0% found this document useful (0 votes)
51 views3 pages

Fun With Partitions: Alex Bols

This document discusses partitions of sets and paths between partitions. It makes 5 claims: 1) The sum of coefficients cP over all partitions P is 0 if n ≥ 2. 2) The sum of cP over partitions respecting a set I is 0 if |I| > 1, and 1 if |I| = 1. 3) The coefficients cP depend only on the number of parts in P. 4) For a graph G, the sum of cP over partitions cutting G is 0 if G is not connected, and relates to subsets of G if G is connected. 5) The coefficients cP are explicitly given by (−1)|P|-1(|

Uploaded by

Alex Bols
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
51 views3 pages

Fun With Partitions: Alex Bols

This document discusses partitions of sets and paths between partitions. It makes 5 claims: 1) The sum of coefficients cP over all partitions P is 0 if n ≥ 2. 2) The sum of cP over partitions respecting a set I is 0 if |I| > 1, and 1 if |I| = 1. 3) The coefficients cP depend only on the number of parts in P. 4) For a graph G, the sum of cP over partitions cutting G is 0 if G is not connected, and relates to subsets of G if G is connected. 5) The coefficients cP are explicitly given by (−1)|P|-1(|

Uploaded by

Alex Bols
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 3

Alex Bols

Fun with partitions


Denote by En = {1, , n} the set containing the first n integers and consider all partitions P of En .
Given two such partitions P and P write P P if P is a strictly finer partition then P, i.e. if for each p P
there is a p P such that p p and P 6= P .
Consider increasing paths P starting from the trivial partition P0 and ending in some other partition
P. Any such path ending in the partition Pn is given by a sequence
P = (P1 , , Pn )

(1)

where Pi1 Pi for all i = 1, , n. Call |P| the length of the path P (number of edges). Allow also the
empty path P0 which goes from P0 to P0 and has length zero.
For each partition P define a number
X
cP =
(1)|P| .
(2)
paths P to P

Because we allow the empty path, this also defines cP0 = 1.


I say that
Claim 1. If n 2, then
X

cP = 0.

(3)

For any collection I of subsets of En , we say that the partition P respects I if for each I I there is
a p P such that I p. we have
Claim 2.
X

cP = 0

(4)

c P = c P0 = 1

(5)

P respecting I

for any I such that |PI | > 1, and


X

P respecting I

otherwise.
Note that there is only one term in the sum if and only if I connects all sites in the sense that for any two
sites a, b there is a sequence {Ii }i=1, ,n in I such that a I1 , b Ik and Ii Ii+1 6= for all i = 1, , k 1.
This term corresponds to the trivial partition P0 .
Proof of claim 1 : Since n 2 there is at least one partition different from the full partition. So take
any partition P different from the full partition and consider a path P ending in that partition. By adding
the full partition to the end of this path we obtain a longer path P = (P, P) which also occurs in the
sum. But if one of these paths contributes +1 to the sum, then the other contributes 1 and vice versa, so
their contributions cancel. Since extending any two different paths not ending in the full partition in this
way yields two different paths ending in the full partition and any path ending in the full partition can be
obtained in this way, there is a pairwise cancellation of all terms and the sum vanishes.

Before proving claim 2 another result is needed:
Claim 3. The numbers cP depend only on |P|, uniformly in n. i.e. if |P| = |P | then cP = cP where P and
P may be partitions of sets of different sizes.
Proof : The paths contributing to cP can only pass partitions obtained by fusing components of P, so
the partitions which can be passed are clearly labelled by partitions of the partition P itself (considering the
components of P as the points for the new labelling partitions). It is then easy to see that cP = c{{1}, ,{|P|}
which proves the claim.

1

Proof of claim 2 : The collection I induces a partition on En . Indeed, say that a b if there is a
sequence {Ii }i=1, ,n in I such that a I1 , b Ik and Ii Ii+1 6= for all i = 1, , k 1. Then is an
equivalence relation and the equivalence classes collect those sites connected by I. Let C1 , , Ck be the
equivalence classes.
Any partition P contributing to the sum must respect the equivalence classes in the sense that for each
class Ci there must be a p P such that Ci p. But the partitions respecting the equivalence classes all
arise as partitions of the set of equivalence classes itself. The number of components of such a partition is
equal to the number of components of the corresponding partition of the set of equivalence classes and by
claim 3, the coefficients cP depend only on the number of components of P uniformly in n, therefore the
claim reduces to claim 1 with n = k.

Consider now graphs G on En and sum only over partitions P such that for any edge (a, b) of the graph
G, a and b belong to different parts of the partition. Call such a partition a partition cutting G.
Claim 4. If G does not connect all sites in En , then
X

cP = 0.

(6)

P cutting G

If G does connect all sites in En , then


X

cP =

P cutting G

(1)|I| .

(7)

IG
connecting En

Proof : The sum can be written as the difference of the sum over all partitions with the sum over all
partitions not cutting G, but the sum over all partitions vanishes by claim 1, so
X
X
cP .
(8)
cP =
P not cutting G

P cutting G

Now, I claim that


X

cP =

P not cutting G

(1)|I|+1

cP .

(9)

P respecting I

IE(G)

The set I is a set of edges on En and is therefore just a set of subsets of En , so it makes sense to speak
of partition respecting I. Say the partition P appears (once and only once) in the sum on the left hand
side because it doesnt cut edges in I E(G). This is just a different way of saying that P respects I, so P
appears (once and only once) in the right hand side in every sum demanding that the partition respect some
J I. Hence, in all, the partition appears
X

(1)|J|+1 =

JI

|I|
X

(1)j+1

j=1

|I|
j

=1

(10)

times. (The last equality is an identity which is obtained easily by considering the binomial expansion of
(1 1)n ). This proves (9).
The right hand side of equation (9) is a sum of terms of the form
X
cP
(11)
P respecting I

with I E(G). If E(G) does connect En then neither does I and the sum vanishes by claim 2, concluding
the proof of the first point of the claim.
If G does connect En then by claim 2, each connecting subgraph I of G contributes (1)|I|+1 cP0 =
(1)|I|+1 to the right hand side of (9), proving the claim.

The numbers cP are given explicitly by
2

Claim 5.
cP = (1)|P|1 (|P| 1)!

(12)

Proof : From claim 3 we already know that cP depends only on |P|, so put c|P| = cP . We have c1 = 1
which agrees with the claim. All other cn are now inductively determined by claim 1, which can be written
in the following way:
n1
X n
cn =
ck
(13)
k
k=1

where nk counts the number of ways in which n points can be divided into k groups, i.e. it counts the
number of partitions of En with k parts (Also known as the Stirling numbers of the second kind). So if we
can show that ck = (1)k1 (k 1)! satisfies (13) for all n, were done. i.e. we want to show that
n
X

k=1

(1)

k1

 
n
(k 1)! = 0.
k

(14)

By googling for a relation between Polylogarithms and Stirling numbers of the second kind, one sees that
this is in fact true.


You might also like