fea-ahlgren.

qxp 8/21/01 2:42 PM Page 978

Addition and Counting:

The Arithmetic of
Partitions
Scott Ahlgren and Ken Ono

At first glance the stuff of partitions seems like computed the values of p(n) for all n up to 200.
child’s play: He found that
4 = 3 + 1 = 2 + 2 = 2 + 1 + 1 = 1 + 1 + 1 + 1. p(200) = 3, 972, 999, 029, 388,

Therefore, there are 5 partitions of the number 4 . and he did not count the partitions one-by-one:
But (as happens in number theory) the seemingly 200 = 199 + 1 = 198 + 2
simple business of counting the ways to break a
= 198 + 1 + 1 = 197 + 3 = . . . . . . . . . . . . . . . .
number into parts leads quickly to some difficult
and beautiful problems. Partitions play important Instead, MacMahon employed classical formal
roles in such diverse areas of mathematics as power series identities due to Euler.
combinatorics, Lie theory, representation theory, To develop Euler’s recurrence, we begin with the
elementary fact that if |x| < 1 , then
mathematical physics, and the theory of special
functions, but we shall concentrate here on 1
= 1 + x + x2 + x3 + x4 + . . . .
their role in number theory (for which [A] is the 1−x
standard reference).
Using this, Euler noticed that when we expand the
infinite product
In the Beginning, There Was Euler… ∞
 1
A partition of the natural number n is any nonin- = (1 + x + x2 + x3 + . . . )
creasing sequence of natural numbers whose sum n=1
1 − xn

is n (by convention, we agree that p(0) = 1 ). The × (1 + x2 + x4 + . . . )

number of partitions of n is denoted by p(n). Eighty
× (1 + x3 + x6 + . . . ) . . . ,
years ago Percy Alexander MacMahon, a major in
the British Royal Artillery and a master calculator, the coefficient of xn is equal to p(n) (think of the
first factor as counting the number of 1’s in a par-
Scott Ahlgren is assistant professor of mathematics at tition, the second as counting the number of 2’s,
the University of Illinois, Urbana-Champaign. His e-mail and so on). In other words, we have the generat-
address is [email protected]. ing function
Ken Ono is professor of mathematics at the University of ∞
 ∞
 1
Wisconsin-Madison. His e-mail address is ono@math. p(n)xn =
wisc.edu. n=0 n=1
1 − xn
Both authors thank the National Science Foundation for = 1 + x + 2x2 + 3x3 + 5x4 + . . . .
its support. The second author thanks the Alfred P. Sloan
Foundation, the David and Lucile Packard Foundation, and Moreover, Euler observed that the reciprocal of
the Number Theory Foundation for their support. this infinite product satisfies a beautiful identity

978 NOTICES OF THE AMS VOLUME 48, NUMBER 9

fea-ahlgren.qxp 8/21/01 2:42 PM Page 979

(also known as Euler’s Pentagonal Number Theo- Rademacher’s series converges is remarkable; for
rem): example, the first eight terms give the approxi-
∞ ∞
 mation
2
(1 − xn ) = (−1)k x(3k +k)/2 p(200) ≈ 3, 972, 999, 029, 388.004
n=1 k=−∞
= 1 − x − x2 + x5 + x7 − x12 − . . . . (compare with the exact value computed by
MacMahon).
These two identities show that To implement the circle method requires a
  detailed study of the analytic behavior of the
∞

 p(n)x 
n generating function for p(n) . Recall that we have
n=0 ∞
  1
× 1 − x − x2 + x5 + x7 − x12 − . . . = 1, F(x) := p(n)xn = .
n=0
(1 − x)(1 − x2 )(1 − x3 ) . . .
which in turn implies, for positive integers n, that
This is an analytic function on the domain |x| < 1 .
p(n) = p(n − 1) + p(n − 2) A natural starting point is Cauchy’s Theorem,
− p(n − 5) − p(n − 7) + p(n − 12) + . . . . which gives
1 F(x)
This recurrence enabled MacMahon to perform his p(n) = dx,
massive calculation. 2π i C xn+1

where C is any simple closed counterclockwise

Hardy-Ramanujan-Rademacher
contour around the origin. One would hope to
Asymptotic Formula for p(n)
adjust the contour in relation to the singularities
It is natural to ask about the size of p(n) . The an- of F(x) in order to obtain as much information as
swer to this question is given by a remarkable as-
possible about the integral. But consider for a mo-
ymptotic formula, discovered by G. H. Hardy and
ment these singularities; they occur at every root
Ramanujan in 1917 and perfected by Hans
of unity, forming an impenetrable barrier on the
Rademacher two decades later. This formula is so
unit circle. In our favor, however, it can be shown
accurate that it can actually be used to compute
that the size of F(x) near a primitive q -th root of
individual values of p(n) ; Hardy called it “one of
unity diminishes rapidly as q increases; moreover
the rare formulae which are both asymptotic and
the behavior of F(x) near each root of unity can be
exact.” It stands out further in importance since
described with precision. Indeed, with an appro-
it marks the birth of the circle method, which has
priate choice of C , the contribution to the integral
grown into one of the most powerful tools in an-
from all of the primitive q -th roots of unity can be
alytic number theory.
calculated quite precisely. The main contribution
Here we introduce Rademacher’s result. He de-
is the function Tq (n) ; a detailed analysis of the
fined explicit functions Tq (n) such that for all n we
errors involved yields the complete formula.
have ∞
 The circle method has been of extraordinary im-
p(n) = Tq (n). portance over the last eighty years. It has played
q=1 a fundamental role in additive number theory (in
Waring type problems, for instance), analysis, and
The functions Tq (n) are too complicated to write
even the computation of black hole entropies.
down here, but we mention that T1 (n) alone yields
the asymptotic formula Ramanujan’s Congruences
1 √
√ eπ 2n/3 . After a moment’s reflection on the combinatorial
p(n) ∼
4n 3 definition of the partition function, we have no
particular reason to believe that it possesses any
(In their original work, Hardy and Ramanujan used interesting arithmetic properties (the analytic
slightly different functions in place of the Tq (n) . formula of the last section certainly does nothing
As a result, their analogue of the series ∞
q=1 Tq (n) to change this opinion). There is nothing, for
was divergent, although still useful.) Moreover, example, which would lead us to think that
Rademacher computed precisely the error incurred p(n) should exhibit a preference to be even
by truncating this series after Q terms. In partic- rather than odd. A natural suspicion, therefore,
ular, there exist explicit constants A and B such might be that the values of p(n) are distributed
that √
An evenly modulo 2 . A quick computation of the
B first 10,000 values confirms this suspicion: of
p(n) − Tq (n) < 1/4 .
q=1
n these 10,000 values, exactly 4,996 are even
and 5,004 are odd. This pattern continues with
Since p(n) is an integer, this determines the exact 2 replaced by 3 : of the first 10,000 values, 3,313;
value of p(n) for large n . The rate at which 3,325; and 3,362 (in each case almost exactly

OCTOBER 2001 NOTICES OF THE AMS 979

fea-ahlgren.qxp 8/21/01 2:42 PM Page 980

Euler
one-third) are congruent respectively to 0 , 1 , and modulo 5 and 7 are quite ingenious but are not
2 modulo 3 . When we replace 3 by 5 , however, terribly difficult, while the proof of the congruence
something quite different happens: we discover modulo 11 is much harder). In these same
that 3,611 (many more than the expected papers he sketched proofs of extensions of
one-fifth) of the first 10,000 values of p(n) are these congruences. For example, we have
divisible by 5 . What is the explanation for this p(25n + 24) ≡ 0 (mod 25),
aberration?
The answer must have been clear to Ramanu- p(49n + 47) ≡ 0 (mod 49).
jan when he saw MacMahon’s table of values of Ramanujan noticed the beginnings of other
p(n) . So Ramanujan would have seen something patterns in these first 200 values:
like the following.
p(116) ≡ 0 (mod 121), p(99) ≡ 0 (mod 125).
1 1 2 3 5 From such scant evidence he made the following
7 11 15 22 30
conjecture:
If δ = 5a 7b 11c and 24λ ≡ 1 (mod δ),
42 56 77 101 135
then p(δn + λ) ≡ 0 (mod δ).
176 231 297 385 490
When δ = 125 , for example, we have λ = 99 . So
627 792 1002 1255 1575 Ramanujan’s conjecture is that
1958 2436 3010 3718 4565 p(125n + 99) ≡ 0 (mod 125).

What is striking, of course, is that every entry in We note that the general conjecture follows easily
the last column is a multiple of 5 . This phenome- from the cases when the moduli are powers of 5 ,
non, which persists, explains the apparent aber- 7 , or 11.
ration above and was the first of Ramanujan’s It is remarkable that Ramanujan was able to
ground-breaking discoveries on the arithmetic of formulate a general conjecture based on such
p(n) . Here is his own account. little evidence and therefore unsurprising that
the conjecture was not quite correct (in the 1930s
I have proved a number of arithmetic Chowla and Gupta discovered the counterexample
properties of p(n) …in particular that p(243) ≡ 0 (mod 73 ) ). Much to Ramanujan’s
p(5n + 4) ≡ 0 (mod 5), credit, however, a slightly modified version of
his conjecture is indeed true; in particular, we
p(7n + 5) ≡ 0 (mod 7).
now know the following:
…I have since found another method If δ = 5a 7b 11c and 24λ ≡ 1 (mod δ),
which enables me to prove all of these b
properties and a variety of others, of then p(δn + λ) ≡ 0 (mod 5a 7 2
+1
11c ).
which the most striking is The task of assigning credit for the proofs of
these conjectures when the modulus is a power of
p(11n + 6) ≡ 0 (mod 11).
5 or 7 poses an interesting historical challenge.
Typically, the proofs have been attributed to
There are corresponding properties in
G. N. Watson. Recently, however, the nature of
which the moduli are powers of 5, 7 , or
Ramanujan’s own contributions [R] has been
11 …. It appears that there are no
greatly clarified. Indeed, a complete outline of the
equally simple properties for any
proof modulo powers of 5 and a much rougher
moduli involving primes other than
sketch for powers of 7 (so rough that it did not yet
these three.
reveal his error in the statement of the conjec-
Ramanujan proved these congruences in a ture) are given by Ramanujan in a long manuscript
series of papers (the proofs of the congruences which he wrote in the three years preceding his

980 NOTICES OF THE AMS VOLUME 48, NUMBER 9

fea-ahlgren.qxp 8/21/01 2:42 PM Page 981

Rademacher
Ramanujan
Hardy

death. In typical fashion, Ramanujan provides in 1

R(5n + 4, m, 5) = · p(5n + 4) for 0 ≤ m ≤ 4,
neither case complete details for all of his asser- 5
tions. This manuscript was apparently in Watson’s 1
R(7n + 5, m, 7) = · p(7n + 5) for 0 ≤ m ≤ 6.
possession from 1928 until his death in 1965. 7
Indeed, a copy of the manuscript in Watson’s The truth of these conjectures would provide a
handwriting (the whereabouts of the original is simple and elegant combinatorial explanation for
unknown) resides in the library of Oxford’s Math- Ramanujan’s congruences. Dyson’s speculation
ematical Institute. In any event, it seems clear was confirmed ten years later by Atkin and
that Ramanujan deserves more credit than he H. P. F. Swinnerton-Dyer in a wonderful paper
has historically been granted for these cases. By which combines classical combinatorial argu-
contrast, the case of powers of 11 is much more ments with techniques from the theory of mod-
difficult; the first published proof of Ramanujan’s ular functions.
conjectures in this case was given by A. O. L. Atkin Unfortunately, Dyson’s rank does not seem
in 1967. to enjoy such simple properties for primes other
than 5 and 7 . However, he conjectured the
Dyson’s Rank and Crank existence of another natural statistic, the crank,
The celebrated physicist Freeman Dyson, when he which explains the congruence
was a college student in 1944, initiated an impor-
tant subject in partition theory by discovering a p(11n + 6) ≡ 0 (mod 11).
delightfully simple phenomenon which appeared
In the late 1980s George E. Andrews and Frank
to explain why Garvan found such a crank [A-G], [G]. Further
work of Garvan, Dongsu Kim, and Dennis Stan-
p(5n + 4) ≡ 0 (mod 5) ton [G-K-S] has produced, for the congruences
with moduli 5, 7, 11, and 25, combinatorial in-
and
terpretations which are rooted in the modular
p(7n + 5) ≡ 0 (mod 7). representation theory of the symmetric group.
Dyson defined the rank of a partition to be the Atkin’s Examples
largest summand minus the number of summands. We return to Ramanujan’s intuition that there
Here, for example, are the partitions of 4 and their are no simple arithmetic properties for p(n)
ranks: when the modulus involves primes greater than
Partition Rank 11. Ramanujan seems to have been correct in this
4 4 − 1 ≡ 3 (mod 5), claim; no new congruence as simple as the
3+1 3 − 2 ≡ 1 (mod 5), originals has ever been found (although it has
2+2 2 − 2 ≡ 0 (mod 5), not been proved that none exists). The 1960s,
2+1+1 2 − 3 ≡ 4 (mod 5), however, witnessed tantalizing discoveries of
1+1+1+1 1 − 4 ≡ 2 (mod 5). further examples (notably by Atkin, Newman,
and O’Brien). Atkin, for example, found elegant
infinite families of congruences modulo 5 , 7 , and
Notice that the ranks of these partitions repre-
13 which are quite different from those previ-
sent each residue class modulo 5 exactly once.
ously known. A simple example of these is the
After computing many more examples, Dyson
congruence
observed that, without exception, numbers of the
form 5n + 4 (respectively 7n + 5 ) have the property p(113 · 13n + 237) ≡ 0 (mod 13).
that their ranks modulo 5 (respectively modulo 7) Atkin also gave more examples, though not so
are equally distributed. More precisely, if systematic, with moduli 17, 19, 23, 29, and 31.
0 ≤ m < M are integers and R(N, m, M) denotes Atkin obtains these results via a detailed
the number of partitions of N with rank congru- study of modular functions. Since these lie at the
ent to m (mod M) , then Dyson conjectured that heart of the proofs of the congruences we have

OCTOBER 2001 NOTICES OF THE AMS 981

fea-ahlgren.qxp 8/21/01 2:42 PM Page 982

seen so far, we will give a brief description here. large role in number theory; their importance, of
Let SL2 (Z) be the set of 2 × 2 integer matrices with course, has been underscored by their central po-
determinant equal to 1 . Then, if N is an integer, sition in the proof of Fermat’s Last Theorem. The
define the congruence subgroup Γ0 (N) by crux of Wiles’ proof is to show that elliptic curves
   are “modular”; in other words, their arithmetic is
a b dictated in part by certain modular forms to which
Γ0 (N) := ∈ SL2 (Z) : c ≡ 0 (mod N) .
c d they are related. What has been learned recently
is that the partition function does not escape the

ab web of modularity; its arithmetic, too, is intimately
An element γ = c d acts on the upper half-plane
connected to the behavior of a certain family of
H of complex numbers via the linear fractional
az+b
modular forms. This connection has allowed the
transformation γz = cz+d . By definition, a modular application of deep methods of Deligne, Serre, and
function on Γ0 (N) is a function f on H which Shimura to the study of p(n). These theories (some
satisfies f (γz) = f (z) for all γ ∈ Γ0 (N) and which of the most powerful of the last half-century) have
in addition is meromorphic on H and “at the important ramifications for p(n) ; in particular,
cusps”. When N is small, the field of these functions properly applied, they imply that p(n) satisfies lin-
is relatively simple; therefore, given several func- ear congruences for every prime $ ≥ 5 . We shall
tions in such a field, one expects to find nontrivial discuss in more detail how modular forms enter
relations among them. If the right functions are in- the picture in the next section; let us first indicate
volved, then such a relation may give information what they enable us to prove.
about values of p(n) . For Atkin’s examples when The second author (inspired by some formulae
$ = 5 , 7 , or 13, the relevant function fields have a of Ramanujan) was the first to notice these con-
single generator; this is responsible for the infinite nections; as a result [O] he proved the following:
families of congruences. As $ increases, however,
things rapidly become more complicated. Atkin’s For any pr ime $ ≥ 5, ther e exist
work is interesting for another reason: it marks an inf initely many congr uences of the f or m
early use of sophisticated computers in mathemat-
p(An + B) ≡ 0 (mod $).
ics. As he says, “It is often more difficult to
discover results in this subject than to prove (We note that if the arithmetic progression An + B
them, and an informed search on the machine may gives rise to such a congruence, then so do any of
enable one to find out precisely what happens.” its infinitely many subprogressions; we do not
count these as new when we speak of “infinitely
A Problem of Erdős many congruences”.) Shortly thereafter the first au-
Even after all of the beautiful discoveries described thor [Ahl] extended this result by showing that the
above, the general arithmetic properties of p(n) prime $ may in fact be replaced by an arbitrary
must seem rather mysterious. Indeed, we have prime power $k ; from this it can be shown that $
said nothing for any prime modulus $ greater than may in fact be replaced by any modulus M which
31, let alone for a general prime modulus. In this is coprime to 6 . An immediate consequence of
context we mention a conjecture of Erdős from the these results is the following:
1980s.
If $ ≥ 5 is pr ime, then a positive
If $ is a pr ime, then ther e exists an n
pr opor tion of natur al number s n
such that p(n) ≡ 0 (mod $).
have p(n) ≡ 0 (mod $).
If we reflect on this conjecture for a moment, we
This provides a very convincing proof of the
are struck by its weakness: it asserts only that
conjecture of Erdős mentioned above.
every prime divides at least one value of the
partition function. On the other hand, (until very More recently, the two authors [Ahl-O] have
recently) the known results were even weaker; the shown that congruences for p(n) are even more
best was a theorem of Schinzel and Wirsing, who widespread than these theorems indicate. To
proved the existence of a constant c such that, for explain this, let us return to Ramanujan’s original
large X, the number of primes $ < X for which results:
Erdős’s conjecture is true is ≥ c log log X . p(5n + 4) ≡ 0 (mod 5),
p(7n + 5) ≡ 0 (mod 7),
Recent Developments
p(11n + 6) ≡ 0 (mod 11).
In the past several years our understanding of the
arithmetic of p(n) has increased dramatically. All As Ramanujan’s conjectures indicate, these
of the advances have arisen from a single source: results may be written in a unified way. Namely,
the fact that values of the partition function are let λ$ denote the inverse of 24 modulo $ (in other
intimately related to the arithmetic of modular words, 24λ$ ≡ 1 (mod $) ). Then they assume the
forms. Modular forms have historically played a following form:

982 NOTICES OF THE AMS VOLUME 48, NUMBER 9

fea-ahlgren.qxp 8/21/01 2:42 PM Page 983

If $ = 5, 7, or 11, then p($n + λ$ ) ≡ 0 (mod $). In addition, f is required to be meromorphic on H

and at the cusps; if f is also holomorphic on H and
Now, for any prime $ ≥ 5 and any exponent k , vanishes at the cusps, then we call f a cusp form.
the results above guarantee the existence of We allow k to be an integer or half an integer (extra
infinitely many progressions An + B such that care must be taken in the latter case); note that the
p(An + B) ≡ 0 (mod $k ) . An important feature modular functions introduced above are just mod-
of the method used to prove the theorems above ular forms of weight zero. Every modular form
is that in every case, the progression An + B which f (z) has a Fourier expansion in powers of x = e2π iz ;
it produces is a subprogression of $n + λ$ (in other if f is a cusp form, then this expansion takes the
words, $ | A and B ≡ λ$ (mod $) ). As an example, form ∞
one of the simplest congruences guaranteed by 
f (z) = af (n)xn .
this theorem is n=1
4
p(59 · 13n + 111247) ≡ 0 (mod 13); When the weight k of a cusp form is integral,
in this case we have 111247 ≡ 1/24 (mod 13) . then the theory of Deligne and Serre is available
What the authors have shown recently is that for the study of the Fourier coefficients af . In par-
congruences are not confined to this single pro- ticular, there is a natural family of operators (the
gression modulo $ . In fact, we now know that if so-called Hecke operators) that act on spaces of
$ ≥ 5 is prime and k is any exponent, then infinitely modular forms. If f is a normalized eigenform for
this family, then Serre conjectured and Deligne
many congruences p(An + B) ≡ 0 (mod $k ) exist
proved the existence of a representation
within each of ($ + 1)/2 progressions modulo $ .
In other words, for each prime, slightly more than ρf : Gal(Q/Q) → GL2 (K)
half of such progressions contain congruences.
When $ = 11, for example, the relevant progressions (for some field K ) such that for all but finitely
are many primes Q we have
11n + 1, 11n + 2, 11n + 3,
Trace(ρf (FrobQ )) = af (Q).
11n + 5, 11n + 6, 11n + 8.
Here FrobQ denotes a Frobenius element at the
Of these, only Ramanujan’s own 11n + 6 had been prime Q . This result is extraordinarily powerful;
distinguished by the previous theory. The latest it allows us to study the Fourier coefficients of mod-
result provides a theoretical framework which ular forms using the structure of Galois groups.
explains every known partition function congru- If the weight k of a cusp form is half-integral,
ence. then we do not have the results of Deligne and Serre
at our disposal. There is, however, a correspon-
Modular Forms dence due to Shimura between cusp forms of half
We will try to indicate briefly how the theory of integral weight and certain forms of integral weight;
modular forms can be applied to the study of p(n) the Shimura correspondence is quite explicit and
in order to yield the results of the preceding sec- commutes in the best possible way with the action
tion. At the heart of the matter are the generating of the Hecke operators on the respective spaces.
function We saw above that the expansion
∞
 ∞  
1 ∞
p(n)xn = n+1 n
1 − xn 1/η(24z) = p x = x−1 + x23 + . . .
n=0 n=1 24
n=−1
and Dedekind’s eta function
contains every value of the partition function. Now
∞
 1/η(24z) is a modular form on Γ0 (576) . However,
η(z) = x1/24 (1 − xn ) (here x := e2π iz ). it has two major deficiencies: the weight is −1/2 ,
n=1
and it has a pole at every cusp. So none of the the-
Combining the last two formulae gives ories above seem to apply. It turns out, however,
  that starting with this expansion, one can con-
∞
 n+1 n struct half-integral weight cusp forms which still
1/η(24z) = p x = x−1 + x23 + . . . .
24 preserve much information about the values of p(n)
n=−1
modulo powers of primes. From these cusp forms
Loosely speaking, a modular form of weight k the theory of Deligne and Serre, filtered through
on the subgroup Γ0 (N) is a function f on the upper Shimura’s correspondence, yields the results of
half-plane H which satisfies a transformation prop- the preceding section.
erty of the form
L -Functions and Arithmetic
  
az + b ab Since modular forms play such an important role
f = (cz + d)k f (z) for all c d ∈ Γ0 (N).
cz + d in partition congruences, it is natural to suspect

OCTOBER 2001 NOTICES OF THE AMS 983

fea-ahlgren.qxp 8/21/01 2:42 PM Page 984

that there may be deeper connections between Conjecture (a vast generalization of the Birch and
partitions and “modular” objects. As it turns out, Swinnerton-Dyer Conjecture) then implies that
this is indeed the case. L(MD,$ , ($ − 3)/2) = ΩD,$ · # X(MD,$ ).
To motivate the connection, consider the fol-
lowing classical Diophantine question (already of Assuming the Bloch-Kato Conjecture, it can be
interest to ancient Greek and Arab scholars): shown, for many n, that
W hich integer s D ar e ar eas of r ight p(n) ≡ 0 (mod $) =⇒ # X(MD,$ ) ≡ 0 (mod $),
tr iangles w ith r ational number sidelengths? where D depends on n. These two conditions are
Such numbers D are known as congruent numbers. probably equivalent, and so it is likely that the
Simple arguments show that a number D is con- divisibility of p(n) often dictates the presence of
gruent precisely when there are infinitely many elements of order $ in these Tate-Shafarevich
rational points (x, y) on the elliptic curve groups. So, perhaps surprisingly, it seems that
congruences like Ramanujan’s are connected to
ED : y 2 = x3 − D 2 x. some highly abstract creations of modern number
theory.
How does one determine whether such a curve
has infinitely many points? The Birch and The Future?
Swinnerton-Dyer Conjecture, one of the main
The beginnings of the partition function are
outstanding conjectures in number theory (and a
extraordinarily humble; after all, what could be
million-dollar Clay Mathematics Institute prob- simpler than addition and counting? Despite its
lem), provides the solution. humble start, the history of the partition function
Let L(ED , s) denote the Hasse-Weil L -function includes connections to many central areas of
attached to ED ; this is an analytic function whose number theory, from the work of Euler to the
definition depends on the behavior of ED modulo birth of the circle method to the modern theory of
primes p. For the congruent number problem the modular forms and L -functions. It will be quite
conjecture implies that interesting to see what further connections the
L(ED , 1) = 0 ⇐⇒ D is congruent. future will reveal.

In addition, the conjecture gives a precise formula References

dictating the analytic behavior of L(ED , s) at s = 1 . [Ahl] S. AHLGREN, The partition function modulo com-
For instance, if L(ED , 1) = 0 , then the conjecture posite integers M , Math. Ann. 318 (2000), 795–803.
asserts that [Ahl-O] S. AHLGREN and K. ONO, Congruence properties for
the partition function, Proc. Nat. Acad. Sci. U.S.A.,
L(ED , 1) = ΩD · # X(ED ). to appear.
Here ΩD is an explicit transcendental number, and [A] G. E. ANDREWS, The Theory of Partitions, Cambridge
X(ED ) is the Tate-Shafarevich group of ED . (The Univ. Press, 1998.
[A-G] G. E. ANDREWS and F. GARVAN, Dyson’s crank of a
Tate-Shafarevich group is a certain Galois
partition, Bull. Amer. Math. Soc. (N.S.) 18 (1988),
cohomology group which measures the extent to 167–171.
which the local-global principle fails for ED .) [R] B. C. BERNDT and K. ONO, Ramanujan’s unpublished
In the early 1980s Jerrold Tunnell, using the manuscript on the partition and tau functions with
works of Shimura and Waldspurger (see [K] for a commentary, The Andrews Festschrift (D. Foata and
good account), constructed two modular forms G. N. Han, eds.), Springer-Verlag, 2001, pp. 39–110.
of weight 3/2 whose coefficients “interpolate” the [G] F. GARVAN, New combinatorial interpretations of
Ramanujan’s partition congruences mod 5, 7 and 11,
square roots of the L(ED , 1) . Together with the
Trans. Amer. Math. Soc. 305 (1988), 47–77.
Birch and Swinnerton-Dyer Conjecture, these [G-K-S] F. GARVAN, D. KIM, and D. STANTON, Cranks and
modular forms provide a complete solution to t-cores, Invent. Math. 101 (1990), 1–17.
the congruent number problem. [G-O] L. GUO and K. ONO, The partition function and the
Recently, Li Guo and the second author [G-O] arithmetic of certain modular L -functions, Internat.
have shown that if 13 ≤ $ ≤ 31 is prime, then Math. Res. Notices 21 (1999), 1179–1197.
certain half-integral weight modular forms [K] N. K OBLITZ , Introduction to Elliptic Curves and
whose coefficients interpolate values of p(n) Modular Forms, Springer-Verlag, 1993.
[O] K. O NO , Distribution of the partition function
modulo $ behave in a manner somewhat similar
modulo m , Ann. of Math. 151 (2000), 293–307.
to Tunnell’s modular forms. In particular, they
showed that there are modular motives MD,$ (these
may be viewed as analogs of elliptic curves) whose
L -functions L(MD,$ , s) have the property that the Note: Photograph of L. Euler courtesy of the
square roots of L(MD,$ , ($ − 3)/2) are related in a Institut Mittag-Leffler. The photograph of
predictable way to the coefficients of these H. Rademacher was provided by Bruce Berndt.
modular forms. The truth of the Bloch-Kato

984 NOTICES OF THE AMS VOLUME 48, NUMBER 9