Invent. math.

101, 1-17 (1990)

[ l l ue~l tioTleS
mathematicae
9 Springer-Verlag 1990

Cranks and t-cores

F r a n k G a r v a n , 1 D o n g s u Kim, 2 and Dennis S t a n t o n *
School of Mathematics, Physics, Computing and Electronics, Macquarie University, Sydney,
NSW 2109, Australia.
2 School of Mathematics, University of Minnesota, Minneapolis, MN 55455, USA

Summary. New statistics on partitions (called cranks) are defined which com-
binatorially prove Ramanujan's congruences for the partition function modulo 5,
7, 11, and 25. Explicit bijections are given for the equinumerous crank classes. The
cranks are closely related to the t-core of a partition. Using q-series, some explicit
formulas are given for the number of partitions which are t-cores. Some related
questions for self-conjugate and distinct partitions are discussed.

1. Introduction

Let p(n) be the number of partitions of n [1]. Dyson [7] proposed a combinatorial
proof of Ramanujan's congruence
p(5n+4)-O mod5.
He defined an integral statistic on partitions, called the rank, whose value rood 5
split the set of partitions of 5n + 4 into 5 equal classes. He further conjectured that
the rank also proved
p(7n + 5) - O m o d 7 ,
and hypothesized a statistic, called the crank, which would similarly prove
p(lln+6)-0 modll.
Atkin and Swinnerton-Dyer [5] proved Dyson's conjecture for 5 and 7. In [9], a
crank for 11 as well as new cranks for 5 and 7 were found relative to vector
partitions. Later, Andrews and Garvan [3] were able to define a crank, in terms of
ordinary partitions, for 5, 7 and 11, thus completing the solution of Dyson's crank
conjecture. New relations for the crank of partitions mod 8, 9 and 10 have been
found in [10].

* This work was partially supported by NSF grant DMS: 8700995

2 F. Garvan et al.

Explicit bijections between these equinumerous classes have not been found in
any of these three cases. This should be related to another possible approach to
these problems: split the set of partitions of 5n + 4 into p ( 5 n . + 4)/5 classes, each of
size 5. Hopefully, the bijections among the five classes should be realized by a 5-
cycle which acts on all partitions of 5n + 4 with no fixed points. The orbits of the 5-
cycle are the p ( 5 n + 4)/5 classes. A crank statistic should be defined which is
{0, 1, 2, 3, 4} on an orbit. A 5-cycle, but no crank, was found by computer in [11].
In this paper, we complete this program by finding dihedral groups of size 10, 14,
and 22 which act on the partitions of 5n + 4, 7n + 5, and 1 In + 6, respectively. The
5 (7 or 11) cycles in these groups have no fixed points, thus proving the con-
gruences. We also define new statistics on partitions, i.e. new cranks, which split the
set of partitions into equinumerous classes. A cycle in the dihedral groups increases
the crank by 1, thus supplying an explicit bijection between the crank classes. Our
definitions of the cycles and the cranks are completely combinatorial.
The key ingredients to our proof are two bijections for partitions and t-cores,
which are given in w They allow us to find dihedral groups as symmetry groups of
quadratic forms. There are interesting q-series attached to these forms, and these q-
series imply some remarkable explicit formulas (Theorems 4 and 5 in w for the
number of partitions which are t-cores, t = 5 or 7. The explicit cranks are given in
Theorems 2 and 3, and in Proposition 1 of w and w A crank for 25n + 24 is given
in w Similar bijections and generating functions for self-conjugate partitions and
partitions with distinct parts are given in w and w
We shall use the notation for generating functions from q-series, e.g.
k-1
(a;q)k = I~ (1 - a q l ) ,
i=O
SO

p(n)q" = 1
.:o (q;q)~

2. Two bijections

In this section we give two bijections relating partitions and t-cores.

First we set the notation. Let P be the set of all partitions. For any 2 ~ P , let 12]
denote the number that 2 partitions. Fix a positive integer t. Let Pt - core be the set of
partitions which are t-cores. Recall that there are two equivalent definitions of a t-
core: a partition 2 is a t-core if, and only if, 2 has no hook numbers that are
multiples of t [13, 2.7.40]; or if, and only if, 2 has no rim hooks that are multiples of
t [13, p. 75]. We let at(n ) be the number of partitions of n which are t-cores.
The first bijection is well-known [13, 2.7.17].
Bijection 1. T h e r e is a b i j e c t i o n (a, : P ~ Pt-core X P • P x . . . x P ,
4~1(2) = (7~, 20, 2~ . . . . . 2 , _ ~) ,
t-1
s u c h t h a t 121 = I-~l + t Zi=o I;,il.
Cranks and t-cores 3

The generating function identity equivalent to Bijection 1 is

p(n)q"- 1 ~ a,(n)q". (2.1)

,=0 (q'; q')~ ,=o
The second bijection is on t-cores only.

Bijeetion 2. There is a bijection 42: P t - c o , e ~ { T t = ( n o , nl . . . . . nt-x): n i e Z ,

no+ ... +nt-1 =0},where
I)~l = tllnll2/2 + / ~ ' n , /~ = (0, 1. . . . . t - 1).
The generating function identity equivalent to Bijection 2 is

a,(n)q"= E q[ll;~ll:+;~ (2.2)

,=o ~.i=0

Clearly (2.1) and (2.2) imply

1 ~ q89 z + ~,.7~
~oP(n)q "
d

(qt;qt)~ ;~.i =0
(2.3)
n=

~eZ'
We now give Bijection 2, and sketch Bijection 1.
Let 2 be a t-core. Define the vector n 42(~) in the following way. Label a cell
=

in the ith row and jth column of ~, by j - i m o d t. (This is called the t-residue
diagram in [13, p. 84].) We also label the cells in column 0 in the same way, and call
the resulting diagram the extended t-residue diagram. A cell is called exposed if it is
at the end of a row. Region r of the extended t-residue diagram of~ is the set of cells
(j,j) satisfying t(r - 1) < j - i < tr. We now define ni to be the m a x i m u m region of
2 which contains an exposed cell labeled i. Since column 0 contains infinitely m a n y
exposed cells, ni is well-defined.
If an exposed cell labeled i lies in region r, then there is an exposed cell labeled i
in each region < r . F o r example, in region r - 1, if i is not exposed, then the rim
h o o k whose northeast head is i in region r, and whose tail is i + ! in region r - 1
has length t. (Note: If i = t - 1, the rim hook will lie entirely in region r.) This
contradicts the assumption that 2 is a t-core.
N o t e that the size of the Durfee square of 2 is the sum of the positive ni's, since
this is the number of exposed cells in positive regions.
We next verify that n o + n 1 + . . . + n,-1 = 0. Let ~' be the conjugate
of ~. First we show that if 4 2 ( ~ ) = ( n o , nl . . . . , n t - 1 ) , then ~ 2 ( 2 ' ) =
( - n t - i, - n t - 2, " ' ", - n o ) . Let a cell labeled i be exposed in region r of 2, so that
i + 1 lies above i. In 2', the cell corresponding to i + 1 will be labeled t - i - 1, lie
in region 1 - r, and not be the last cell in its row. Thus t - i - ! is not exposed
in region 1 - r of z'. This shows that if i is exposed in region r of ~, r < ni; then
t - 1 - i is not exposed in region r of ~,', r > 1 - n~. A similar argument shows that
if a b o u n d a r y cell labeled t - i - 1 is not exposed in region 1 - r, then i is exposed
in region r. This verifies the claim about 42(2'). Since the size of the Durfee square is
unchanged by conjugation, the sum of the positive n~'s must be equal to minus the
sum of the negative n~'s; that is, n o + n t + . . . + n t_ 1 = O.
4 F. Garvan et al.

The inverse m a p to ~bz is easy to find: if (no, n~ . . . . . n,_ 1) is given, then the
exposed cells in each region are known. This in turn gives the lengths of each row of
the Ferrers diagram of ~, and thus ~..
Finally we show that if ~b2(~) = n, then ~ is a partition of the exponent in
Bijection 2. The n u m b e r of cells to the right of the main diagonal of the Ferrers
diagram of 2 is

ini + t(21 ) 9
nz>O

For the cells below the main diagonal, we take ~', to find

- ~ (t- l-j)nj+t
n~<O
There are

Z
nt>O
/'/i

cells on the main diagonal. The sum of these three terms is the exponent ofq in (2.3).
We quickly sketch q~l using the notation in Bijection 2. Given 2, we construct t
biinfinite words in the letters N and E, w o . . . . . wt- 1. T h e j t h letter o f w i is E ifi is
exposed in region j of 2, otherwise the j th letter is N (not exposed). If 2 is a t-core,
then each word wl is an infinite sequence of E's followed by an infinite sequence of
N's. In this case, 2 = ~, and 21 = ~ for all i. Otherwise there is an E to the right of
some N. Find the rightmost E, say in positionj. Find the rightmost N to the left of
this E, say in position k < j. Delete a rim hook in 2 whose northeast head is the cell
corresponding to E, and whose southwest tail is adjacent to the cell corresponding
to N. Place a part of sizej - k in 21. The new partition has N ' s in w~ from positionj
on, thus the E's have been pushed to the left. This operation can be done in any
order whatsoever, to finally obtain the t-core of 2, ~,. The parts of each 2~ a p p e a r in
increasing order.

3. Cranks

In this section we give the cranks (Theorem 2) and the cycles which give bijections
for these crank classes.
Let r be a residue class of t, and consider p(tn + r). O n the right side of (2.3), all
exponents of q which are equivalent to r mod t lie in the multisum. Since ft. 1 = 0,
we have 89IIn II2 ~_ 0 m o d t. Thus we have

p(tn + r)r "+r = - - 1 ~ q-~ ik~ll2 +r,.~ (3.1)

,=o (qt;qt)~n".~---
b modt
n-1 = 0
heY'
The symmetry groups of the quadratic forms in (3.1) have been c o m p u t e d by
c o m p u t e r for t < 8 in [1 l]. Until now, we have assumed that the class t and residue
r are arbitrary. We shall see that the choices of 5n + 4, 7n + 5, and 11 n + 6 simplify
(3.1).
Cranks and t-cores 5

For 5n + 4 we now change variables so that the symmetry group is obviously

the dihedral group D 5 of order 10. The five vectors ~ for ~b2(p(4)) are
(1) ~o = (1, - 1, o, o, o),
(2) ~ -- (0, 1, - 1, O, 0),
(3) ~2 = (o, o, 1, - 1, o),
(4) ~3 = (0, 0, 0, 1, -- 1), a n d
(5) ~4 = (1, 1, 0, - 1 , - 1 ) .
They form a non-planar pentagon whose center ~ = (2/5, 1/5, 0, - 1/5, - 2 / 5 ) .
We now change variables: write fi = Cr o + . . . + e4v4. A vector ~ e Z s satis-
fies h']" = 0 and b . h = 4 m o d 5 if, and only if,/7 = ~oVo + . . . + agV4 for some
~ e Z s such that ~'1" = 1.
It is easy to see that the quadratic form in (3.1) in terms of the vector ~ is
511&ll z - 5(=o~i + cq~2 + . . . + =4=o)- 1 .
Thus we obtain

as(5n + 4)q,+ l = ~ qOA;O,

n=o ~'i=l
~eZ'
w h e r e Q(~) = II ~112 - (eo~l + ~1~2 + . . . + ~4~0). C l e a r l y the d i h e d r a l g r o u p D5
is the a u t o m o r p h i s m group.
It is also clear that the 5-cycle which cyclically permutes the ei's has no fixed
points. Any fixed point fi must have all entries equal. Since ~.-1 = 1, this would
imply that the entries are not integral.
The construction of the quadratic form and the group are completely analog-
ous for 7n + 5 and l l n + 6. We collect these results in a theorem.

T h e o r e m 1. For (t, r) = (5, 4), (7, 5) or (11, 6),

p(tn + r)q " + 1 - 1 F qQ(~), (3.2)

.=o (q;q)~ ~.'~= 1
fieZ'
where Q(fi) =
l]]xl]2 --
,-1
Ei=O~iO~i+ 1 .

It should be noted that other identities for the generating functions of p(tn + r),
which yield R a m a n u j a n ' s congruences, have been found by others. The most
famous of these is R a m a n u j a n ' s [19, (17) p. 213] identity for p(5n + 4), which was
considered by H a r d y in agreement with M a c M a h o n as R a m a n u j a n ' s most beau-
tiful identity [19, p. xxxv]. R a m a n u j a n [19, (18) p. 213] also found an analog for
p(7n + 5). In fact for t = 5", 7" similar identities have been found by Watson [20].
Using the theory of m o d u l a r functions an analog for p(1 In + 6) was found by Fine
[8, (3.25)] and extended to higher powers of 11 by Atkin [4]. However, these other
identities are of a different nature than (3.2) and do not yield obvious cranks.
Further, Fine's identity for t = 11 and R a m a n u j a n ' s identity for t = 5 or t = 7 are
dissimilar so it is surprising that an identity like (3.2) holds for all three values t = 5,
7 and 11.
6 F. Garvan et al.

Finally we come to the cranks. An obvious statistic, which is increased by

1 m o d 5 each time the e's are cyclically permuted is ~ = 0 ie~. So in terms of the
basis the crank is clear, but in terms of the fi basis (given combinatorially in w the
crank is not clear.

Theorem 2. A crank statistic for partitions 2 of 5n + 4, 7n + 5, or 1 In + 6 is 9iven by

the following alyorithm.

(1) Find the t-core ~ of 2, (t = 5, 7, or 11) by Bijection 1.

(2) Find dp2(-2) = ~ by Bijection 2.
(3) Let crank(2) be the followin(r m o d t linear combination:

(t = 5) 4no + nl + n3 + 4n4 f o r 5n + 4,
(t = 7) 4no + 2nl + n2 + n4 + 2n5 + 4n6for 7n + 5,
(t = 11) 4n o + 9n~ + 5n 2 + 3n 3 + n 4 + n 6 + 3n 7 + 5n 8 + 9n 9 + 4n~ o for
lln+6.

The new crank is not equal to the rank or the old crank. One can verify this on
the 5 partitions of 9 which are 5-cores: 621,522, 51111, 33111, and 321111.

4. More cranks

T h e cranks in w depend only upon the t-core ~ of 2, and thus use Bijections 1 and 2.
In this section we give two simpler ways (Proposition 1 and T h e o r e m 3) of
computing crank(2), (one which is a polynomial in the parts of 2), thus avoiding
Bijections 1 and 2.
First, let rk(2) be the n u m b e r of cells in the t-residue diagram of 2 which are
labeled k rood t. Suppose that 2 is a t-core. A relation between the rk'S and the ni's
can be found by counting the cells in a row ending with i in region nl. It is
t--I
r k = ~, (n2/2 + niz(i >= k)) (4.1)
i=O
where
1/2 i f S is true,
x(S)= -1/2 if S is false .

It is easy to see that (4.1) implies

nk = rk -- rk + 1 , (4.2)
SO that the relation in (4.1) for r o for t-cores becomes
t--I
ro = (# -- r,r,+ l) .
i=0

T h e cranks in T h e o r e m 2 are defined by a linear combination of the ni's. Using

(4.2), we can find rood t equivalent linear combinations of the rk'S.
Cranks and t-cores 7

Proposition I. T h e f o l l o w i n g m o d t linear c o m b i n a t i o n s o f t h e t-residue n u m b e r s r k

are c r a n k statistics:

(1) r 1 + 2 r 2 - 2 r 3 - r 4 f o r 5n+4,
(2) 2rl + r2 + r3 - r 4 - r 5 - 2r 6 f o r 7n + 5,
(3) - 6r 1 - 4r z - 2r 3 - 2r 4 - r 5 + r 6 + 2r 7 + 2r 8 + 4r 9 + 6 r i o f o r l l n + 6 .
P r o o f W e a l r e a d y k n o w from T h e o r e m 2 that c e r t a i n l i n e a r c o m b i n a t i o n s give the
c r a n k for t-cores, e.g. (2, - 1, 1, - 2) for t = 5. If we m u l t i p l y this l i n e a r c o m b i n a -
tion by 3, we o b t a i n (1) for t-cores. W e reverse the sign to o b t a i n (2), a n d (3) is
i m m e d i a t e from T h e o r e m 2. N o w c o n s i d e r (1)-(3) as definitions of a c r a n k for a n y
p a r t i t i o n 2, n o t j u s t t-cores. R e m o v i n g a rim h o o k of l e n g t h t from 2 reduces each
r k by one. T h e a b o v e linear c o m b i n a t i o n s do n o t change, since the s u m of all
of the coefficients is zero. T h u s the c r a n k c a n be defined b y P r o p o s i t i o n 1 for all
p a r t i t i o n s 2. []

Next, we give a d e f i n i t i o n of a c r a n k based u p o n the parts of

2:21 ~ 2 2 ~ . . . 2> Am"

T h e o r e m 3. F o r t = 5, 7 or 11, let p , ( x ) = ( x - (t - 1)/2)'- 3 m o d t. A c r a n k s t a t i s t i c

f o r p a r t i t i o n s 2: 2, > 2 2 > . . . > 2,. o f tn + r ( w h e r e r = 4, 5, a n d 6, r e s p e c t i v e l y ) is
given b y

crank(A) = ~ (p,(2i - i) - p,(i - 1)) modt.

i=1
Proof Let al be the coefficient of r i in P r o p o s i t i o n 1. Recall that
al +a2+ ... +at-~ =0.
First we e v a l u a t e the c o n t r i b u t i o n to the c r a n k in P r o p o s i t i o n 1 from the first
row of 2. T h e t-residue d i a g r a m of 2 has a n x -= 2 a - 1 m o d t at the e n d of the first
row. T h e n the c o n t r i b u t i o n of this r o w to the c r a n k is

a l+a 2+ ... +a~. (4.3)

It r e m a i n s to find a p o l y n o m i a l whose values at x = 0, 1. . . . . t - 1 agree with (4.3)
m o d t. Here are such p o l y n o m i a l s :

(1) qs(x) = 3(ps(x ) - 4) for t = 5,

(2) qT(x) = - ( p T ( x ) - 4) for t = 7, a n d
(3) q ~ ( x ) = p ~ ( x ) - 4 for t = 11.
T h u s we have s h o w n t h a t the first r o w c o n t r i b u t e s q,(x) =- q,(2~ - 1) - qt(O) m o d t.
C o n s i d e r row i o f the t-residue d i a g r a m of 2: the first e n t r y is 1 - i m o d t, a n d
the final e n t r y is 21 - i m o d t. The c o n t r i b u t i o n of this r o w to the c r a n k is

aa + a 2 + . . . + a~,_ i + (at + 1 - i + 9 9 . + a t - 1) =--

a l + a2 + . . . + a ~, _ i -- (al + . . . +at-i) =- q,(2i - i) - qt(t - i) m ~ t.

Thus

crank(A) = ~ (q,(2i - i) - qt(t - i)) m o d t .

i=1
8 F. G a r v a n et al,

Since q t ( t - 1 - i) = q~(i) rood t, and q,(i) is a linear function of pt(i), the result
follows. []

5. Associated q-series

From (2.1) there is an explicit generating function for the partitions of n which are t-
cores. We investigate these functions for t = 5, 7, and 11, and give explicit formulas
for as(n ) and a7(n ) in Theorems 4 and 5.
The five cycle in w implies that as(5n + 4) - 0 mod 5. However, much more is
true,
as(5n + 4) = 5as(n ) . (5.1)
We give two proofs of (5.1): one which is analytic, and proves more (Theorem 4),
and another which is a bijection.
Theorem 4. L e t n + 1 = 5CP~ 1 . 9 9 es'a~'~b'~l . . . qb, b e t h e p r i m e f a c t o r i z a t i o n of n + 1
into primes Pi = 1, 4 mod 5, qj ~ 2, 3 mod 5. T h e n
and
p~,+l _ 1 bi+l
as(n) = Y 1!I ~i2_~ 1LI qj + ( - 1 ) b~
i=i j=l qj +
Proof Clearly (2.1) implies
(qS;qS)~
as(n)q" =
,=o (q;q)oo
However an identity of Ramanujan ([2, (3.46)] or [6]) is
(q,;qS)~
q (q;q)~ -,=o g g-r
where ( b ) is the Legendre symbol. Clearly the right side of (5.2) is

n=l

so

Since (~)/d is multiplicative, (5.3) and [12, Theorem 265] implies that a s ( n - 1) is
multiplicative. It remains to evaluate a s ( p " - 1) for primes p, which can be done by
(5.3). []
The combinatorial proof of (5.1) is explicit: given a partition ~. of n which is a 5-
core, find five partitions 0o, 01, 02, 03, 04, of 5n + 4 which are also 5-cores. Let
q52(~.) = ff~. We now give the image under 4~2 of the five partitions of 5n + 4.
(1) ~bz(0o)=(m l + 2 m 2 + 2 m 4 + l , -m 1-mE+m3+m4+l,2ma+m2
+2m3, - 2 m 2 - 2 m 3 - m 4 - 1 , -2ml-m3-2m4-1),
Cranks and t-cores 9

(2) q52(01) = (2m 1 + 2m4 + 1 + m3, -- 1 -- m 2 -- 2m 3 -- 2m4,

- - 2 m l - - 2 m 2 - - m 4 , mt + 2 m z + 2 m 3 , --rex + m 2 + m 4 - - m 3 ) ,
(3) q~2(02) = (ml -- m2 -- m4 + m3, 2m2 + m4 + 1 + 2m3, -- m 1 -- 2m,
- - 1 - - 2 m 3, - - 2 m 1 - 2 m 2 - m a , 2 m l + m 2 + 2 m * ) ,
(4) ~b2(03) = ( - 2ml - m2 - 2m4, - ms - 2m2 - 2m3, 2m2 + 2m~ + 1 + m3,
ml + m E - - m 4 - - 1--m3,2ml +2m3+m4),
(5) ~b2(04) = ( - 2m 1 - m 4 - 2m3, 2ml + 2m2 + m3, ml - m2 + m4 - m3,
m2+2m4+ 1 +2m3, -m 1-2mz-2m4-1).
We delete the verification of this claim. In fact, these five partitions Oi form an
orbit under the five cycle, and
crank(Oi) - 3i rood 5 .
We denote by y, the 5 - to - 1 m a p from the set of 5-cores of 5n + 4 to the set of
5-cores of n, y,(0i) = ~. for 0 < i < 4.
F o r t = 7, we have a7(7n + 5) -- 0 mod 7. It is not true that aT(7n + 5) = 7aT(n ).
There are however two multiplicative functions here. The identity which is analog-
ous to (5.2) is due to Ramanujan [8, p. 159]

8q2(qT;qT)~+q(qV;qT)a~(q;q'3=(q;q)~ ,=1~'(7) q"(l(]-

~_q,~g+q") . (5.4)

Let a'(n), b(n), c(n) denote the coefficient of q" in the expansion of each of the
three functions in (5.4) so that
8aT(n) = a'(n + 2),
a'(n) + b(n) = c(n).
We will give in Lemmas 1 and 2 explicit formulas for b(n) and c(n), resulting in an
explicit formula for aT(n) in Theorem 5.
The proof of Theorem 4 analogously establishes the multiplicativity of c(n). We
delete the details. The resulting explicit formula is given in L e m m a 1.
Lemma 1. I f n . .7Cp"11
. . Ps,,sqlbi 9 9 9 qb, is the prime factorization of n into primes
Pl - l, 2, 4 m o d 7 , and qj --- 3, 5, 6 m o d 7 , then
q2bj+E+(__l)bJ
c(n) 72c 15 p2a,+2__l h
''
i=1 p l2 = i- j "="l q~ + 1

T o establish the multiplicativity of the sequence b(n), we apply the theory of

modular forms.
Proposition 2. The sequence b(n) is multiplicative, moreover

b(pm) = b(p)b(pm-1) - ( P ) p 2 b ( p m- 2)

for any prime p and integer m > 2.

10 F. Garvan et al.

Proof I f q = e 2~iz, the generating function B(z) for b(n) (see (5.4)) can be considered
a function of z in the upper half plane. In terms of the classical q-function it is
B(z) = (rl(z)q(7z)) 3, which is a cusp form of level 7, weight 3, and character
z(n) = (~) [14, Prob. 12, 13, p. 145]. Since this space of cusp forms is one-
dimensional, B(z) must be an eigenform for the Hecke operators T,. This proves the
multiplicativity of b(n), and the three term recurrence follows from the Euler
product for the associated Dirichlet series [14, p. 160 (5.11)]. []
To find an explicit formula for b(n), we need only find b(p) for all primes p. The
three term recurrence in Proposition 2 gives b(pm), and multiplicativity gives b(n)
for all n. The next proposition gives the values of b(p).
Proposition 3. Suppose p is an odd prime, l f p --- 3, 5 or 6 m o d 7 , then b(p) = O. I f
p - 1, 2 or 4 m o d 7, let p = x 2 + 7y 2, where x and y are positive integers. Then
b(p) = 2(x 2 - 7y2).
Remark. The result also holds for p = 2 except in this case x = y = 89
Proof F r o m Jacobi's identity for (q;q)3 1-12, T h m 357] we find that
b(p) = ~ ( - 1)m+"(2m + 1)(2n + 1).
m,n>O
( 2 m + 1) 2 + 7 ( 2 n + 1) 2 = 8 p

We need to determine all of the solutions (m, n) to

8p = (2m + 1)2 + 7(2n + 1) 2 , (pl)
where m, n > 0. Clearly the right side is a quadratic residue m o d 7, so ifp = 3, 5 or 6
m o d 7 , there are no solutions (m, n) to (pl) and b(p) = O.
Thus we can assume that p =- 1, 2 or 4 m o d 7. We first show in this case that p
has a unique representation p = x 2 + 7y 2 for positive integers x, y. We work in

K = Q(~/- 7), whose ring of algebraic integers, (gK = Y [1 + x/~--

2 7 1 , is a unique

factorization domain [12, Th. 246]. Since

1 =(P)=(~-) (by quadratic reciprocity),

p divides a 2 + 7 for some integer a. Since p does not divide a 4. ~ 7 - , p is not

prime. Thus p splits
p = tiff,
where fl = x + Y x / - 7 is prime in (ilk and x, y > 0. Now, x, y e Z , otherwise
x=m+ 89 y=n+ 89 nEZ),p=m(m+l)+7n(n4-1)4-2whichis
even and contradicts the assumption that p was odd. Hence p = tiff = x 2 + 7y 2 for
some positive integers x, y. The uniqueness of the representation follows from
unique factorization, and the only units being 4- 1.
We next show that (pl) always has two solutions (m, n), whose contribution to
b(p) is 2(x 2 - 7yZ). (pl) is equivalent to 2p = 79, where
(2m 4- 1) + (2n 4- 1 ) ~ _ xl 4- y l x / - 7
-- , Xl, Yl > 0 .
7= 2 2
Cranks and t-cores 11

Let

2 = 2 = ~ "

It follows t h a t 2p factors as

2p = (7~fl)(ff/~) = (~/~)(fffl) = y~7.

By u n i q u e f a c t o r i z a t i o n we m u s t h a v e ? = + ~fl, + ~fl, + ~/~, or 7 = + ~/~ This
a l l o w s eight p o s s i b l e solutions (Xl, Yl), six of which violate x l , Yl > 0. T h e two
s o l u t i o n s can be explicitly given, d e p e n d i n g u p o n which of the three inequalities
0 = y < 7y < x, 0 < x < y < 7y, or 0 < y < x < 7y occur. In each of these three
cases we find b(p) = 2 ( x 2 - 7 f ) as required. []
F o r p = l, 2 or 4 m o d 7 note t h a t P r o p o s i t i o n 3 implies t h a t b(p) =# 0, since
b(p) -- 0 w o u l d i m p l y t h a t x 2 = 7y z, a n d thus p = 1 4 f -- 0 m o d 7. W e also n o t e
t h a t b(p) = 2(x 2 - 7y 2) = fiE +/~2 where p = fl/~, fl = x + y x f - 7 a n d x, y > 0.
U s i n g the three t e r m r e c u r r e n c e in P r o p o s i t i o n 2 a n d the value of b(p) in
P r o p o s i t i o n 3, the value of b(p") is easily found. It is given in the next l e m m a .

L e m m a 2. L e t n = 7c/x 1 . . . ps"~qlbl . . . q~t be the prime factorization o f n into primes

Pi = l, 2, 4 m o d 7, and qj - 3,5,6 m o d 7. L e t Pi = x2 + 7yE f or some positive integers
xl, Yi in the case when Pi is odd and x 1 = Yl = 89 the case Pl = 2. Then

b(n) = 0 if some bj is o d d ,

b(n)= ( - 7) c I~I fl/2.,+2 _ fli2.,+z(-Iqb,otherwise,

where fli = Xi + YiJ -7"

N o t e t h a t L e m m a 2 implies t h a t b ( p ' ) = 0 if, a n d o n l y if, p - 3, 5 or 6 m o d 7

a n d m is odd. W e can n o w state the explicit f o r m u l a for aT(n).

T h e o r e m 5. L e t n + 2 = 7Cp~'. . . Ps,s ql~, . . . q~' be the prime factorization o f n + 2

into primes Pi = l, 2, 4 m o d 7, and qi - 3, 5, 6 m o d 7, then

aT(n) = ~(c(n + 2) - b(n + 2 ) ) .

where c(n) and b(n) are given in L e m m a s 1 and 2 respectively.

W e state r e l a t i o n s which are c o r o l l a r i e s of T h e o r e m s 4 a n d 5.

C o r o l l a r y 1. The f o l l o w i n g relations hold:

(1) as(5~m - 1) = 5~as(m - 1) -- 0 m o d 5 ~ ,
(2) aT(7~m - 2) -= 0 m o d 7 ~ ,
(3) aT(49n + 19) = 49aT(7n + 1),
(4) aT(49n + 33) = 49av(7n + 3 ) ,
(5) av(49n + 40) = 49aT(7n + 4).
P r o o f (1)-(5) are i m m e d i a t e f r o m T h e o r e m s 4 a n d 5. []
12 F. Garvan et al.

F o r a11(n ) there are identities for cusp forms which prove

a11(1 l ' m - 5) = 0 m o d 11 ~ ~5.5)
and
all(43n - 5) = 0 m o d 4 4 for (n, 43) = 1 . (5.6)
We delete the proofs of (5.5) and (5.6).

6. A crank for p(25n + 24)

We can define a group G which acts on the set of partitions oftn + r: G = G 1 x St,
where G x is the a u t o m o r p h i s m group of the quadratic form in Bijection 2, and St is
the symmetric group on t letters. F o r example, for 5n + 4, G -- D 5 x $5, where D 5
is the dihedral group of order 10. We have concentrated on the action of G1 for
congruences. We can also use St for congruences, and do so in this section to find a
crank for
p(25n + 24) = 0 m o d 2 5 . (6.1)
The group G -- D 5 x $5 for 5n + 4 contains two c o m m u t i n g five cycles which
generate a subgroup of order 25. If each five cycle has no fixed points, then G will
have no fixed points, and the orbits of G will have size 25. We have seen that the five
cycle inside D 5 has no fixed points. Since the five cycle in $5 could have fixed points,
we consider two cases to define the crank, which will be an ordered pair of integers,
each between 0 and 4. This ordered pair could be considered as the base 5
expansion for a n u m b e r between 0 and 24.
Theorem 6. A crank statistic for partitions 2 of 25n + 24 is given by the following
algorithm.
(1) Find ~b1(2) = (~., 20, 21,2z, 23, 24) given by Bijection 1.
(2) Let crank(2) be the following ordered pair of integers:
Suppose that 20 = 2A = 22 = 23 = 24, so that ~ is a partition of 25s + 24 for
some s. Let #1 = 75s+4(2) and/~2 = 7s(Pl) be partitions of Ss + 4 and s respectively
(see w Put
crank(2) = (crank(lq ), crank(p2)). (i)
Suppose that some 2 i 4: 2~, so that the five cyclic permutations of(20, 21, 22, 23,
24) are distinct. Order these five tuples in lex order, and define the crank of the ith
tuple to be i - 1. Put
crank(2) = (crank(2), crank((2 o, 21, 22, 23, 24))). (ii)
Proof. The vectors Vn(0i) = ~. in w form an orbit under C5 _-< G1, so that the cranks
of these five partitions are {0, 1, 2, 3, 4}.
The second part of the definition similarly interprets C 5 x C5, as a s u b g r o u p
of G. []
C r a n k s a n d t-cores 13

In Theorem 6(2)(ii) the lex order crank can be replaced by the following
statistic. Find ~b1(2), and if

12ol + I;~11 + 1221 + 1231 + 1241 - 0 m o d 5 ,

define the crank as in Theorem 6(2)(i). Otherwise define
crank(2) = (crank(~2), 1211 + 21221-4- 31231 + 41241 m o d 5 ) .
It is clear that the five cyclic permutations of the 21 will give five distinct cranks for
the second component.

7. S e l f - c o n j u g a t e partitions

In this section we consider the restriction of Bijections I and 2 to self-conjugate

partitions. Let asct(n) be the n u m b e r of t-cores which a r e self-conjugate.
In Bijection 1, suppose that 2 is self-conjugate. Then 2 must be self-conjugate,
and 21 = 2 t - i - 1, 0 < i < t - 1. Since ( - q ; q2)o~ is the generating function for self-
conjugate partitions, the self-conjugate version of (2.1) is
0o
1
( -q;q2)oo - (q2t;q2t)t/2 ~ asct(n)q" for t e v e n , (7.1a)
n=O

and
( • q• • q 2 • ) •

( __ q;q2)~ = (q-Zt~,,qZt)~- 1)~/2 n = 0 asc,(n)q" for t odd . (7.1b)

While there are congruences for qo(n), the n u m b e r of self-conjugate partitions of

n, e.g. qo(125n + 99) -= 0 m o d 5, it is not obvious how to prove these congruences
from this approach. Nevertheless, for t = 5 there is a nice L a m b e r t series, and a
theorem analogous to Theorems 4 and 5.
The L a m b e r t series is
(q10;qlO)2 q"
q(_q;q2)~(_qS;qlO)~ L Z2o(n) -l - q " '
- .=1 (7.2)

where (
Z2o(n) = ) 0 1 if (n, 10) > 1,
[
( _ 1)yr,- 1) otherwise.

Thus only odd n occur in (7.2), and the appropriate signs for odd n m o d 20 are
+ - 0 - + - + 0 + - . (7.2) follows from the 61//6 summation formula 1-2, (3.1)].
The identity (7.2) also follows from an identity of R a m a n u j a n [21, (1) p. 60] and
Jacobi's formula for r2(n ), the n u m b e r of representations of n as a sum of two
squares 1-12, T h m 278]. The functions r2(n ) and ascs(n) are related by

r2(n) - r2 ~ = 4ascs(n - 1) . (7.3)

As before we obtain from (7.2) the following theorem.

14 F. Garvan et al.

Theorem 7. L e t n + 1 = 2C5dp~1 9 9 . ~'s"a~"b~ . . . qb, be the prime factorization o f n + 1

into primes Pl -= 1 mod 4, and q~ = 3 mod4. Then

ascs(n) = ~ (a, + 1) ]ZI (1 + ( -1)b,)/2.

i=1 j=l

A surprising corollary results.

Corollary 2. The followin9 relations hold:

(1) ascs(2n + 1) = ascs(n),
(2) ascs(5n + 4) = ascs(n),
(3) ascs(n) = 0 if and only if there exists a prime q - 3 m o d 4 and an odd inteyer
b such that qb e x a c t l y divides n + 1.

There is a nice combinatorial proof of Corollary 2(2). Suppose ~. is a self-

conjugate partition of 5n + 4 which is a 5-core. It is easy to see that only 0 o ofw is
self-conjugate. Thus, the restriction of the map ~, to self-conjugate partitions is a
bijection.
A similar bijection exists for Corollary 2(1). Just replace (nl, n2) in (7.4) by
( - n l + n2, - n l - n2 - 1). We do not have a combinatorial proof of Corollary
2(3).
In Bijection 2, clearly we have ng = - n t - 1 - ~ , so the generating function
becomes

asct(n)q"= ~ q tlrnll2+c'~ (7.4)

n= 0 ~Zft/2]

where
J(2, 4 . . . . . t - 1) for t odd
((1,3, ,t 1) forteven,
and It/2] is the integral part of t/2. Using (7.4) dissections of ( _q;q2)~ due to
Kolberg [16] can be easily derived.

8. Partitions with distinct parts

A bijection analogous to Bijection 1 for partitions with distinct parts has been
given by Morris and Yaseen 1-17]. In this section we give the appropriate version of
Bijections 1 and 2 for these partitions.
We begin with notation. Let pd(n) be the number of partitions 2 of n with
distinct parts. Given such a 2, the shifted Ferrers diagram of 2, S(2), is the Ferrers
diagram of 2 with the ith row shifted to the right by i - 1 cells. The doubled
partition of 2, denoted 22, is defined by adding 2g cells to the i - 1st column of S(2).
For example, if 2 = 31, then 22 = 431. We let D D denote the set of doubled distinct
partitions 22. We define the t-core of 22 as the usual t-core of 22. Let DDt-core be
the set of partitions which are t-cores of doubled distinct partitions, and let add,(n)
Cranks and t-cores 15

be the number of these which are partitions of n. We state without proof that
DD~_ ~o~ ~ DD, thus addt(n) = 0 if n is odd.
By restricting Bijection 1, we find a bijection for doubled distinct partitions. We
must take the conjugate of the 2o in ~bx to find 20 for ~b3.

Bijection 3. There is a bOection ~3: DD ~ DDt_core x DD x P x . . . x P,

~ l (~2) = (2~, 20 ' 21 2~ -- 1 ) '

t-1 ~ t
such that 2 1 2 1 = 1 2 2 1 = 1 2 ~ . [ + t ~ i = o 1 2 i l , and2t-~ 21,for 1 <_i<_t-1.
The generating function identities equivalent to Bijection 3 are

n= 0
pd(n)q ~ " = (_q2~;qZt)~
(q2t;qZt)~- 1)/2 n = O
~ aad,(n)q" for t odd (8.1a)

and
( __qt;qt)~ ~ ~o
~, pd(n)q 2" = (q2t;~2t)~-2)/~ n=~Oaddt(n)q" for t e v e n . (8.1b)
n=O =

Bi]ection 4. There is a bijection ~)4: D D t - r {n = (no, n 1. . . . . nt-1): nleZ,

t~ 0 = 0 , n i -nt-i, 1 < i <_ t - 1},where
IA~l=tll~ll2/2 +b'n, b = ( O , 1. . . . . t-l).
The generating function identity equivalent to Bijection 4 is

add,(n)q" = Z qtll~l? + ~~ (8.2)

n = 0 ~ ~ ~[(t - 1 )/21

where
= ~'(1, 3 . . . . . t -- 2) for t o d d ,
/. (2, 4, , t 2) for t e v e n .

9. Remarks

Naturally it is desirable to prove the general congruence for powers of 5, 7, and 11

found in [4, 20]. It would be natural to assume that the symmetry group of the
quadratic form for t = 5"7bl 1c contains a cycle of the appropriate length. We have
not been able to find this cycle. Our crank for p(25n + 24) was inspired by Bailey's
([6], [2, p. 465]) proof of Ramanujan's congruence for p(5n + 4). In fact Bailey's
method also yields the congruence for p(25n + 24). It may be possible to extend our
methods to obtain a crank for p(49n + 47) and p(121n + 116). First we need to find
a crank for aT(49n + 47) and all(121n + 116). Our present method does not
succeed in these cases since there is no analog of (5.1). It may be easier to find a
mod 49 crank for p(49n + 19), p(49n + 33), p(49n + 40) in view of Corollary 1 (3),
(4), (5).
In [ l l ] the symmetry groups of the forms associated with p(tn + r) for t < 8
and all r were computed. The symmetry groups for p(1 In + r) for r =1= 6 need to be
16 F. Garvan et al.

found. There is reason to believe these groups are larger than C 2. It seems that an
unusually high proportion of the coefficients of al l(n) are divisible by 11 so there
may be more 11-cycles. A higher proportion of the coefficients are divisible by 5. In
fact, using modular forms which appear in an analog of (5.4) for t = 11, we can
prove the following theorem.

Theorem 8. If(fir) = 1 and n not divisible by 5, then a11(n - 5) = 0 m o d 5 .

We can prove by a bijection from (7.4) that

asc7(4n + 6) = asc7(n) . (9.1)

Further, by utilizing Gauss' result for sums of three squares [12, p. 311] it can be
shown that
ascT(n)=O if and only i f n + 2 = 4 " ( 8 m + 1). (9.2)
(This observation was made by D o u g McDoniel.)
Kolberg [15, 16] has given several infinite products for dissections of p(n), q(n)
and qo(n). M a n y of these results (for q(n) and qo(n)) follow easily from (7.4), (8.2) and
the Jacobi triple product formula.
John Stembridge informed the authors of the existence of [17, 18] for distinct
partitions. It should be noted that the definition of t-core given in w for t-even does
not agree with [17, 18].

References

1. Andrews, G.E.: The theory of partitions. Encyclopedia of Mathematics and Its Applications,
Vol. 2. Rota, G.-C. (ed.), Reading, MA: Addison-Wesley 1976 (reissued by Cambridge Univ.
Press, London and New York, 1985)
2. Andrews, G.E.: Applications of basic hypergeometric functions. SIAM Rev. 16, 441-484
(1975)
3. Andrews, G.E., Garvan, F.G.: Dyson's crank of a partition. Bull. Am. Math. Soc. 18, 167 171
(1988)
4. Atkin, A.O.L.: Proof of a conjecture of Ramanujan. Glasgow Math. J. 8, 14-32 (1967)
5. Atkin, A.O.L., Swinnerton-Dyer, P.: Some properties of partitions. Proc. Lond. Math. Soc.,
III. Ser. 4, 84 106 (1954)
6. Bailey, W.N.: A note on two of Ramanujan's formulae. Q. J. Math. Oxf. II. Ser. 3, 29-31 (1952)
7. Dyson, F.: Some guesses in the theory of partitions. Eureka (Cambridge), 8, 11~15 (1944)
8. Fine, N.: On a system of modular functions connected with the Ramanujan identities. Tohoku
Math. J. 8, 149 164 (1956)
9. Garvan, F.: New combinatorial interpretations of Ramanujan's partition congruences mod 5,
7 andl 1. Trans. Am. Math. Soc. 305, 47-77 (1988)
10. Garvan, F.: The crank of partitions mod 8, 9 and I0. Preprint
11. Garvan, F., Stanton, D.: Sieved partition functions and q-binomial coefficients. Math.
Comput. (to appear)
12. Hardy, G.H., Wright, E.M.: An Introduction to the Theory of Numbers. London: Oxford
Univ. Press 1979
13. James, G., Kerber, A.: The Representation Theory of the Symmetric Group. Reading, MA:
Addison-Wesley 1981
14. Koblitz, N.: Introduction to Elliptic Curves and Modular Forms. New York: Springer, 1984
Cranks and t-cores 17

15. Kolberg, O.: Some identities involving the partition function. Math. Scand. 5, 77-92 (1957)
16. Kolberg, O.: An elementary discussion of certain modular forms. Univ. Bergen Arb. naturv, r.
Nr. 19. (1959)
17. Morris, A., Yaseen, K.: Some combinatorial results for shifted Young diagrams. Math. Proc.
Camb. Philos. Soc., 99, 23-31 (1986)
18. Olsson, J.: Frobenius symbols for partitions and degrees of spin characters. Math. Scand. 61,
223-247 (1987)
19. Ramanujan, S.: Collected Papers of S. Ramanujan. London, New York: Cambridge Univ.
Press 1927 (reprinted by Chelsea, New York, 1962)
20. Watson, G.N.: Ramanujans Vermutung /iber Zerf/illungsanzahlen. J. Reine Angew. Math.
179, pp. 97-128 (1938)
21. Watson, G.N.: Proof of certain identities in combinatory analysis. J. Ind. Math. Soc. 20, 57-69
(1933)

Oblatum 16-IX-1989