![MathJax Logo](/templates/jsp/_style2/_tandf/pb2/images/math-jax.gif)
Abstract
Motivated by arithmetic properties of partition numbers , our goal is to find algorithmically a Ramanujan type identity of the form
, where
is a polynomial in products of the form
with
. To this end we multiply the left side by an appropriate factor such the result is a modular function for
having only poles at infinity. It turns out that polynomials in the
do not generate the full space of such functions, so we were led to modify our goal. More concretely, we give three different ways to construct the space of modular functions for
having only poles at infinity. This in turn leads to three different representations of
not solely in terms of the
but, for example, by using as generators also other functions like the modular invariant
.
1. Introduction
This note, despite its algebraic nature, has been inspired by classical additive number theory having partition numbers as one of its primary objects. For a fixed integer
,
is defined as the number of additive decompositions of n; e.g.
: 4, 3 + 1, 2 + 2, 2 + 1 + 1, 1 + 1 + 1 + 1. Notice that the order in which the parts occur is considered irrelevant. From numerical tables, Ramanujan [Citation1] observed congruences satisfied by subsequences of
; for example,
,
, or
. For the cases 5 and 7, Ramanujan established identities as a witness for the corresponding divisibility property,
(1.1)
(1.1) and
(1.2)
(1.2) For the 11 case Ramanujan did not present any such identity. Only recently, with the help of his Ramanujan-Kolberg algorithm, Radu was able to derive such kind of a witness identity of Ramanujan type; see [Citation2, (58)]. Radu's work triggered further algorithmic developments on this theme. We mention a few examples.
First, another witness identity was derived by Hemmecke in a generalized algebraic setting [Citation3, (9)]; this identity reveals the 11 divisibility in explicit manner.
Another kind of witness identity is presented in [Citation4, Thm. 1.1]: Suppose
(1.3)
(1.3) and
(1.4)
(1.4) Then
(1.5)
(1.5) The divisibility
follows immediately from the fact that all coefficients of powers of q on the right-hand side of (Equation1.5
(1.5)
(1.5) ) are integers containing 11 as a factor. This property clearly carries over to
since
is an element of an integral domain – regardless whether the
-series/products involved are considered as formal Laurent series or as analytic functions.
Analytically, when taking
the witness identities (Equation1.1
(1.1)
(1.1) ), (Equation1.2
(1.2)
(1.2) ), and (Equation1.5
(1.5)
(1.5) ) can be explained in the world of modular functions. Let
denote the
-algebra of modular functions for
having a pole (of some order) at most at infinity; definitions of these notions are given in Section 2. Then (Equation1.1
(1.1)
(1.1) ) means that its left side is a modular function in
which turns out to be the constant function 5. Similarly, (Equation1.2
(1.2)
(1.2) ) means that its left side is a modular function in
which can be written in the form as on its right side as a linear combination of 49 times the constant modular function 1, plus 7 times a modular function in
representable as a
-product.
Despite that fact that f and t being again modular functions, in , from an algebraic point of view the structure of (Equation1.5
(1.5)
(1.5) ) is more involved – and also more interesting. Namely, it expresses the algebraic fact that
(1.6)
(1.6) This means that the
-algebra of polynomials in
and
with complex coefficients can be represented as a module freely generated by
over the ring
of polynomials in
with complex coefficients. The free generation is obvious, as in all cases we will consider, owing to the fact that the pole order of the generators
are pairwise different.
In general, for non-constant modular functions in
, our notation for such modules is:
In various applications one needs a
-module representation of the whole space
. For example, when using modular functions to prove Ramanujan's congruences for powers of 11 (i.e.
,
, etc.) one needs to work with a
-module representation of
. According to the Weierstraß gap theorem, see [Citation5, Thm. 12.2] for a version in the context of modular functions, there is a representation
(1.7)
(1.7) with
as in (Equation1.3
(1.3)
(1.3) ) and with
of the form
. Atkin [Citation6] was the first to construct such
explicitly. In [Citation7] Atkin's construction was revisited and a simpler representation of the
was found by using a trace operator; more precisely, a special instance of [Citation8, (1)]. An explicit discussion of the representation (Equation1.7
(1.7)
(1.7) ) can also be found in [Citation5].
Summarizing, despite the usefulness of the module representation (Equation1.6(1.6)
(1.6) ), it does not give the full space, for instance,
in view of the pole orders 4 and 5 of
and
, respectively. It is the main objective of this note to show how a basis of the full space
can be obtained algorithmically. To be as concrete as possible, we will do this in the form of a case study where we fix
. Despite being a special case, we feel this specialization will allow to illustrate general features and, on the other hand, will be sufficiently general to lead also to interesting non-trivial number theoretic applications.
To construct module bases for the full space such that
two concepts turn out to be fundamental: order-completeness and the notion of an integral basis; see Definition 4.1. The connection to the classical notion of integral elements is made by
Lemma 1.1
[Citation5, Lemma 4.2]
Let be a modular function for
as defined in Section 2. Let
with q-expansion
. Suppose
. Then f satisfies an algebraic relation
with polynomials
(i.e.
is integral over
) if and only if
Moreover, if
, then there exists an algebraic relation with
.
The main reason to aim at the computation of an integral basis is the following. Many of the modular functions arising in q-series identities can be modified in a straight-forward manner to turn them into members of , for example, by multiplying with eta quotients. In such cases the
-expansions at the pole ∞ often are available in a ‘natural fashion’. The knowledge of an integral basis
for
that is computed from known functions such as eta-quotients or the Klein
function then allows to algorithmically express
as a
-linear combination of the elements of
.
The content of our note is structured as follows. Section 2 recalls the most important modular function notions needed. Section 3 gives a brief summary of the main problem of this note which is solved by three different methods in Sections 6–8. To describe these solutions we need some preparations. Section 4 discusses the problem of using eta-quotients for module representations. Section 5 returns to the theme of representing the generating function and prepares the ground for the computation of integral bases. Section 6 solves the problem of computing an integral basis for
by using the modular invariant; i.e. Klein's
function. Section 7 solves the integral basis problem using the trace operator already mentioned in connection with (Equation1.7
(1.7)
(1.7) ). Finally, Section 8 explains how the Maple package algcurves can be invoked to derive the desired integral basis.
2. Notation
Let denote the complex upper half-plane. In the following N denotes a positive integer. We define the groups
acts on
by
. This action induces an action on meromorphic functions
, namely
A modular function for
is a meromorphic function
such that
for all
:
; and
if
then
admits a Laurent series expansion in powers of
, where
, of the form
(2.1)
(2.1)
Because of , we say that (Equation2.1
(2.1)
(2.1) ) is a q-expansion of f at a/c. Understanding
, this extends to defining q-expansions at ∞. Note that if
then
for some
and, thus,
(2.2)
(2.2) i.e. we can (uniquely) extend the definition of f to points on
as
Let
be the set of modular functions for
that only have a pole (if any) at infinity. An element
has a representation as a Laurent series in q.
We denote by the pole order (at infinity) of f; here
is defined as the index of the least non-zero coefficient in the expansion (Equation2.1
(2.1)
(2.1) ) of f in powers of q. In view of (Equation2.2
(2.2)
(2.2) ) with c = 0 and thus
, we note that q-expansions at infinity are unique in integer powers of q.
Denote by the elements of
whose q-series expansion have rational coefficients. From Theorem 3.52 of [Citation9] it follows that
is generated as a
-vectorspace by elements of
.
The action of on
extends in an obvious way to an action on
. The orbits of the action of the subgroup
are denoted by
The set of all such orbits is denoted by
There are only finitely many cosets with respect to
; more precisely, for
,
Owing to this fact together with the observation
, there are only finitely many orbits
with
. These orbits are called cusps of
.
As usual,
denotes the Dedekind eta function.
Let be the positive divisors of N. For convenience, we allow to index n-dimensional vectors by the divisors of N, instead of the usual index set
.
We define to be the set of integer tuples
. With
we denote the subset of all tuples
of
that fulfill the following conditions:
(2.3)
(2.3)
(2.4)
(2.4)
(2.5)
(2.5)
(2.6)
(2.6) Note that
is an additive monoid.
To elements we associate eta-quotients
By [Citation10, Theorem 1], the elements of
are modular functions for
. Moreover, we define
If L is a ring and S is a subset of an L-module, we denote by
the set of L-linear combinations of elements of S. If L is a field, then
is a vector space. If
, then
is an ideal of L.
3. The problem
As pointed out in the introduction, this case study was inspired by recent algorithmic progress made in connection with classical number theoretic observations made by Ramanujan. For N = 121, the problem is to find an integral basis of the space of modular functions having a pole (if any) only at infinity.
We present a solution to this problem by following three different approaches, namely, by using Klein's j-invariant, by using series that are obtained by the trace operator, applied to some eta-quotients living in , and by employing the integral_basis command from Maple.
Essentially, in each of the three approaches we construct a basis for . These bases cannot be shown explicitly in this paper because of size. For this reason we have put the explicit expressions of the bases at https://risc.jku.at/people/hemmecke/papers/integralbasis/. Each basis at the above URL has been computed in the computer algebra system FriCASFootnote1 by the package QEta.Footnote2
In this article, we explicitly mark the references to these bases by giving the name of the file that contains the respective expression(s). This filename has to be appended to the above URL in order to retrieve the data from the internet.
4. A basis for the eta-quotients in ![](//:0)
![](//:0)
In [Citation2], Radu shows that is a finitely generated (multiplicative) monoid; i.e. there exist
such that any element of
can be written as
. Radu also describes an algorithm to compute such monoid generators. Furthermore, Radu gives an algorithm to compute elements
such that
,
, and
is order-complete. This latter notion is defined in
Definition 4.1
[Citation5]
An n element subset is called order-complete if
and for each
there exists
such that
. Let
and let
be a
-module. Then B is called an order-complete basis for M, if
and
.Such an order-complete basis B is called integral basis for
if
; i.e.
.
One can use the reduction process from [Citation3] (see, in particular, Theorem 3.2) to algorithmically determine polynomials T an indeterminate, and a remainder series
such that for a given element
,
(4.1)
(4.1) with r = 0 or
and
. Then r = 0 if and only if
.
For this article, we use an implementation of the algorithm samba from [Citation3] in our QEta package. Given modular functions with
, the algorithm samba computes a basis
such that
and
. In this paper the input
is always such that
is order-complete.
For N = 121, Radu's algorithm [Citation2] delivers two monoid generators in , namely
(4.2)
(4.2) and
(4.3)
(4.3) Next, by application of the algorithm samba to t and u one obtains the element
(4.4)
(4.4) such that
(4.5)
(4.5) i.e.
forms an order-complete basis of
.
We remark that in this simple example the reduction expressed by the left equality in (Equation4.4(4.4)
(4.4) ) can be ‘seen’ immediately. Also note that, by using
, one can easily show that the q-series of z has integer coefficients.
Definition 4.2
[Citation5, Definition 12.1]
Let M be a subalgebra of . A positive integer n is called a gap in M, if there is no
with
. We also define the gap number
as the total number of gaps in M; i.e.
.
The gap number for
can be determined by an application of relation (Equation4.5
(4.5)
(4.5) ) as
The Riemann surface
is a curve of genus 6, see https://oeis.org/A001617. The genus can be computed in FriCAS with our QEta package by calling
in the computer algebra system Sage.Footnote3
By the Weierstraß gap theorem (see [Citation11, Thm. III.5.3], respectively [Citation5, Thm. 12.2] for the given context), one has Consequently,
is a proper submodule; i.e.
is not generated as a
-vectorspace by using only eta-quotients from
.
5. The generating function for ![](//:0)
![](//:0)
In this section, we use a method described in the proof of Proposition 4.3 in [Citation5] to find a new relation for the generating function of that shows
for all
. We aim at computing the cofactor d and the coefficients
as described in the following Lemma.
Lemma 5.1
[Citation5, Proposition 4.3]
Let , with
and
. Then
for some polynomials
,
.
Using the method described in [Citation2, Section 4.1] and implemented in the package QEta,Footnote4 one can find (and prove!) that
(5.1)
(5.1) where
, is a modular function for
with a pole (of order 54) only at infinity.
However, by reducing f by the basis B above we obtain
and, therefore,
.
Since the maximal pole order of an element of the basis from (Equation4.5(4.5)
(4.5) ) is
, it is possible by (Equation4.1
(4.1)
(4.1) ) to (algorithmically) reduce any element
to an element r of pole order
.
Let us consider the 192 coefficients of the principal part of the (reduced) elements for , for
and put them into a matrix (one row for any element), i.e. the
th entry of the matrix is
(i.e. the coefficient of
in the q-series expansion of
). Since there are more rows than columns, it is clear that there must be a
-linear relation among the rows of this matrix. We can thus find a polynomial
with the property that
can be reduced to a modular function with vanishing principal part, i.e.
.
It turns out that the polynomial d that we have computed can be factored as where
Keeping track of the reduction steps of
with respect to the basis B from (Equation4.5
(4.5)
(4.5) ) then leads to a polynomial of
such that
, in other words, we have found another identity for the generating function for
in term of eta-quotients. The polynomial c can be factored to reveal a factor of 11 and the degrees of the
are 75, 66, 56, 46, and 36, respectively, see dc.input.
As mentioned above the q-series expansion of z has integer coefficients. Thus the identity reveals and proves divisibility by 11 of for all
.
With (Equation4.5(4.5)
(4.5) ) we have found an order-complete basis of
with
. The computations described above not only gave us an identity for f, but they also showed that
.
Adding f to the generators, we can determine an order-complete basis of
, i.e.
, with respective pole orders 0, 66, 42, 33, 24; see bf.input. Thus,
has gap number 31 and
is not an integral basis for
.
Remark
Inspired by the Ramanujan congruence one could, for instance, add to the generators the element
But this does not lead to a better basis, because
which is seen by the relation displayed at f2.input.
6. An integral basis by using the Klein j function
Let us come back to the basis computed in Section 5, see bf.input. This basis is not an integral basis, so we must consider to include other elements of
.
Klein's j-invariant (also called modular invariant or absolute invariant) is a modular function for ,
In the theory of modular functions the j-invariant is fundamental because every modular function can be expressed as a rational function in j. For a definition and further properties of j see, for example, Chapter VII of [Citation12].
We attempt to add more modular functions for and call samba in the hope to get an integral basis. We know, for example, that
is a modular function for
and consequently also for
. Since for
we have
, also
is a modular function for
.
We multiply and
by suitable eta-quotients with the goal to arrive at elements in
. Let us define
where t and u are the eta-quotients defined in (Equation4.2
(4.2)
(4.2) ) and (Equation4.3
(4.3)
(4.3) ), then
.
Calling samba from our QEta package with input t, u, ,
leads to an order-complete basis
, with
where
i.e.
has gap number 6. In other words,
is an integral basis for
.
We were also able to compute the representation of the elements of in terms of the original functions t, u,
and
, but these polynomials are too big to be presented in this article; see bj.input on our website.
Since is an integral basis for
, we can construct a polynomial
such that
by reducing f with respect to
and keeping track of the cofactors of this reduction, see fj.input. Unfortunately, this relation cannot be used to demonstrate
.
7. An integral basis obtained with the trace map
We can generate a new modular function by applying the trace operator to a modular function from .
The trace is given through the Atkin-Lehner involution, see, for example, Equation (1) in [Citation8].
In our case we have
where
is the matrix corresponding to the Atkin-Lehner involution, and
is the operator on functions
so that
For concrete computations with such trace maps the reader is referred to [Citation7]. Here we only remark that if
,
, is the q-series expansion of φ, then the effect of the action of
is
Also, notice that the action of
is defined via the slightly more general action of the general linear group
: for all
,
Similar to the computation of (Equation4.2
(4.2)
(4.2) ) and (Equation4.3
(4.3)
(4.3) ), Radu's algorithm [Citation2] yields 94 eta-quotients
as a complete set of monoid generators of
; i.e.
.
There are such that
By exploiting the modular transformation properties it is straight-forward to compute the q-expansions
We first apply samba to the input t, u, and g, where t and u are defined in (Equation4.2
(4.2)
(4.2) ) and (Equation4.3
(4.3)
(4.3) ). That leads to an order-complete basis
with gap number 8 for the module
; i.e. we still did not succeed to obtain an integral basis for
; see bg.input.
However, since the reduction of f with respect to that basis yields non-zero, in fact, a series of pole order 8, we call samba with input t, u, f, and g and arrive at an integral basis for
; see bfg.input
Alternatively, we also arrive at an integral basis for
, if we call samba with input t, u, and h; see bh.input. Similar to the computation of the polynomial d in Section 5, we can compute a polynomial
with
. In addition to the factors of d in Section 5,
has one more factor,
; i.e. we have
where
See ds.input on our website.
That the additional polynomial appears is, in fact, not a surprise, but can be explained by the following Theorem 7.1. It is a factor of the discriminant (wrt. y) of a polynomial
such that
, see dz.input.
Theorem 7.1 is an extension of Lemma 5.1 in the sense that we do not just claim the existence of a polynomial , but rather state that choosing the discriminant (see exact formulation in the theorem) will work for any
.
Theorem 7.1
Let , with
and
. Let
be the minimal polynomial between t and z, that is
. Let
be the usual discriminant of
with respect to y. Then
for some polynomials
,
.
Proof.
Note that the existence of such a monic polynomial is given by Lemma 1.1. Definition 7.1 of [Citation5] defines the discriminant polynomial
for an order-complete basis
of the
-module
. In that definition we see that the discriminant is the square of the determinant of a certain Vandermonde matrix:
. By comparing the definition of the entries
in [Citation5] with the definition of
in the proof of Theorem 7.1 of [Citation7], we observe that they coincide. However, in that paper G corresponds to
and F to
where
and
denote the functions
corresponding to t and z, see Remark 5.1 in [Citation5]. Furthermore, in the proof of Theorem 7.1 of [Citation7] it is also shown that the symmetric functions are exactly the coefficients of the polynomial
, i.e.
,
. By considering modular functions instead of functions
, we can remove the star and, therefore,
. We see that by definition
which coincides with
and also with
.
As described in Section 4 of [Citation5], we can by successive pole-order-reduction steps transform the initial order-complete basis into an integral basis
for
, see in particular [Citation5, Proposition 4.6].
In the th step of this process one replaces the order-complete basis
with another order-complete basis
where
for some
, i.e.
.
The discriminant polynomials is a special case of the concept of an order-reduction polynomial
, see [Citation5, Definition 5.3]. By [Citation5, Proposition 8.1] we have the following relation
for some
, i.e.
divides the discriminant
. Collecting all the
factors accordingly, we can express the elements
of the final integral basis as follows:
(7.1)
(7.1) for some
with
. By construction,
divides the discriminant
.
Since and
is an integral basis for
, we can write
for some polynomials
. Replacing the
by their respective representation given by (Equation7.1
(7.1)
(7.1) ) and multiplying by the discriminant
, we see that
as claimed by the theorem.
8. An integral basis by using Maple's algcurves
In this section we explain a third method to obtain an integral basis for ; namely, by using the Maple package algcurves. By using (Equation4.5
(4.5)
(4.5) ) and Lemma 1.1, together with Prop. 3.5 and Prop. 4.3 in [Citation5] one can prove in a fairly straightforward fashion that
where
. Note that
, so Prop. 3.5 applies. However
. Hence
Next note that
is isomorphic to
, were
is the minimal polynomial such that
. This polynomial can be found at p.maple. To compute the set of all integral elements
we can use Mark van Hoeij's Maple package algcurves via
This returns the basis where v is a rational function in T and U that is rather huge; see v.input. The q-series expansion of
has pole order 1670.
By specification of Maple's integral_basis function (see [Citation13]) we know that all that are integral over
, i.e. all elements
, can be expressed in the form
where
,
. However, the basis
is not order-complete. In order to make this basis order-complete, we apply the samba algorithm to t, u,
,
,
. This computation takes about 30 minutes and yields a basis
. Note that if also the non-leading terms of the q-expansion of the basis elements are reduced by the other basis elements, then
agrees with the basis
computed in Section 6.
Since , we have
Unfortunately, we cannot give the explicit form of the
, because they are too big and therefore only listed on our website in the file bv.input. However, we give their degree for the reader to get an idea. We define the matrix
, where
and
9. Conclusion
There exist already several identities for expressing in terms of eta-quotients. In [Citation3] we gave a relation in terms of eta-quotients for
. Initially, our goal was to get rid of the factor 2, which in this context seems unnatural, and try to work with eta-quotients for
. As shown in Section 5, f cannot be expressed in that way. However, it is possible to find a polynomial d such that
indeed is a sum of eta-quotients from
.
The wish to avoid such a polynomial prefactor d in turn led us to attempts to express f by other functions, like the Klein j invariant, an eta-quotient from , and a modular function
coming from the output of van Hoeij's Maple package. Any of these additions not only gave us a way to express f, but, more generally, a way way to compute an integral basis B such that any function of
can be expressed as a
-linear combination of elements of B. Moreover, in contrast to the basis returned by van Hoeij's Maple package, our basis can be used to algorithmically find polynomials
with
.
Acknowledgments
Paule and Radu were supported by grant SFB F50-06 of the Austrian Science Fund (FWF). We thank the anonymous referees for their comments and careful reading of our manuscript.
Disclosure statement
No potential conflict of interest was reported by the authors.
Additional information
Funding
Notes
1 FriCAS 1.3.2 [Citation14].
2 QEta 2.1 [Citation15].
3 Sage 8.0 [Citation16].
4 The Mathematica package RaduRK by Nicolas A. Smoot (see [Citation17]) also implements this method.
References
- Ramanujan S. Some properties of p(n), the number of partitions of n. In: Proceedings of the Cambridge Philosophical Society; Vol. 19; 1919. p. 207–210.
- Radu CS. An algorithmic approach to Ramanujan–Kolberg identities. J Symb Comput. 2015;68(Part 1):225–253. doi: 10.1016/j.jsc.2014.09.018
- Hemmecke R. Dancing samba with Ramanujan partition congruences. J Symb Comput. 2018;84:14–24. doi: 10.1016/j.jsc.2017.02.001
- Paule P, Radu CS. A new witness identity for 11|p(11n+6). In: Andrews GE, Garvan F, editors. Analytic number theory, modular forms and q-hypergeometric series. Cham: Springer; 2016. p. 625–640. Available from: https://link.springer.com/chapter/10.1007/978-3-319-68376-8_34
- Paule P, Radu CS. A proof of the Weierstraß gap theorem not using the Riemann-Roch formula. Ann Comb. 2019;23(3):963–1007. doi: 10.1007/s00026-019-00459-2
- Atkin AOL. Proof of a conjecture of Ramanujan. Glasgow Math J. 1967;8:14–32. doi: 10.1017/S0017089500000045
- Paule P, Radu CS. A unified algorithmic framework for Ramanujan's congruences modulo powers of 5, 7, and 11. Johannes Kepler Universität, 4040 Linz, Austria, Europe: Research Institute for Symbolic Computation; 2019. (RISC Report Series 19-09). Available from: https://www.risc.jku.at/publications
- Kohnen W. A short remark on Weierstrass points at infinity on X0(N). Monatsh Math. 2004;143:163–167. doi: 10.1007/s00605-003-0054-1
- Shimura G. Introduction to the arithmetic theory of automorphic functions. Princeton University Press; 1994. (Kanô Memorial Lectures).
- Newman M. Construction and application of a class of modular functions (II). Proc Lond Math Soc. 1959;s3–9(3):373–387. doi: 10.1112/plms/s3-9.3.373
- Farkas HM, Kra I. Riemann surfaces. 2nd ed. New York: Springer-Verlag; 1991.
- Serre JP. A course in arithmetic. New York: Springer-Verlag; 1973. Graduate Texts in Mathematics.
- van Hoeij M. An algorithm for computing an integral basis in an algebraic function field. J Symb Comput. 1994;18(4):353–363. doi: 10.1006/jsco.1994.1051
- FriCAS team. FriCAS – an advanced computer algebra system; 2020. Available from: https://fricas.github.io/
- Hemmecke R. QEta – a FriCAS package to compute with Dedekind eta functions; 2019. [cited 2020 Mar 31]. Available from: https://risc.jku.at/sw/qeta and https://hemmecke.github.io/qeta
- The Sage Developers. Sagemath, the Sage Mathematics Software System (Version 8.0); 2017. Available from: http://www.sagemath.org
- Smoot NA. An implementation of Radu's Ramanujan–Kolberg algorithm. Johannes Kepler Universität, 4040 Linz, Austria, Europe; 2019. (RISC Report Series 19-08). Available from: http://www.risc.jku.at/publications