Euler's pentagonal number theorem

Tags: #theorem

Statement

The partition function satisfies the following recurrence relation:

p(n)=k=±1,±2,(1)k1p(nk(3k1)2)

for n1. (note it is 0 for n<0 and p(0)=1). Equivalently,

(n0p(n)xn)1=i1(1xi)=kZ(1)kxk(3k1)/2

Note that k(3k1)2 are the pentagonal numbers

Equivalence of two statements

First let us justify why the two statements are equivalent. Suppose we know that

n1(1xn)=kZ(1)kxk(3k1)/2

Then, multiplying this with the generating function for the partition function yields

1=(n0p(n)xn)(kZ(1)kxk(3k1)/2)

From this, we see that p(0)=1 and for any other term n, we have

p(n)=p(n1)p(n2)+p(n5)+p(n7)

With Jacobi's Triple Product

This is a special case of Jacobi's triple product formula with x=q^{3/2}, y { #2} = -q^{1/2}.

Franklin's bijective proof (1881)

Now let us proceed with the proof of the second equation in the statement.

(1x)(1x2)(1x3)(1x4)=1xx2+((1)21)x3+((1)21)x4+

where we note that the coefficient of each xn is the sum of the ways of partitioning n into distinct parts times the (1) to the power of the number of distinct parts. In other words,

=λn with distinct parts(1)number of parts of λ

which is the (number of partitions with distinct, even number of parts) - (number of partitions with distinct, odd number of parts).

We claim that this is equal to (1)k if n=k(3k1)2 for some kZ, otherwise it is 0. To prove this, we construct a partial involution on partitions with distinct parts that changes the parity of the number of parts.

Consider the Ferrers shape of a partition of n.

Define the involution τ as follows:

  1. if |A|<|B|, move all the dots in A to the bottom to make a new row.
    • Note that since it is strictly less, this creates a new, distinct part
  2. if |A||B|, move all the dots in B to create a new diagonal on the right of A
    • If the original parts were distinct, then adding one to each row will keep them distinct.

Now note that this is an involution. Let A be the diagonal after applying τ and let B be the last row after applying τ.

There is one exception, and that is when A,B have a common element and |A|=|B| or |A|=|B|1.

The first case happens when the partition is of the form λ=(2k1,2k2,,k) and so n=k(3k1)2. In the second case, we have λ=(2k,2k1,,k+1) and n=k(3k+1)2.

Putting this together (noting that the two cases have the same parity number of parts), the original sum becomes

1+k1(1)k(xk(3k1)/2+xk(3k+1)/2)

where we have the 1 for when the partition is just (1). We notice that k(3k+1)2=k(3k1)2 so

=kZ(1)kxk(3k1)/2

as desired.

Example

Suppose we have
20260316_131602.jpg|500
Then we see that we are in case I. Applying the operation, we have
Pasted image 20260317150359.png|400
and applying τ again, we have:
Pasted image 20260317150447.png|400

Here is another example, in which we are in case 2 first, and then 1:
20260316_132435.jpg|500