when the coefficients Ha,b,χ(n) may be defined without using asymptotic expansions of partial theta functions. To describe this situation, we require the Habiro ring, which was brilliantly studied by Habiro in [22] and further connected to the F 1 story and to the theory of Tate motives in [34]. This ring is defined as the completion
: = lim n 0 Z [ q ] / ( ( q ) n ) ,
which may, as usual, be realized as the set of formal expansions of the form
n 0 a n ( q ) ( q ) n ,
where a n Z [ q ] . Associated to any element of the Habiro ring and to any root of unity ζ, there is a power series expansion in (ζq) [22]. That is, there is a map
ϕ ζ : Z [ ζ ] [ [ ζ q ] ] .
For example, the map ϕ1 may be realized explicitly (and be computed efficiently) for any F = n 0 a n ( q ) ( q ) n as the expansion n 0 c n 1 q n by recursively solving for c n in the expansion
n 0 a n ( 1 q ) ( 1 q ; 1 q ) n = n 0 c n q n .

The resulting expressions defining c n always terminate, since (1−q;1−q) n =O(q n ). In particular, recalling the definition of ξ(n), the Fishburn numbers are the coefficients of ϕ1(F), where F is Kontsevich’s function.

A central result of Habiro (see Theorem 5.2 of [22]) states that the map ϕ ζ is injective for any ζ, so that in fact, an element of the Habiro ring is uniquely determined by its power series expansion at any given root of unity. As Habiro points out, this is similar to the fact that a holomorphic function is determined by its power series expansion at a single point, so that may be thought of as a ‘ring of analytic functions on the roots of unity’.

Returning to the question of congruences, we now suppose that Pa,b,χ satisfies a ‘strange identity’, i.e., that there is an F a , b , χ such that
P a , b , χ e t F a , b , χ e t .
(4)
Defining ca,b,χ(n) as the coefficients of ϕ1(Fa,b,χ), it directly follows from (4) and the definition of Ha,b,χ(n) that
H a , b , χ ( n ) = c a , b , χ ( n ) .

Hence, we have given congruences for the coefficients of power series expansions of any element of the Habiro ring which satisfies a strange identity connecting it to a partial theta function Pa,b,χ satisfying the conditions of Theorem 1. In the ‘Fishburn numbers and Hikami’s functions’ section, we study the half-derivatives of Andrews-Gordon functions, which satisfy strange identities as shown by Hikami in [25]. Furthermore, the resulting elements of the Habiro ring arise naturally in the context of Kashaev’s invariant for the (2m+1,2) torus knots and include Kontsevich’s function as a special case.

More generally, it is very interesting to ask which partial theta functions satisfy strange identities, and what the general structure of such identities should be. For work related to this question, the interested reader is referred to the important work of Coogan, Lovejoy, and Ono in [15],[32], which studies connections between asymptotic expansions of partial theta series and q-hypergeometric series.

The paper is organized as follows. In the ‘Preliminaries’ section, we recall some standard facts on p-adic measures and L-functions, and we prove a fundamental result concerning congruences of polynomials in L-values, as well as a useful expression of the coefficients Ha,b,χ(n) in terms of L-values. Together, these results allow us to reduce the proof of Theorem 1 to an elementary statement on congruences for binomial coefficients, which we prove along with Corollary 2 in the ‘Proof of Theorem 1 and Corollary 2’ section. Finally, in the ‘Fishburn numbers and Hikami’s functions’ section, we give an illuminating example of importance in knot theory, which also shows that Theorem 1 yields all known congruences for the Fishburn numbers, as well as new congruences for a class of sequences naturally generalizing the Fishburn numbers.

Preliminaries

Congruences for polynomials in L-values

In this subsection, we prove a useful theorem regarding the p-adic properties of certain polynomials in L-values. For any periodic sequence χ(n) of mean value zero, we define for Re(s)0 the L-function
L χ ( s ) : = n 1 χ ( n ) n s .

This L-function has an analytic continuation to all of , which can easily be seen by writing L as a linear combination of specializations of the Hurwitz zeta function and using the well-known continuation for the Hurwitz zeta function, which only has a simple pole of residue one at s=1. Thus, in particular, the values of L(s) for s N 0 are well defined.

In order to state our theorem, we first define a linear operator L χ : Q [ x ] C by its action on generators as
L χ x n : = L χ ( n ) ,

and we recall that f Q [ x ] is called a numerical polynomial on a set X Z if f ( x ) Z for all xX. We also denote by supp(χ) the support of χ. Our main result concerning congruences of L-values is as follows.

Theorem 3.

Let χ be a good function with period M, p a prime with (p,M)=1, and A N . If f is a numerical polynomial on supp(χ) such that for all nsupp(χ), we have
f ( n ) 0 mod p A ,
then we also have
L χ ( f ) 0 mod p A .

Remark.

This theorem is a specialization of fairly standard facts on p-adic Dirichlet L-functions, or, if one prefers, p-adic Mazur measures (see, e.g., [17], [23], Chapter 3]). However, for the reader’s convenience, we present a full and elementary proof below.

Proof.

We begin by recalling the definition of generalized Bernoulli numbers Bk,χ, which are given by the generating function
a = 0 M 1 χ ( a ) t e at e Mt 1 = : k 0 B k , χ t k k ! .
As mentioned above, L χ has an analytic continuation to all of . At non-positive integers, this is realized by the well-known relation
L χ ( n ) = B n + 1 , χ n + 1 .
Furthermore, it is possible (see [29], Chapter XIII, Theorem 1.2]) to present these numbers as p-adic limits of related power sums:
B k , χ = lim n 1 M p n a = 0 M p n 1 χ ( a ) a k .
We thus have that
L χ x k = L χ ( k ) = B k + 1 , χ k + 1 = lim n 1 M p n a = 0 M p n 1 χ ( a ) a k + 1 k + 1 .
Now let c>1 be a positive integer such that (c,M p)=1, and define
L χ c ( f ) : = L χ ( f ) L χ c f c ,
where the functions f c and χ c are defined as f c (x):=f(c x) and χ c (x):=χ(c x). We now write
f ( x ) = m = 0 N d m x m
with d m Q and use the above to compute
L χ c ( f ) = m = 0 N d m lim n 1 M p n a = 0 M p n 1 χ ( a ) a m + 1 m + 1 χ ( ca ) ( ca ) m + 1 m + 1 .
Let a c be the element of {0,1,…M p n −1} such that a c c a (mod M p n ). Since (c,M p)=1, multiplication by c permutes residues modulo M p n , and we can rearrange the sum as
L χ c ( f ) = m = 0 N d m lim n 1 M p n a = 0 M p n 1 χ ( ca ) a c m + 1 ( ca ) m + 1 m + 1 .
We now let a c =c a+M p n t a with t a Z , make use of the congruence
a c m + 1 ( ca ) m + 1 = ( ca + M p n t a ) m + 1 ( ca ) m + 1 ( m + 1 ) ( ca ) m M p n t a mod M p n 2 ,
and reorder the summations to conclude that
L χ c ( f ) = lim n a = 0 M p n 1 t a χ ( ca ) f ( ca ) .
In particular, we derive that for every c>1 such that (c,M p)=1,
L χ c ( f ) 0 mod p A

whenever f(n)≡0(mod p A ) for every nsupp(χ).

However, this does not suffice since we need the congruence for L χ ( f ) itself, not merely for the modified version L χ c ( f ) . Since χ is a good function, it suffices to prove the congruence for the function ψu,v(n) where u and v satisfy the conditions of the definition of a good function. We assume that u and v do satisfy these conditions for the remainder of the proof. Now note that for an integer c such that (c,M p)=1, we have
ψ u c ( x ) = ψ u ( cx ) = ψ u c 1 ( x ) ,
where c c−1≡1 (mod M). We may now choose two integers c1>1 and c2>1 satisfying
( c 1 , Mp ) = ( c 2 , Mp ) = 1 uc 1 v ( mod M ) c 2 1 ( mod M ) c 1 c 2 ( mod p C )

for any large C, which we choose later.

We already know that
L ψ v c 1 ( f ) = L ψ v ( f ) L ψ u ( f c 1 ) 0 ( mod p A )
and
L ψ u c 2 ( f ) = L ψ u ( f ) L ψ u f c 2 0 ( mod p A ) ,
and it suffices to prove that C can be chosen big enough to guarantee the congruence
L ψ u f c 1 L ψ u f c 2 = L ψ u f c 1 f c 2 0 ( mod p A ) .
However, the assumptions on u and v imply that
L ψ u f c 1 f c 2 = m = 0 N d m c 1 m c 2 m B m + 1 , ψ u m + 1

for some rational quantities B m + 1 , ψ u / ( m + 1 ) , which makes our claim obvious.

A useful formula for the sequence H a,b,χ (n)

In this subsection, we relate the values of the sequences Ha,b,χ(n) to polynomials in L-values. Namely, we show the following result.

Theorem 4.

Assume the conditions of Theorem 1. Then, we have
H a , b , χ ( n ) = ( 1 ) n L χ x x 2 a 2 b n .

Proof.

Plugging in q=et into the definition of Pa,b,χ(q), we find
P a , b , χ e t = n 1 ( n ) e n 2 a 2 b t .
Using a shifted version of the Euler-Maclaurin summation formula or a simple Mellin transform argument (see, for example, the proposition on page 98 of [28]), we find as t→0+ the asymptotic expansion
P a , b , χ e t e a 2 b t n 0 ( 1 ) n L χ ( 2 n 1 ) n ! t b n = n 0 t n n ! b n j = 0 n ( 1 ) j L χ ( 2 j 1 ) a 2 ( n j ) n j = n 0 ( t ) n n ! b n L χ x x 2 a 2 n .
(5)
Now from the definition of the coefficients Ha,b,χ(n), we have another expression for the asymptotic expansion of Pa,b,χ(et), given by
P a , b , χ e t n 0 H a , b , χ ( n ) 1 e t n .
Letting n k denote the Stirling numbers of the second kind, which are defined by the generating series
n k n k x n n ! : = 1 k ! e x 1 k ,
we find
P a , b , χ e t n 0 ( t ) n n ! j = 0 n n j j ! H a , b , χ ( j ) ( 1 ) j .
(6)
By comparing coefficients of t n in (5) and (6), we find
j = 0 n n j j ! H a , b , χ ( j ) ( 1 ) j = 1 b n L χ x x 2 a 2 n .
(7)
We now require the Stirling numbers of the first kind, which are defined by the relation
j = 0 n s ( n , j ) x j : = ( x ) n ,

where ( x ) n : = j = 0 n 1 ( x j ) is the usual Pochhammer symbol.

Recall that the two types of Stirling numbers satisfy the inversion relationship
k = 0 n s ( n , k ) k j = δ n , j ,
(8)
where
δ x , y : = 1 if x = y , 0 if x y.
Thus, if we have two sequences u n ,v n satisfying the relationship v n = j = 0 n n j u j , then we can easily invert to find
k = 0 n s ( n , k ) v k = j = 0 n u j k = j n s ( n , k ) k j = j = 0 n u j k = 0 n s ( n , k ) k j = j = 0 n u j δ n , j = u n ,
(9)

where we used the fact that k j = 0 for j>k.

Combining (7) and (9), we obtain
H a , b , χ ( n ) = ( 1 ) n n ! L χ x j = 0 n x 2 a 2 b j s ( n , j ) = ( 1 ) n n ! L χ x x 2 a 2 b n ,

which implies the desired formula for Ha,b,χ(n).

Proof of Theorem 1 and Corollary 2

We begin with a result giving a family of congruences for the sequences Ha,b,χ(n) which we will shortly see implies Theorem 1. Firstly, however, we derive an elementary lemma on congruences of binomial coefficients. Specifically, we now recall Kummer’s theorem, which allows us to easily study such congruences.

Theorem 5(Kummer, [33]).

Let p be a prime, and suppose n Z , k N . Then, the p-adic order of n k equals the number of carries when adding k to nk in base p.

From this, one can easily obtain the following lemma (see also Lemma 3.4 of [38]).

Lemma 6.

Let p be a prime, s{0,1,…,p−1}, and α N . Then, the following are true.
  1. (1)
    If B{1,2…,ps−1}, then for all A , n N ,
    s + p A n B 0 ( mod p A ) .
     
  1. (2)
    If B{1,2…,p−1} and α−1 (mod p), then for all A , n N ,
    s + p A n B 0 ( mod p A 1 ) .
     

Proof.

Write n:=s+p α and k:=p A nB, and denote the base p coefficients of n (resp. k) by n0,n1,… (resp. k0,k1,…). By assumption, n0=s and k0=pB. Moreover, we have k1=k2=…=kA−1=p−1. We now split into cases.

Proof of (1): As B{1,2…,ps−1}, we have k0>n0. Denoting the base p coefficients of nk by m0,m1,…, we find that m0=pk0+n0, so that there is a carry when m0 is added to k0. Since k1=k2=…=kA−1=p−1, there are at least A carries occurring when k is added to nk, so by Theorem 5, we find the desired congruence.

Proof of (2): By part (1) of the Lemma, we may suppose that B{ps,ps+1,…,p−1}, i.e., that k0n0. Note that m0=n0k0, and by the assumption on α, we find n1p−1, so that m1>0. Hence, when adding k to nk, a carry occurs when adding m1 to k1. Since k2=k3=…=kA−1=p−1, there is also a carry when m i is added to k i for i=2,…A−1, so that there are at least A−1 carries when k is added to nk. By Theorem 5, the result follows.

In order to state our generalized version of Theorem 1, we first set
S a , b , χ , p : = S : = s N 0 : s < p , x supp ( χ ) , x 0 ( mod p ) , x 2 a 2 b s ( mod p ) ,
and we define an analogous set
S a , b , χ , p : = S : = s N 0 : s < p , x supp ( χ ) , x 2 a 2 b s ( mod p ) .

Then, our main result is as follows.

Theorem 7.

Let χ be a good function and p a prime, and suppose a , b Z are chosen so that P a , b , χ Z [ [ q ] ] . Then, the following are true.
  1. (1)
    If B { 1 , 2 , , p 1 max s S s } , then for all n , A N , we have
    H a , b , χ p A n B 0 ( mod p A ) .
     
  1. (2)
    If β≠(p−1) and (b,p)=1, then for any B{1,2,…,p−1− maxsS s} and for all n , A N , we have
    H a , b , χ p A n B 0 ( mod p A ) .
     

Proof.

We begin by splitting into cases. Proof of (1): By Theorem 3 and Theorem 4, it suffices to show that for B { 1 , 2 , , p 1 max s S s } , we have
x x 2 a 2 b p A n B 0 ( mod p A )
for all A , n N , xsupp(χ). For fixed x, let y : = x 2 a 2 b , and let s N be the reduction of y modulo p with 0≤s<p. We claim that we can choose α N so that
y p A n B s + p A n B ( mod p A ) .
For this, it is enough to choose α satisfying
( y ) p A n B ( s + ) p A n B ( mod p A + C ) ,
where C:= ordp((p A nB)!). Clearly, it suffices to choose α with
y s + ( mod p A + C ) .
Now we can set ys=:z p, where z Z . The last equation is then equivalent to
α z ( mod p A + C 1 ) ,
which is clearly possible to solve for α. Thus, for such an α, we have
y p A n B s + p A n B ( mod p A ) .
Now (1) of Lemma 6 directly implies that
s + p A n B 0 ( mod p A )

for all B{1,2,…,p−1− maxsSs}, n N , xsupp(χ).

Proof of (2): To complete the proof, it suffices to show that if x≡0 (mod p), β≠(p−1), and (b,p)=1, then for all 0≤Bp−1 n N , we have
x x 2 a 2 b p A n B 0 ( mod p ) .
In this case, we find that y a 2 b ( mod p 2 ) . Hence, we may choose α as above and assume without loss of generality that ys+p α (mod p2). As β≠(p−1), we have that α−1 (mod p). By (2) of Lemma 6, we find that
s + p A n B 0 ( mod p A 1 ) .
Hence,
x x 2 a 2 b p A n B x s + p A n B 0 ( mod p A ) ,

as desired.

We are now in a position to prove Theorem 1. First, consider the sets
T a , b , χ , p : = T : = s N 0 : s < p , x Z , x 0 ( mod p ) , x 2 a 2 b s ( mod p ) ,
T a , b , χ , p : = T : = s N 0 : s < p , x Z , x 2 a 2 b s ( mod p ) .

Clearly, since ST, and ST if B satisfies the conditions of Theorem 7 with S replaced by T or with S replaced by T, then the same congruence for Ha,b,χ holds. Theorem 1 follows from this observation, together with the following elementary result, whose proof follows from a straightforward calculation.

Lemma 8.

Let a , b Z , let p be a prime and suppose (p,b)=1. Then, the following are equivalent conditions for B N .
  1. (1)
    We have that B satisfies
    a 2 b p = 1 , a 2 2 b p = 1 , , a 2 Bb p = 1 .
     
  1. (2)
    We have that
    B 1 , 2 , , p 1 max t T t .
     
Moreover, the following are also equivalent conditions for B N .
  1. (1)
    We have that B satisfies
    a 2 b p 1 , a 2 2 b p 1 , , a 2 Bb p 1 .
     
  1. (2)
    We have that
    B 1 , 2 , , p 1 max t T t .
     

Remark.

An elementary argument shifting integers x by multiples of p shows that if (p,M)=1, then we have S=T and S=T. Hence, for all but finitely many primes p, the congruences given in Theorem 7 are exactly the same as the congruences implied by Theorem 1.

We now prove Corollary 2.

Proof (Proof of Corollary 2).

It suffices to check that if a2b is not a square and a2b3 (mod 4), then exactly 50% of primes p satisfy a 2 b p = 1 , since Theorem 1 then implies that Ha,b,χ(p n−1)≡0 (mod p) for all n N . To see that this is the case, note that given the conditions on a2b, the function a 2 b · is a non-principal Dirichlet character and hence takes on values 1 and −1 equally often (as the sum over a complete set of representatives of residue classes modulo the modulus of the character is zero). Moreover, the values where this character are non-zero are exactly those values which are coprime to the modulus of the character. Hence, by the Chebatorev density theorem applied to primes in arithmetic progressions, exactly 50% of primes satisfy the desired condition.

Fishburn numbers and Hikami’s functions

In this section, we work out a particularly important family of examples of Theorem 1. Specifically, we consider a collection of quantum modular forms whose beautiful properties were laid out by Hikami in [25]. For further important results ‘inverting’ these functions and relating them to indefinite theta series and mock theta functions, see also [11],[26]. The quantum modular forms defined by Hikami are then given for m N and α{1,2,…,m−1} by
F m ( α ) ( q ) : = k 1 , k 2 , , k m = 0 ( q ) k m q k 1 2 + + k m 1 2 + k α + + k m 1 i = 1 i α m 1 k i + 1 k i q · k α + 1 + 1 k α q ,
where the usual q-binomial is defined by
n k q : = ( q ) n ( q ) k ( q ) n k if 0 k n , 0 otherwise .
We note that the function F 1 ( 0 ) ( q ) reduces simply to Kontsevich’s function F(q). Moreover, it is clear from the definition that F m ( α ) . The connection to partial theta functions is shown in (15) of [25], which states that F m ( α ) shares an asymptotic expansion with a half-derivative of an Andrews-Gordon function. Specifically, Hikami shows that there is a strange identity connecting F m ( α ) and
1 2 n 1 n χ 8 m + 4 ( α ) ( n ) q n 2 2 m 2 α 1 2 8 ( 2 m + 1 ) ,
(10)
where
χ 8 m + 4 ( α ) ( n ) : = 1 if n 2 m 2 α 1 ( mod 8 m + 4 ) , 1 if n 2 m + 2 α + 3 ( mod 8 m + 4 ) , 1 if n 6 m 2 α + 1 ( mod 8 m + 4 ) , 1 if n 6 m + 2 α + 5 ( mod 8 m + 4 ) , 0 otherwise .
We now note that P 2 m 2 α 1 , 8 ( 2 m + 1 ) , χ 8 m + 4 ( α ) Z [ [ q ] ] , since, as in (10) of [25],
1 ( q ) n 0 χ 8 m + 4 ( α ) ( n ) q n 2 ( 2 m 2 α 1 ) 2 8 ( 2 m + 1 ) = n 1 n 0 , ± ( α + 1 ) ( mod 2 m + 1 ) 1 q n 1 .
Moreover, by the discussion in the ‘Introduction and statement of results’ section, in this case, the coefficients H 2 m 2 α 1 , 8 ( 2 m + 1 ) , χ 8 m + 4 ( α ) may be defined combinatorially as the coefficients of the expansion
n 0 ξ m ( α ) ( n ) q n : = F m ( α ) ( 1 q ) ,
so that
H 2 m 2 α 1 , 8 ( 2 m + 1 ) , χ 8 m + 4 ( α ) ( n ) = ξ m ( α ) ( n ) .
As an example, consider
F 2 ( 0 ) ( q ) = n 0 ( q ) n k = 0 n q k ( k + 1 ) n k q ,
so that the first few coefficients ξ 2 ( 0 ) ( n ) are given by 1,2,6,23,109,…. Numerical calculations suggest that the following congruences hold for all n , A N :
ξ 2 ( 0 ) 3 A n 1 0 ( mod 3 A ) ,
ξ 2 ( 0 ) 1 1 A n a 0 ( mod 1 1 A ) , where a { 1 , 2 , 3 } ,
ξ 2 ( 0 ) 1 3 A n a 0 ( mod 1 3 A ) , where a { 1 , 2 , 3 , 4 } .

These congruences follow immediately from Theorem 1 and Lemma 8, along with a short computation, once the following result is checked.

Lemma 9.

For any m N , a{1,2,…,m−2}, χ 8 m + 4 ( α ) ( n ) is a good function.

Proof.

Let x:=2m+1 and y:=α+1. Then, the period of χ 8 m + 4 ( α ) is M=4x, and by the definition of the sequence, it is clearly supported on odd integers. Furthermore, one easily checks that
χ 8 m + 4 ( α ) = ψ x 2 y , x + 2 y ψ 3 x 2 y , 3 x + 2 y .
Now it is easy to verify that
( 4 x , x 2 y ) = ( 4 x , x + 2 y ) = ( x , y )
and
( 4 x , 3 x 2 y ) = ( 4 x , 3 x + 2 y ) = ( x , y ) .

Hence, ψx−2y,x+2y and ψ3x−2y,3x+2y both satisfy the conditions of a good function, so that χ 8 m + 4 ( α ) is good as well.

Thus, we may apply Theorem 1 to the coefficients ξ m ( α ) , which directly implies congruences for ξ m ( α ) ( n ) . Using Corollary 2 and the fact that
2 m 2 α 1 2 8 ( 2 m + 1 ) 0 , 1 ( mod 4 ) ,

we immediately deduce the following result.

Corollary 10.

Choose α , m N with α<m such that (2m−2α−1)2−8(2m+1) is not a square. Then
ξ m ( α ) p A n 1 0 ( mod p A )

for all n , A N for at least 50% of primes p.

Remark.

The condition that (2m−2α−1)2−8(2m+1) is not a square is necessary in Corollary 10. For example, if α=1, m=7, then we have a2=121, b=120. Hence, Theorem 1 yields a congruence modulo p only when a 2 b p = 1 p 1 , which does not hold for any p.

We now take a closer look at the congruences for ξ(n), which inspired this paper. Using the well-known fact that for p≥5, p 2 1 24 Z , we find that if p 2 1 24 s ( mod p ) with 0≤s<p, then
β p = p 2 1 24 s p = p 2 1 24 p ( p 1 ) .
By Theorem 1, we find that
ξ p A n 1 0 ( mod p A )
for all A , n N whenever
23 p 1 .
Using quadratic reciprocity, it is easy to check that this occurs exactly when
p = 23 or p 5 , 7 , 10 , 11 , 14 , 15 , 17 , 19 , 20 , 21 , 22 ( mod 23 ) ,

which is remarked in [4] and [38].

Finally, we show that our set of congruences for ξ(n) is the same as that which was given in [38]. Indeed, one easily finds by an elementary argument that the set S in Theorem 7 may be replaced by the set T (since (p,M)=1). Comparing with Theorem 1.2 of [38], it suffices to show that the set of reductions of values of x 2 1 24 modulo p as x ranges over Z pℤ is equal to the set of reductions of pentagonal numbers 1 2 x ( 3 x 1 ) modulo p with x0 (mod p) for any prime p≥5. However, this follows immediately by noting that x 2 1 24 becomes 1 2 x ( 3 x 1 ) upon substituting x with 6x−1, which simply permutes the residue classes of x modulo p. Thus, thanks to the extensive calculations of Straub [38], we have conjecturally given all linear congruences for ξ(n). That is, following Andrews and Garvan and Straub, we make the following conjecture, which we leave as an important challenge for future work.

Conjecture.

Let p be a prime. Then, there exists a B N such that there is a congruence
ξ p A n B 0 ( mod p A )
for all n precisely when
p = 23 or p 5 , 7 , 10 , 11 , 14 , 15 , 17 , 19 , 20 , 21 , 22 ( mod 23 ) .
We conclude by noting that there are other congruences for linear combinations of Fishburn numbers. Specifically, in unpublished work, Garthwaite and Rhoades observed that for all n N , we have
ξ ( 5 n + 2 ) 2 ξ ( 5 n + 1 ) 0 ( mod 5 ) ,
ξ ( 11 n + 7 ) 3 ξ ( 11 n + 4 ) + 2 ξ ( 11 n + 3 ) 0 ( mod 11 ) .

In Theorem 1.3 of [21], Garvan established an infinite set of congruences which includes this example. It is likely that the methods of this paper extend to prove these congruences of Garvan and that similar congruences hold for other quantum modular forms. We leave the details to the interested reader.

Declarations

Acknowledgements

The research of the first author is supported by the Simons Foundation Collaboration Grant. This research was conducted while the first author was a guest at MPIM, and he is grateful to the Institute for making this research possible. The first author thanks Don Zagier for an elucidating communication. The third author is grateful for the support of the DFG through a University of Cologne postdoc grant.

Authors’ Affiliations

(1)
Department of Mathematics, University of Hawaii
(2)
Mathematical Institute, University of Cologne

References

  1. Ahlgren, S, Kim, B: Dissections of a “strange” function. Int. J. Number Theory (2014). in press.Google Scholar
  2. Andrews G, Dyson F, Hickerson D: Partitions and indefinite quadratic forms. Invent. Math 1988,91(3):391–407. 10.1007/BF01388778MATHMathSciNetView ArticleGoogle Scholar
  3. Andrews G, Jelínek V: Some q -series identities related to interval orders. European J. Combin 2014, 39: 178–187. 10.1016/j.ejc.2014.01.003MATHMathSciNetView ArticleGoogle Scholar
  4. Andrews, G, Sellers, J: Congruences for the Fishburn numbers. J. Number Theory (2014). in press.Google Scholar
  5. Bajpai, J, Kimport, S, Liang, J, Ma, D, Ricci, J: Bilateral series and Ramanujan’s radial limits. Proc. Amer. Math. Soc (2014). in press.Google Scholar
  6. Bousquet-Mélou M, Claesson A, Dukes M, Kitaev S: (2+2)-free posets, ascent sequences and pattern avoiding permutations. J. Comb. Theory Ser. A 2010,117(7):884–909. 10.1016/j.jcta.2009.12.007MATHView ArticleGoogle Scholar
  7. Bringmann, K, Creutzig, T, Rolen, L: Negative index Jacobi forms and quantum modular forms. Research in the Math. Sci (2014). in press.Google Scholar
  8. Bringmann, K, Li, Y, Rhoades, R: Asymptotics for the number of row Fishburn matrices. J. Comb. Ser. A (2014). in press.Google Scholar
  9. Bringmann, K, Rolen, L: Half-integral weight Eichler integrals and quantum modular forms (2014). arxiv.org/abs/1409.3781 (preprint).Google Scholar
  10. Bringmann, K, Rolen, L: Radial limits of mock theta functions (2014). arxiv.org/abs/1409.3782 (preprint).Google Scholar
  11. Bryson J, Ono K, Pitman S, Rhoades R: Unimodal sequences and quantum and mock modular forms. Proc. Nat. Acad. Sci. USA 2012,109(40):16063–16067. 10.1073/pnas.1211964109MathSciNetView ArticleGoogle Scholar
  12. Chmutov S, Duzhin S: An upper bound for the number of Vassiliev knot invariants. J. Knot Theory Ramifications 1994, 3: 141–151. 10.1142/S0218216594000101MATHMathSciNetView ArticleGoogle Scholar
  13. Claesson, A, Dukes, M, Kitaev, S: A direct encoding of Stoimenow matchings as ascent sequences (2014). (preprint)., [http://arxiv.org/abs/0910.1619]
  14. Cohen H: q -identities for Maass waveforms. Invent. Math 1988,91(3):409–422. 10.1007/BF01388779MATHMathSciNetView ArticleGoogle Scholar
  15. Coogan G, Ono K: A q-series identity and the arithmetic of Hurwitz zeta-functions. Proc. Amer. Math. Soc 2003, 131: 719–724. 10.1090/S0002-9939-02-06649-2MATHMathSciNetView ArticleGoogle Scholar
  16. Costa E, Debaene K, Guerreiro J: Asymptotic expansions, L -values, and a new quantum modular form. Ramanujan J 2014,35(1):141–148. 10.1007/s11139-013-9508-9MATHMathSciNetView ArticleGoogle Scholar
  17. Courtieu, M, Panchishkin, A: Non-Archimedean L-Functions and Arithmetical Siegel Modular Forms. Second edition. Lecture Notes in Mathematics, 1471. Springer-Verlag, Berlin (2004).Google Scholar
  18. Choi, D, Lim, S, Rhoades, R: Mock modular forms and quantum modular forms. preprint. [http://math.stanford.edu/~rhoades/FILES/mockquantum.pdf]
  19. Fishburn P: Intransitive indifference with unequal indifference intervals. J. Math. Psychol 1970, 7: 144–149. 10.1016/0022-2496(70)90062-3MATHMathSciNetView ArticleGoogle Scholar
  20. Folsom A, Ono K, Rhoades R: Mock theta functions and quantum modular forms. Forum Math. Pi 2013,1(e2):27.MathSciNetGoogle Scholar
  21. Garvan, F: Congruences and relations for the r-Fishburn numbers (2014). arxiv.org/abs/1406.5611 (preprint).Google Scholar
  22. Habiro K: Cyclotomic completions of polynomial rings. Publ. Res. Inst. Math. Sci 2004,40(4):1127–1146. 10.2977/prims/1145475444MATHMathSciNetView ArticleGoogle Scholar
  23. Hida H: Elementary Theory of L-Functions and Eisenstein Series. Cambridge University Press, Cambridge; 1993.MATHView ArticleGoogle Scholar
  24. Hikami, K: Decomposition of Witten-Reshetikhin-Turaev invariant: linking pairing and modular forms. In: Chern-Simons Gauge Theory: 20 Years After AMS/IP Studies in Advanced Mathematics 50 , pp. 131–151 (2011). AMS.Google Scholar
  25. Hikami K: q -series and L -functions related to half-derivatives of the Andrews-Gordon identity. Ramanujan J 2006, 11: 175–197. 10.1007/s11139-006-6506-1MATHMathSciNetView ArticleGoogle Scholar
  26. Hikami, K, Lovejoy, J: Torus knots and quantum modular forms. Research in the Math. Sci. (2014). in press.Google Scholar
  27. Jelínek V: Counting self-dual interval orders. J. Comb. Theory Ser. A 2012,119(3):599–614. 10.1016/j.jcta.2011.11.010MATHView ArticleGoogle Scholar
  28. Lawrence R, Zagier D: Modular forms and quantum invariants of 3-manifolds. Asian J. Math 1999, 3: 93–108.MATHMathSciNetGoogle Scholar
  29. Lang S: Introduction to Modular Forms. Springer-Verlag, Berlin; 1995.Google Scholar
  30. Levande P: Fishburn diagrams, Fishburn numbers and their refined generating functions. J. Comb. Theory Ser. A 2013,120(1):194–217. 10.1016/j.jcta.2012.07.012MATHMathSciNetView ArticleGoogle Scholar
  31. Li, Y, Ngo, H, Rhoades, R: Renormalization and quantum modular forms, part II: mock theta functions (2014). arxiv.org/abs/1311.3044 (preprint).Google Scholar
  32. Lovejoy J, Ono K: Hypergeometric generating functions for the values of Dirichlet and other L -functions. Proc. Nat. Acad. Sci. USA 2003,100(12):6904–6909. 10.1073/pnas.1131697100MATHMathSciNetView ArticleGoogle Scholar
  33. Kummer E: Über die Ergänzungssätze zu den allgemeinen Reciprocitätsgesetzen. J. reine angew. Math 1852, 44: 93–146. 10.1515/crll.1852.44.93MATHMathSciNetView ArticleGoogle Scholar
  34. Lo, C, Marcolli, M: F-zeta geometry, Tate motives, and the Habiro ring. Int. J. Num. Theory (2014). in press.Google Scholar
  35. Ngo, H, Rhoades, R: Integer partitions, probabilities, and quantum modular forms. preprint ., [http://www-personal.umich.edu/~hieu/pdfs/quantum_circle.pdf]
  36. Rolen L, Schneider R: A “strange” vector valued quantum modular form. Archiv der Mathematik 2013, 101: 43–52. 10.1007/s00013-013-0529-9MATHMathSciNetView ArticleGoogle Scholar
  37. Stoimenow A: Enumeration of chord diagrams and an upper bound for Vassiliev invariants. J. Knot Theory Ramifications 1998, 7: 93–114. 10.1142/S0218216598000073MATHMathSciNetView ArticleGoogle Scholar
  38. Straub, A: Congruences for Fishburn numbers modulo prime powers (2014). arxiv.org/abs/1407.7521 (preprint).Google Scholar
  39. Zagier, D: Quantum modular forms. In: Quanta of Maths: Conference in Honor of Alain Connes, Clay Math. Proc. 11 , pp. 659–675. AMS and Clay Math. Institute (2010).Google Scholar
  40. Zagier D: Vassiliev invariants and a strange identity related to the Dedekind eta-function. Topology 2001,40(5):945–960. 10.1016/S0040-9383(00)00005-7MATHMathSciNetView ArticleGoogle Scholar

Copyright

© Guerzhoy et al.; licensee Springer. 2014

This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited.