13.11.2014 Views

proofs

proofs

proofs

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Identities versus bijections 211<br />

As experienced formal series manipulators we notice that the introduction<br />

of the new variable y yields<br />

∏<br />

(1 + yx k ) = ∑<br />

p d,m (n)x n y m ,<br />

k≥1<br />

n,m≥0<br />

where p d,m (n) counts the partitions of n into precisely m distinct summands.<br />

With y = −1 this yields<br />

∏<br />

− x<br />

k≥1(1 k ) = ∑ (E d (n) − O d (n))x n , (10)<br />

n≥0<br />

where E d (n) is the number of partitions of n into an even number of distinct<br />

parts, and O d (n) is the number of partitions into an odd number. And here<br />

is the punchline. Comparing (10) to Euler’s expansion in (8) we infer the<br />

beautiful result<br />

⎧<br />

1 for n =<br />

⎪⎨<br />

3j2 ±j<br />

2<br />

when j ≥ 0 is even,<br />

E d (n) − O d (n) = −1 for n = 3j2 ±j<br />

2<br />

when j ≥ 1 is odd,<br />

⎪⎩<br />

0 otherwise.<br />

An example for n = 10:<br />

10 = 9 + 1<br />

10 = 8 + 2<br />

10 = 7 + 3<br />

10 = 6 + 4<br />

10 = 4 + 3 + 2 + 1<br />

and<br />

10 = 10<br />

10 = 7 + 2 + 1<br />

10 = 6 + 3 + 1<br />

10 = 5 + 4 + 1<br />

10 = 5 + 3 + 2,<br />

so E d (10) = O d (10) = 5.<br />

This is, of course, just the beginning of a longer and still ongoing story. The<br />

theory of infinite products is replete with unexpected indentities, and with<br />

their bijective counterparts. The most famous examples are the so-called<br />

Rogers–Ramanujan identities, named after Leonard Rogers and Srinivasa<br />

Ramanujan, in which the number 5 plays a mysterious role:<br />

∏<br />

k≥1<br />

1<br />

(1 − x 5k−4 )(1 − x 5k−1 )<br />

= ∑ n≥0<br />

x n2<br />

(1 − x)(1 − x 2 ) · · · (1 − x n ) ,<br />

∏<br />

k≥1<br />

1<br />

(1 − x 5k−3 )(1 − x 5k−2 )<br />

= ∑ n≥0<br />

x n2 +n<br />

(1 − x)(1 − x 2 ) · · · (1 − x n ) .<br />

The reader is invited to translate them into the following partition identities<br />

first noted by Percy MacMahon:<br />

Srinivasa Ramanujan<br />

• Let f(n) be the number of partitions of n all of whose summands are<br />

of the form 5k + 1 or 5k + 4, and g(n) the number of partitions whose<br />

summands differ by at least 2. Then f(n) = g(n).<br />

• Let r(n) be the number of partitions of n all of whose summands are<br />

of the form 5k + 2 or 5k + 3, and s(n) the number of partitions whose<br />

parts differ by at least 2 and which do not contain 1. Then r(n) = s(n).<br />

All known formal series <strong>proofs</strong> of the Rogers–Ramanujan identities are<br />

quite involved, and for a long time bijection <strong>proofs</strong> of f(n) = g(n) and<br />

of r(n) = s(n) seemed elusive. Such <strong>proofs</strong> were eventually given 1981<br />

by Adriano Garsia and Stephen Milne. Their bijections are, however, very<br />

complicated — Book <strong>proofs</strong> are not yet in sight.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!