From The Cover: Partition congruences and the Andrews-Garvan-Dyson crank

Karl Mahlburg
PNAS 2005;102;15373-15376; originally published online Oct 10, 2005;
doi:10.1073/pnas.0506702102
This information is current as of March 2007.

Online Information High-resolution figures, a citation map, links to PubMed and Google Scholar,
& Services etc., can be found at:
www.pnas.org/cgi/content/full/102/43/15373
Related Articles A related article has been published:
www.pnas.org/cgi/content/full/102/43/15275
This article has been cited by other articles:
www.pnas.org/cgi/content/full/102/43/15373#otherarticles
E-mail Alerts Receive free email alerts when new articles cite this article - sign up in the box
at the top right corner of the article or click here.
Rights & Permissions To reproduce this article in part (figures, tables) or in entirety, see:
www.pnas.org/misc/rightperm.shtml
Reprints To order reprints, see:
www.pnas.org/misc/reprints.shtml

Notes:
 SEE COMMENTARY
Partition congruences and the
Andrews–Garvan–Dyson crank
Karl Mahlburg†
Department of Mathematics, University of Wisconsin, 418 Van Vleck Hall, E B, 480 Lincoln Drive, Madison, WI 53706

Communicated by George E. Andrews, Pennsylvania State University, University Park, PA, August 4, 2005 (received for review June 1, 2005)

In 1944, Freeman Dyson conjectured the existence of a ‘‘crank’’ 1

function for partitions that would provide a combinatorial proof of N共m, 5, 5n ⫹ 4兲 ⫽ 䡠 p共5n ⫹ 4兲 0 ⱕ m ⱕ 4, [1.4]
5
Ramanujan’s congruence modulo 11. Forty years later, Andrews
and Garvan successfully found such a function and proved the where N(m, N, n) is the number of partitions ␭ of n for which
celebrated result that the crank simultaneously ‘‘explains’’ the rank( ␭ ) ⬅ m (mod N). His observations were proven 10 years
three Ramanujan congruences modulo 5, 7, and 11. This note later by Atkin and Swinnerton-Dyer (6). However, even the
announces the proof of a conjecture of Ono, which essentially smallest examples show that the rank does not equally dissect the
asserts that the elusive crank satisfies exactly the same types of Ramanujan congruence modulo 11.
general congruences as the partition function. Instead, Dyson conjectured that there would be a ‘‘crank’’
function for the final Ramanujan congruence, although it wasn’t
1. Introduction and Statement of Results until 40 years had passed that Andrews and Garvan defined the
function and showed that
Whether these guesses are warranted by the evidence, I
leave to the reader to decide. Whatever the final verdict of 1
posterity may be, I believe the ‘crank’ is unique among M共m, 11, 11n ⫹ 6兲 ⫽ 䡠 p共11n ⫹ 6兲. [1.5]
11
arithmetical functions in having been named before it was
discovered. May it be preserved from the ignominious fate Here M(m, N, n) is defined for the crank just as N(m, N, n) was
of the planet Vulcan! for the rank (7, 8). In these historic works, they also showed that
Freeman Dyson (1) the crank dissects the Ramanujan congruences modulo 5 and 7
in a different way than the rank.
If ␭ 1 ⫹ ␭ 2 ⫹ 䡠 䡠 䡠 ⫹ ␭ s ⫹ 1 ⫹ 䡠 䡠 䡠 ⫹ 1 has exactly r ones, then
A partition of n is a non-increasing list of positive integers
␭ 1 ⱖ ␭ 2 ⱖ 䡠 䡠 䡠 ⱖ ␭ k that sum to n; we write 兩 ␭ 兩 ⫽ n. The
partition function p(n) is defined to count the number of distinct
let o( ␭ ) be the number of parts of ␭ that are strictly larger than
r. The crank is given by

再
partitions of a given integer n.
␭1 if r ⫽ 0,
Ramanujan’s celebrated congruences for the partition func- crank共 ␭ 兲 :⫽ [1.6]
o共 ␭ 兲 ⫺ r if r ⱖ 1.
tion state that
Clearly p(n) ⫽ M(0, N, n) ⫹ 䡠 䡠 䡠 ⫹ M(N ⫺ 1, N, n), and for
p共5n ⫹ 4兲 ⬅ 0 共mod 5兲, the Ramanujan congruences, all of these summands are equal.
p共7n ⫹ 5兲 ⬅ 0 共mod 7兲, [1.1] However, this behavior is atypical, and an unpublished conjec-
p共11n ⫹ 6兲 ⬅ 0 共mod 11兲. ture of Ono asserted that a different approach would show that
congruences for the partition function are related to the crank
Following the spirit of Ramanujan’s own work, Watson and in a universal manner.
Atkin extended these congruences to arbitrary powers of 5, 7,
and 11 (2). Sporadic progress was made in proving congruences Conjecture (Ono). For every prime ᐉ ⱖ 5 and integer ␶ ⱖ 1, there are
for primes up to 31, until Ono’s seminal work from 2000 (3) infinitely many non-nested arithmetic progressions An ⫹ B for
achieved a surprising improvement. He proved the existence of which
infinite families of partition congruences for every prime ᐉ ⱖ 5
by developing the p-adic theory of half-integral weight modular M共m, ᐉ, An ⫹ B兲 ⫽ 0 共mod ᐉ ␶兲,
forms. This result was expanded to include congruences for
for every 0 ⱕ m ⱕ ᐉ ⫺ 1.
every modulus coprime to 6 by Ahlgren and Ono (4, 5). These
In fact, the following theorem shows that the crank function
results are typically much more complicated than Ramanujan’s
actually satisfies congruences beyond those predicted by Ono.
original congruences, as displayed by the example
Theorem 1.1. Suppose that ᐉ ⱖ 5 is prime and that ␶ and j are
p共48037937n ⫹ 1122838兲 ⬅ 0 共mod 17兲. [1.2]
positive integers. Then there are infinitely many non-nested arith-
metic progressions An ⫹ B such that
However, there is another side to the story, which began when
Freeman Dyson wondered whether a simple statistic might M共m, ᐉj, An ⫹ B兲 ⬅ 0 共mod ᐉ ␶兲,
group the partitions into natural classes and explain the Ra-
manujan congruences. The rank of the partition ␭ 1 ⫹ ␭ 2 ⫹ 䡠 䡠 䡠 ⫹ simultaneously for every 0 ⱕ m ⱕ ᐉ j ⫺ 1.
␭ k, is defined by Remark: The frequency of such congruences is quantified later
in this note by Theorem 4.1. An obvious implication of Theorem
MATHEMATICS

rank共 ␭ 兲 :⫽ ␭ 1 ⫺ k. [1.3]

Dyson (1) observed empirically that this function decomposes See Commentary on page 15277.
the Ramanujan congruences modulo 5 and 7 into classes of equal †E-mail: mahlburg@math.wisc.edu.
size. For example, © 2005 by The National Academy of Sciences of the USA

www.pnas.org兾cgi兾doi兾10.1073兾pnas.0506702102 PNAS 兩 October 25, 2005 兩 vol. 102 兩 no. 43 兩 15373–15376

1.1 is that p(An ⫹ B) ⬅ 0 (mod ᐉ␶) as well, corresponding to the 1. For a prime p ⱉ N and a non-integral half-integer k, the action
congruences found by Ahlgren and Ono. of T(p 2) is space-preserving, i.e.,
This announcement begins with a review of modular forms in
Section 2. Following that, Section 3 explains how to write the f共z兲兩T共p2兲 僆 Mk共⌫0共N兲, ␹兲.
generating function of the crank in terms of Klein forms. A
2. If ␺ is a character with modulus M, then
condensed proof of Theorem 1.1 is found in Section 4.
f共z兲 丢 ␺ 僆 Mk共⌫0共NM2兲, ␹␺2兲.
2. Half-Integral Weight Modular Forms
This section contains the basic definitions and properties of 3. If f(z) is a cusp form, t ⱖ 1 and 0 ⱕ r ⱕ t ⫺ 1, then
modular forms that will be needed in Section 4 (see ref. 2 for
details). Let ⌫ :⫽ SL 2(⺪) denote the full modular group of 2 ⫻
2 matrices, and for a given modulus N, let ⌫ 0(N) and ⌫ 1(N)
冘
n⬅r 共mod t 兲
a共n兲q n 僆 S k共⌫ 1共Nt 2兲兲.

denote the subsets of matrices that are congruent to (ⴱ0ⴱⴱ) and (10ⴱ1)
(mod N), respectively. If k 僆 21⺪, and ⌫⬘ 傺 ⌫ is such a congruence
subgroup, then M !k(⌫⬘) denotes the vector space of nearly
holomorphic modular forms of weight k for the subgroup ⌫⬘
(these are holomorphic on the upper half-plane H and mero-
morphic at the cusps of ⌫⬘). The forms that are holomorphic at
the cusps are denoted by M k(⌫⬘), and the forms that vanish at
the cusps are denoted by S k(⌫⬘). If ⌫⬘ ⫽ ⌫ 0(N), then M !k(⌫ 0(N),
␹ ), M k(⌫ 0(N), ␹ ), or S k(⌫ 0(N), ␹ ) denotes the appropriate space
of modular forms of weight k on ⌫ 0(N) with Nebentypus
character ␹.
For a meromorphic function f on the upper half plane H, and
an integer k, the ‘‘slash operator’’ is defined by f(z) 兩 k( ac bd) :⫽
(cz ⫹ d) ⫺k f ((az ⫹ b)兾(cz ⫹ d)), for any ( ac bd) 僆 ⌫. A key
property is that this is a group action in the sense that for any M 1,
M 2 僆 ⌫, (f(z) 兩 k M 1) 兩 k M 2 ⫽ f(z) 兩 k M 1M 2.
Let q :⫽ e 2␲iz. If f ⫽ ¥ ⬁n⫽0 a(n)q n 僆 M k(⌫ 0(N), ␹ ) and ␺ is
a Dirichlet character, then the twist of f by ␺ is
In Section 4 we will need to simultaneously find congruences
for two half-integral weight modular forms of different weights
and levels. This situation is partially addressed by Ono’s theorem
2.2 in ref. 9, Ahlgren and Ono’s proof of lemma 3.1 in ref. 5, and
Serre’s arguments in ref. 10. The additional ingredients needed
to prove the next theorem are the integral weight Hecke
operators, the famous Shimura correspondence, and the decom-
position S k(⌫ 1(N)) ⫽ QS k(⌫ 0(N), ␹ ), where the sum is over all
even characters ␹.
Theorem 2.2. Suppose that k i and N i are positive integers for 1 ⱕ i ⱕ
r, and let g 1(z), . . . , g r(z) be half-integer weight cusp forms with
algebraic integer coefficients such that g i(z) 僆 S ki⫹1/2(⌫ 1(N i)). If
M ⱖ 1, then a positive proportion of primes p ⬅ ⫺1 (mod
N 1 䡠 䡠 䡠 N rM) have the property that for every i,
gi共z兲兩T共p2兲 ⬅ 0 共mod M兲.

f共z兲 丢 ␺ :⫽ 冘
nⱖ0
␺ 共n兲a共n兲q n. [2.1] 3. The Crank Generating Function and Klein Forms
Let M(m, n) be the number of partitions ␭ of n such that
crank( ␭ ) ⫽ m. Using the generating function found by Andrews
Simple facts about Gauss sums allow one to rewrite the twist and Garvan (8), define
of a modular form using the slash operator in a manner that

冘 冘 冘
is independent of the weight. If p is a prime and ⬁

F共x, z兲 :⫽ M共m, n兲x mq n ⫽ x crank共␭兲q 兩␭兩

冘冉 冊
p⫺1
v 2␲iv/p m⫽⫺⬁ nⱖ0 ␭
gp :⫽ e
p
写 1⫺q n
v⫽1
⫽ . [3.1]
共1 ⫺ xq n兲共1 ⫺ x ⫺1q n兲
is the standard Gauss sum, then nⱖ1

冉 冊 冘冉 冊 冉 冊
p⫺1 Consider a positive integer N and set ␨ :⫽ e 2␲i/N. For any
• gp v 1 ⫺v兾p residue class m (mod N), elementary calculations give the
f共z兲 丢 ⫽ f共z兲 兩 . [2.2]
p p p 0 1 generating function for the crank
v⫽1

冘 冘
N⫺1
The half-integral weight Hecke operators are important tools 1
for finding congruences among the coefficients of modular M共m, N, n兲qn ⫽ F共␨s, z兲␨⫺ms
N
forms. If f(z) ⫽ ¥ nⱖ0 a(n)q n 僆 M k(⌫ 0(N), ␹ ) and k is not an nⱖ0 s⫽0
integer, then for a prime p ⱉ N the Hecke operator is defined by

冘冉 冉冊 冘 冉写 冊
N⫺1
1 共1 ⫺ qn兲
n k⫺3/2 ⫽ ␨⫺ms ⫺s n .
f共z兲兩T共p 兲 :⫽
2
a共p 2n兲 ⫹ ␹ *共p兲 p a共n兲 N 共1 ⫺ ␨ q 兲共1 ⫺ ␨sqn兲
p s⫽0 nⱖ1
nⱖ0

冉 冊冊
[3.2]
n
⫹ ␹ *共p 兲p2 2k⫺2
a 2 n
q , [2.3] To prove congruences for this function, we need to show that
p
it is a modular form. Recall Dedekind’s eta-function
where
␩共z兲 :⫽ q 1/24 写 共1 ⫺ q n兲,
冉 冊
[3.3]
共 ⫺ 1兲k⫺1/2 nⱖ1
␹*共n兲 :⫽ ␹共n兲 .
n
which is a modular form of weight 1兾2. Perhaps less familiar are
All of these operators act on spaces of modular forms in an the Klein forms, which were studied extensively by Kubert and
easily described manner. Lang (11).
Definition 3.1: Let 1 ⱕ s ⱕ N ⫺ 1. The (0, s)–Klein form is
Proposition 2.1. Suppose that f(z) ⫽ ¥ nⱖ0 a(n)q n 僆 M k(⌫ 0(N), ␹ ). given by

15374 兩 www.pnas.org兾cgi兾doi兾10.1073兾pnas.0506702102 Mahlburg

 SEE COMMENTARY
t0,s共z兲 :⫽
␻s
2␲i
䡠 写
nⱖ1
共1 ⫺ ␨ sq n兲共1 ⫺ ␨ ⫺sq n兲
共1 ⫺ q n兲 2
,
Let G m(z) and P(z), respectively, denote the two summands in
the final expression of Eq. 4.2.
As explained in ref. 5, if t is a positive integer, then there is a
Dirichlet character ␹ ᐉ,t such that
where ␻ s :⫽ ␨ s/2(1
⫺ ␨ ⫺s).
Now write d៮ for the least residue of d modulo N and set ␩ ᐉ 共z兲
t

exp(z) :⫽ e 2␲iz. Understanding the action of ⌫ on the Klein forms Et共z兲 :⫽ 僆 M ᐉt⫺1共⌫ 0共ᐉ t兲, ␹ ᐉ,t兲. [4.3]
is an important aspect of the proof of Theorem 1.1. The following ␩ 共ᐉ tz兲 2

formula comes from equation K2 on p. 28 of ref. 11.

This form vanishes at every cusp a兾c with ᐉ t not dividing c, and
␶

Proposition 3.2. If ( ac bd) 僆 ⌫ 0(N), then also satisfies E t(z) ᐉ ⬅ 1 (mod ᐉ ␶⫹1) for any ␶ ⱖ 0. Similar facts
about eta-quotients, along with Corollary 3.3, show that
t0,s共z兲兩⫺1 冉 冊 a
c
b
d
⫽ ␤䡠t0,ds共z兲,
P共z兲 ⫽
␩ᐉ共ᐉz兲
僆 Mᐉ⫺1 ⌫0共ᐉ兲,
•
冉 冉 冊冊 , [4.4]
␩共z兲 2 ᐉ
where ␤ is a root of unity given by ␤ :⫽ exp(cs⫹(ds⫺ds))兾2N ⫺
cds 2兾2N 2). and
A simple calculation shows that this ‘‘multiplier system’’ is
!
always trivial for a certain congruence subgroup. Gm共z兲 僆 M ᐉ⫹1共⌫1共2N2兲兲. [4.5]
2
Corollary 3.3. If 1 ⱕ s ⱕ N ⫺ 1, then t 0,s(z) 僆 M !⫺1(⌫ 1(2N 2)).
Returning to the crank generating function, for 1 ⱕ s ⱕ N ⫺ For any function f(z) with a q-expansion, define

冉冊
1 Eq. 3.1 becomes
•
f̃共z兲 :⫽ f共z兲 ⫺ ␧ᐉf共z兲 丢 .
1 ␻sq1/24 ᐉ
F共␨s , z兲 ⫽ 䡠 . [3.4]
␩共z兲t0,s共z兲 2␲i
The line of argument in ref. 5 implies that if ␶ is sufficiently large,
The generating function for the partition function is ¥ nⱖ0 p(n)q n then there is some integer ␭⬘ ⱖ 1 and some character ␹ such that
⫽ ⌸ nⱖ1 1兾(1⫺q n), which is the s ⫽ 0 term in Eq. 3.2. Hence
P̃共24z兲 ␶
䡠 E 共24z兲ᐉ 僆 S␭⬘⫹1/2共⌫0共576ᐉmax兵3, j⫹1其兲, ␹ 兲.
冘 冘 冘 ␩ᐉ共24ᐉz兲 j⫹1
N⫺1 ⫺ms
1 ␻ s␨ 1
M共m, N, n兲qn ⫽ 䡠 q1/24 ⫹ p共n兲qn .
2␲iN ␩共z兲t0,s共z兲 N [4.6]
nⱖ0 s⫽1 nⱖ0

[3.5] We now show that a similar cusp form exists for G m(z).

4. The Proof of Theorem 1.1 Lemma 4.2. If ␶ is sufficiently large, then there is some ␭ ⱖ 1 such that

For a prime ᐉ ⱖ 5, set ␦ ᐉ :⫽ (ᐉ 2 ⫺ 1)兾24, and define ␧ ᐉ :⫽

Then define the set
␦ᐉ
ᐉ
. 冉 冊 Gm共24z兲 ␶
䡠 E 共24z兲ᐉ 僆 S␭⫹1/2共⌫1共576ᐉ2N2兲兲. [4.7]
␩ᐉ共24ᐉz兲 j⫹1

Sᐉ :⫽ 再 0ⱕ␤ⱕᐉ⫺1 冏 冉 ␤ ⫹ᐉ ␦ 冊 ⫽ 0 or ⫺␧ 冎 .
ᐉ
ᐉ Proof: Recalling Eq. 4.3, if ␶ is sufficiently large, then it only
needs to be shown that G m(z)兾 ␩ ᐉ(ᐉz) vanishes at each cusp a兾c
[4.1] with ᐉN 兩 c. If ( ac bd) 僆 ⌫ 0(ᐉN), then the expansion of 1兾 ␩ ᐉ(ᐉz)
2
at that cusp is (q ⫺ᐉ /24 ⫹ . . .) up to a root of unity. Thus, it must
The following theorem is a more precise description of the be proven that the expansion of G m(z) at a兾c is (*q h ⫹ . . .) for
congruences satisfied by the crank function, and it clearly implies some h ⬎ ᐉ 2兾24.
Theorem 1.1. Using Proposition 3.2 and Eq. 4.4, calculate

冉 冊 冉冊
Theorem 4.1. Suppose that ᐉ ⱖ 5 is prime, ␶ and j are positive
冏 冘
N⫺1
integers, and ␤ 僆 S ᐉ. Then a positive proportion of primes Q ⬅ ⫺1 a b 1 d ␩ᐉ共ᐉz兲 ␻s␨⫺ms
Gm共z兲 ᐉ⫹1 ⫽ , [4.8]
(mod 24ᐉ) have the property that for every 0 ⱕ m ⱕ ᐉ j ⫺ 1 2
c d 2␲i ᐉ ␩共z兲 ␤st0,ds共z兲
s⫽1

冉
M m, ᐉj ,
Q 3n ⫹ 1
24
⬅0 冊 共mod ᐉ ␶兲, where the ␤ s are the roots of unity described by Proposition 3.2.
To find the expansion of
for all n ⬅ 1 ⫺ 24 ␤ (mod 24ᐉ) that are not divisible by Q.
For the rest of this section, let N :⫽ ᐉ j be a fixed power of a
fixed prime ᐉ ⱖ 5. Theorem 2.2 applies to modular forms with
Gm共z兲 丢 冉冊
•
ᐉ
,

algebraic integer coefficients, and thus Eq. 3.5 must be rescaled

first observe that for any v⬘
in defining

gm共z兲 :⫽ 冉冘 N䡠M共m, N, n兲qn⫹␦ᐉ 冊写 共1 ⫺ qᐉn兲ᐉ

冉 1
0
⫺v兾ᐉ
1 冊冉 冊 冉a
c
b
d
⫽
a⬘
c⬘
b⬘
d⬘ 冊冉 1
0
⫺v⬘兾ᐉ
1
, 冊 [4.9]
MATHEMATICS

nⱖ0 nⱖ1
where

冘 冉 冊 冉 冊
N⫺1 ᐉ ⫺ms ᐉ
1 ␩ 共ᐉz兲 ␻s␨ n 共ᐉz兲 a⬘ b⬘ a ⫺ cv兾ᐉ b ⫺ cvv⬘兾ᐉ2 ⫹ 共av⬘ ⫺ dv兲兾ᐉ
⫽ 䡠 ⫹ . [4.2] :⫽ .
2␲i ␩共z兲 t0,s共z兲 ␩共z兲 c⬘ d⬘ c d ⫹ cv⬘兾ᐉ
s⫽1

Mahlburg PNAS 兩 October 25, 2005 兩 vol. 102 兩 no. 43 兩 15375

Pick v⬘ ⬅ d 2v (mod ᐉ) for the subsequent arguments, and let g ᐉ
be the Gauss sum as in Eq. 2.2. Proposition 3.2 and Eqs. 4.8 and
hm,␤共z兲 :⫽ 冘
n⬘⬅␤⫹␦ᐉ 共mod ᐉ 兲
NM共m, N, n⬘ ⫺ ␦ ᐉ兲q 24n⬘⫺ᐉ
2

4.9 show that

冉 Gm共z兲 丢 冉 冊冊冏 冉 冊
•
ᐉ
ᐉ⫹1
a
c
b
d
⫽ 冘
n⬅24 ␤ ⫺1 共 mod 24ᐉ 兲
冉
NM m, N,
n⫹1 n
24
q . 冊
2
[4.11]

冉 冊 冘 冘冉 冊 冏冉10 冊
N⫺1 ᐉ⫺1
gᐉ d⬘ v ␩ᐉ共ᐉz兲 ␻s␨⫺ms ⫺v⬘兾ᐉ But Theorem 4.3 implies that
⫽ 䡠 .
2␲iᐉ ᐉ ᐉ ␩共z兲 ␤⬘st0,d⬘s共z兲 1
s⫽1 v⫽1 hm,␤共z兲 ⬅ Fm,␤共z兲 ⫹ F⬘␤共z兲 共mod ᐉ ␶兲, [4.12]
[4.10] where Fm,␤(z) and F⬘␤(z) are defined by restricting Fm(z) and
Since d⬘ ⬅ d (mod N) and ᐉN 兩 c, a short computation verifies that F⬘(z) to only those indices with n⬘ ⬅ ␤ ⫹ ␦ᐉ (mod ᐉ).
␤s⬘ ⫽ ␤s. An additional calculation then shows that the first term of Proposition 2.1 and Theorem 2.2 then show that a positive
Eq. 4.8 is ␧ᐉ times the first term of Eq. 4.10. Thus, the expansion of proportion of primes Q ⬅ ⫺1 (mod 24ᐉ) have the prop-
G̃m(z) has the form (*q␦ᐉ⫹1 ⫹ . . .), and ␦ᐉ ⫹ 1 ⬎ ᐉ2兾24. erty that
The following theorem is now proved, as F⬘(z) is defined by
Fm,␤共z兲兩T共Q2兲 ⬅ F⬘␤共z兲兩T共Q2兲 ⬅ 0 共mod ᐉ ␶兲,
Eq. 4.6, and F m(z) is described by Lemma 4.2.
for all m. Replace n by Qn in Eq. 2.3 to see that
Theorem 4.3. For any ␶ ⱖ 0 and 0 ⱕ m ⱕ N ⫺ 1, there is a character
␹, positive integers ␭ and ␭⬘, and modular forms F m(z) 僆
S ␭⫹1/2(⌫ 1(576ᐉ 2N 2)) and F⬘(z) 僆 S ␭⬘⫹1/2(⌫ 0(576ᐉ max{3, j⫹1}), ␹ )
such that
冉
NM m, N,
Q 3n ⫹ 1
24
⬅0 冊 共mod ᐉ ␶兲, [4.13]

for all n ⬅ 1 ⫺ 24 ␤ (mod 24ᐉ) that are not divisible by Q.

g̃m共24z兲
⬅ Fm共z兲 ⫹ F⬘共z兲 共mod ᐉ ␶兲. Dividing by N then completes the proof.
␩ᐉ共24ᐉz兲
I thank Ken Ono for his great support throughout the course of this
Deduction of Theorem 4.1 from Theorem 4.3: Suppose that ␤ 僆 work. I also thank the anonymous referees for comments that im-
Sᐉ. Starting from the definition of gm(z) in Eq. 4.2, restrict g̃m(24z)兾 proved the exposition and presentation of the results in this work. I was
␩ᐉ(24ᐉz) to those indices n⬘ ⬅ ␤ ⫹ ␦ᐉ (mod ᐉ), which then gives a supported by a National Science Foundation Graduate Research
new series (scaled by a factor of 1兾2 when ␤ ⬅ ⫺␦ᐉ (mod ᐉ)) Fellowship.

1. Dyson, F. (1944) Eureka (Cambridge, U.K.) 8, 10–15.
2. Ono, K. (2004) The Web of Modularity, CBMS Regional Conference Series in
Mathematics (American Mathematical Society, Providence, RI), Vol. 102.
3. Ono, K. (2000) Ann. Math. 151, 293–307.
4. Ahlgren, S. (2000) Math. Ann. 318, 795–803.
5. Ahlgren, S. & Ono, K. (2001) Proc. Natl. Acad. Sci. USA 98, 12882–12884.
6. Atkin, A. L. & Swinnerton-Dyer, P. (1954) Proc. London Math. Soc. 4, 84–106.
7. Garvan, F. (1988) Trans. Am. Math. Soc. 305, 47–77.
8. Andrews, G. & Garvan, F. (1988) Bull. Am. Math. Soc. 18, 167–
171.
9. Ono, K. (2001) J. Reine Angew. Math. 533, 81–97.
10. Serre, J.-P. (1976) L’Ens. Math. 22, 227–260.
11. Kubert, D. & Lang, S. (1981) Modular Units, Grundlehren der mathematischen
Wissenschaften (Springer, New York), Vol. 244.

15376 兩 www.pnas.org兾cgi兾doi兾10.1073兾pnas.0506702102 Mahlburg