Partitions and q-Series Identities, I 1 Denitions and Examples

A partition of a number n is = (1 , 2 , , r ) with 1 2 r > 0 with i Z+ and

|| =
i=1

i = n.

A strict partition is one in which we have strict inequalities 1 > 2 > > r > 0 We call () = r the length of and refer to the i as its parts. Notationally, we may sometimes write = (1m1 2m2 . . .), where mi is the (multiplicity) number of parts of that are equal to i.

EXAMPLE: = (8, 6, 6, 5, 1) = (11 51 62 81 ) is a partition of 26, with length () = 5. = (11, 7, 5, 3, 1) is a strict partition of 27 and has length 5.

For a given n we denote the set of all partitions of n by Pn and the set of all strict partitions by SPn . We let On and En denote the sets of partitions with all odd or even parts, respectively. If H is any set of positivie integers, we denote by H the set of partitions with parts from H, and we let H( d) be the set of partitions with parts from H and with no part appearing more than d times. EXAMPLE: If H is the set of positive odd integers, then H = O. If H is the set of all positive integers, then H( 1) = SP.

We let p(S, n) denote the number of partitions of n that belong to a subset S of P. EXAMPLES: p(4) = 5, p(O, 4) = 2, p(SP, 4) = 2. If H is the set of positive odd integers, then p(H, n) = p(O, n). If H is the set of all positive integers, then p(H( 1), n) = p(SP, n) and p(H, n) = p(P, n) = p(n).

If = (1 , 2 , , r ) is a partition of n, we dene its conjugate to be = (1 , 2 , , s ), where i is the number of parts of that are greater than or equal to i.

EXAMPLE: If = (8, 6, 6, 5, 1) then = (5, 4, 4, 4, 4, 3, 1, 1). If we graphically represent and its conjugate , we see that they are both partitions of n and one graph can be obtained by transposing the other. In our example here, we can write graphically as: and we see that the graphical representation of is actually the transpose of this:

Counting Partitions

Theorem 1: Let H be a set of positive integers and let f (q) =

p(H, n)q n , p(H( d), n)q n .

fd (q) =
n0

For |q| < 1 we have f (q) =

(1 q n )1 ,

and fd (q) =
nH

(1 + q n + + q dn ) =
nH

(1 q (d+1)n )(1 q n )1 .

Before moving to the proof, we mention that the second equality in the second identity is obtained from the formula for the sum of a nite geometric series.

Proof of Theorem 1: To prove the rst identity, write H = {h1 , h2 , h3 , . . .} to get (1 q n )1 =

(1 + q n + q 2n + q 3n + )

= (1 + q h1 + q 2h1 + ) (1 + q h2 + q 2h2 + ) (1 + q h3 + q 2h3 + ) =

q a1 h1 +a2 h2 +a3 h3 + .

Note that the exponent of q is the partition (ha1 ha2 ha3 ), and thus q N will appear 1 2 3 once for each partition of n into parts in H. This gives us (1 q)1 =
nH n0

p(H, n)q n .

The proof of the second identity is similar except that the innite geometric series is replaced by a nite geometric series. In that case we get

(1 + q n + q 2n + q 3n + q dn ) =
nH da1 0 da2 0 da3 0

q a1 h1 +a2 h2 +a3 h3 + p(H( d), n)q n .

We mention that we are manipulating these as formal power series and thus not considering matters of convergence, but one can in fact justify our manipulations analytically.

Corollary 2 (Euler): For any n, the number of odd partitions and the number of strict partitions are equal. Proof: By Theorem 1, we have

p(O, n)q =
n0

(1 q 2n1 )1 ,
n=1

and p(SP, n)q n =

(1 + q n ).
n=1

Using the identity (1 q 2n ) 1 (1 + q ) = = , n) (1 q 1 q 2n1 n=1 n=1 n=1

we get p(O, n)q n =

p(SP, n)q n

and thus p(O, n) = p(SP, n).

We can in fact generalize the above result as follows: Corollary 3: p(Z+ ( d), n) = p(Z+ , n), where Z+ is the set of positive d+1 d integers not divisible by d. Proof: We again use Theorem 1 to get p(Z+ ( d), n)q n =
n0

(1 q (d+1)n ) (1 q n ) n=1

= =
n0

(d+1) n

1 (1 q n )

p(Z+ , n)q n , d+1

giving us the desired result. This result says that the number of partitions of n with no part appearing more than d times is equal to the number of partitions made up of parts not divisible by d + 1. Letting d = 1 here gives the previous result.

Combinatorial Identities

In the previous section, we were interested in using well-known identities to obtain results related to counting partitions. In this section we do just the opposite, using partitions to prove combinatorial identities. We being with the following Theorem 4 (Euler): 1+
n1

tn q n 1 = 2 ) (1 q n ) (1 q)(1 q (1 tq n ) n1

Proof: As in Theorem 1, we compute 1 = (1 tq n ) n1 =

p(P, m, n)tm q n
m0 n0

t () q || tn
n P, ()=n

q ||

We now note that using the transpose of gives us q || =

q | | =

qn . (1 q)(1 q 2 ) (1 q n )

Thus, we have

1 = (1 tq n ) n1 =

tn
n P, ()=n

q || q | |
P,1 =n

tn
n

= 1+
n1

tn q n (1 q)(1 q 2 ) (1 q n )

and this gives us the desired result.

Theorem 5 (Euler): 1+
n1

tn q n(n+1)/2 = (1 + tq n ) (1 q)(1 q 2 ) (1 q n ) n1

Proof: We again proceed as in Theorem 1 to compute (1 t2 q 2n ) (1 + tq ) = (1 tq n ) n1 n1

=
m0 n0

p(SP, m, n)tm q n t () q ||
SP

= =
n

tn q ||
SP, ()=n

We now look at a strict partition graphically and count the boxes in a specic way. (Since I dont know how to create this picture with TeX, Ill describe it as best I can). Although this is just as easily described algebraically (for example, we can consider a strict partition = (1 , . . . , r ) to be a combination of the strict partition (r, r 1, . . . , 1) and the regular partition = (1 , . . . , s ) where s r and i = i r i), this gives a nice description of what is happening. A strict partition will have a staircase on the right, since, as we move down, each row has at least one block less than the previous row. We then shift the rst row to the right by 1, the second row to the right by 2, etc. We then split the partition vertically so that the last (rth) row has exactly one block to the left of the vertical split. Then the number of blocks to the left of the vertical split is 1 + 2 + + r = r(r + 1)/2 as we count from the bottom up. Note that to the left we still have a strict partition. We call the partition on the right and note that this partition is not necessarily strict, and that there is not necessarily any blocks in the last (rth) row. That is, P and () r.

10

Thus, we see that if is a strict partition, we can write || = n(n + 1)/2 + ||, where is now in P and has length less than or equal to n. This gives us =
n SP, ()=n

tn q || tn q n(n+1)/2 q ||
n P, ()n

= =
n

tn q n(n+1)/2
P, ()n

q || q | |
P, (1 )n

=
n

tn q n(n+1)/2 tn q n(n+1)/2 (
n

= = 1+

1 ) (1 q)(1 q 2 ) (1 q n )

n1

tn q n(n+1)/2 (1 q)(1 q 2 ) (1 q n )

11

We conclude with an application of Theorem 5, known as Jacobis triple product identity. Theorem 6 (Jacobi):

znqn =
n=

(1 q 2n+2 )(1 + zq 2n+1 )(1 + z 1 q 2n+1 )

Proof: To reduce clutter, we rst introduce some notation. We let

(a)n = (a; q)n = (1 a)(1 aq) (1 aq n1 ), (a) = (a; q) = (a)0 = 1. We may then dene (a)n = (a) /(aq n ) . Now, applying Theorem 5 to the second product gives us (1 + zq
n0 2n+1

(a; q)n ,

zmqm )= , (q 2 ; q 2 )m m0

and this can be written as 1 2; q2) (q z q

(q

2m+2

; q )

1 = 2 2 (q ; q )

z m q m (q 2m+2 ; q 2 ) ,
m=

since (q 2m+2 ; q 2 ) vanishes for negative m.

12

We have (1 + zq
n0 2n+1

)=

1 2 z m q m (q 2m+2 ; q 2 ) 2n+2 ) n0 (1 q m=

and thus we need to show

z q
m=

m m2

(1 q
neven

2m+n

)=

1 2 znqn . 1 q 2n+1 ) n0 (1 + z n=

We rewrite the LHS of this to get

z q
m=

m m2

(1 q
neven

2m+n

) =
m=

zmqm

(1 q 2m q n )
neven

=
m=

zmqm

(1 q 2m1 q n+1 )
neven

=
m=

zmqm

(1 q 2m1 q n+1 )
n0

We again apply Theorem 5 (with t = q 2m1 ) to get

z q
m=

m m2 n0

(1)n q n +2mn+n = (q 2 ; q 2 )n =

n0

(1)n z n q n (q 2 ; q 2 )n (q/z)n (q 2 ; q 2 )n

z m+n q (m+n) z q

m= m m2

n0

= =

1 (q/z; q 2 )

m=

zmqm
m=

1 2 znqn 1 q 2n+1 ) n0 (1 + z n=

and this yields the desired result.

13