Cayley Tree 1

Download as pdf or txt
Download as pdf or txt
You are on page 1of 5

JOURNALOF COMBINATORIALTHEORY,Series A 71, 154-158 (1995)

Note

A New Proof of Cayley's Formula for


Counting Labeled Trees
PETER W. SHOR

A T & T Bell Laboratories, Room 2D-149, 600 Mountain Avenue,


Murray Hill, New Jersey 07974
Communicated by the Managing Editors
Received July 29, 1993

We give a new proof of Cayley's formula, which states that the number of labeled
trees on n nodes is n'-2. This proof uses a difficult combinatorial identity, and it
could equally well be regarded as a proof of this identity that uses Cayley's formula.
The proof proceeds by counting labeled rooted trees with n vertices and j improper
edges, where an improper edge is one whose endpoint closer to the root has a larger
label than some vertex in the subtree rooted on the edge. © 1995AcademicPress,Inc.

A w e l l - k n o w n t h e o r e m of C a y l e y gives the n u m b e r o f l a b e l e d trees on n


nodes. A n u m b e r of proofs of Cayley's f o r m u l a are k n o w n [ 5 ], a n d we a d d
a new one to the collection. This p r o o f uses a c o m b i n a t o r i a l identity which
is n o t easy to derive; the p r o o f c o u l d equally well be r e g a r d e d as a c o m -
b i n a t o r i a l p r o o f of this identity. A l t h o u g h the p r o o f as given b e l o w does
n o t p r o v e the identity in its full generality, it can be altered to d o so; we
discuss this in the s e c o n d section of this paper.

1. Tim PROOF

Cayley's formula states t h a t the n u m b e r of labeled trees on n nodes is n ~ - 2.


Since we can r o o t a tree at a n y of its nodes, this is trivially equivalent to the fact
t h a t the n u m b e r of labeled r o o t e d trees on n nodes is n " - 1. O u r starting p o i n t
is a c o m b i n a t o r i a l identity. Define a function Q(i, j ) = Q(i, L 0) as

Q(1, 0 ) = 1
Q(i, - 1 ) = 0 , i~>l,
Q( 1, j ) = 0, j~>l,
154
0097-3165/95 $12.00
Copyright © 1995 by Academic Press, Inc.
All rights of reproduction in any form reserved.
NOTE 155

and

Q ( i , j ) = ( i - 1) Q ( i - 1, j ) + ( i + j - 2 ) Q ( i - 1, j - 1), otherwise.

Then,
i--1
Q ( i , j ) = i i-1. (1)
j=0

This is a specialization of an identity due to Meir [-3, p. 259], with the


notational change Sz(m, n) = Q(m + 1, n, z - 1); the full identity is given in
the next section. Two proofs of this identity can be found in [ 3 ]. Given this
identity, all we need for a p r o o f of Cayley's formula is a classification of
labeled rooted trees that partitions the trees on i nodes into classes, where
the j t h class contains Q(i, j) elements. Such a classification is given in
Theorem 1.
We assume that the nodes of a tree are labeled 1, 2, ..., n. We call the
node labeled i the node i. For any edge of the tree, we call the node closer
to the root the parent node of the edge, and the node farther from the root
the child node of the edge. For a node x we let fl(x) be the smallest label
on any node in the subtree rooted at x (it is possible that/~(x) = x). We call
a tree edge e a proper edge if the label on the parent node of e is smaller
than the label on all descendant nodes of e, i.e., if fl(child(e))> parent(e).
Otherwise, we will call the edge improper. We can now state the main
theorem of this section.

THEOREM 1. The number of labeled rooted trees on n nodes with j


improper edges is Q(n, j).
Mallows and Riordan have defined a similar quantity [4, 6]: they con-
sider the number of inversions in labeled rooted trees, where an inversion
is a pair of vertices one of which is an ancestor of the other and in which
the ancestor has the larger label. By weighting each improper edge by the
number of descendant vertices which "make" it improper, one obtains the
number of inversions.
Proof of Theorem 1. We give a mapping q5from trees on n nodes to trees
on n - 1 nodes which gives rise to the recurrence for Q(n, j). To find qS(T),
where T is a labeled tree, find the node n in T which has the largest label. If
this node is a leaf, remove it. If it is not a leaf, then from the children of n, say
xl, x2 ..... x d, choose the child x a with the largest fl(x~). Contract the edge
nxa (so that the former children of both n and x a are now siblings) and let the
resulting node have label x~. This is now a tree on n - 1 labeled nodes. Note
that in the first case (when n is a leaf) the number of improper edges is con-
served and in the second case this number is reduced by one.
156 NOTE

We now consider the inverse of the m a p ~b. First, if we have a tree R on


n -- 1 nodes, we can add node n as the child of any of the nodes of R and
obtain a tree T in which n is a leaf and ~b(T)=R. We obtain n - 1
preimages T = ~b-I(R) of every tree R on n - 1 nodes in this way. F o r these
preimages, T and R have the same number of improper edges.
Next, suppose we have a tree R on n - 1 nodes with j improper edges.
To find a T = ~b-l(R) with j + 1 improper edges, we can take any node x,
put node n in place of node x, and make node x a child of node n. We now
have to decide what to do with the children of x in R. In T, they must
become children of either n or x. Let b be the number of improper edges
out of x in R, and let these edges have child nodes x~, x 2 .... , xb, with
fl(X1) < fl(X2) "~ "'" ~ / ~ ( X b ) < X. In order to construct a T with ~b(T) = R we
must make all the proper children of x in R be children of x in T, and we
can choose an a, 0 ~<a ~<b, and make the nodes x~, x2 .... , xa children of n
in T and the nodes Xa+l, ..., xb children of x in T. It is easy to verify these
are the only ways of partitioning the children of x in R between x and n
so as to make fl(x) the largest fl value among the children of n in T. There
are b + 1 such ways; thus we have one preimage T for each improper edge out
of x and one additional preimage. Summing over all nodes x gives j + n - 1
preimages T = ~b-a(R), where T has one more improper edge than R.
We have shown that any tree R with n - 1 nodes and j improper edges
has n - 1 preimages ~b-l(R) with the same number, j, of improper edges,
and n + j - 1 preimages ~b-~(R) with j + 1 improper edges. Thus if Q'(n, j)
is the number of trees on n nodes with j improper edges, Q'(n, j) satisfies

Q'(n,j)=(n-1) Q'(n- l,j) +(n+ j - 2) Q'(n- l , j - 1 ) ,


which is the recurrence for Q. To show that Q = Q' and thus complete the
proof, we must show that the boundary conditions of Q are satisfied for
trees with n nodes and j improper edges; i.e., we verify that there is exactly
one labeled rooted tree on one node, which has no improper (or proper)
edges. |
The number of improper edges in a r a n d o m tree on n vertices can be
calculated using the generating function given in Ruehr's p r o o f [ 3 ] of the
identity (1). The expected number of improper edges in a r a n d o m tree is
n

and the variance of this quantity is

2n+1 (1 (l÷!/n÷,l 3e+


NOTE 157

2. T i m GENERAL FORM OF THE IDENTITY

The identity (1) given above is actually a special case ( k = 0 ) of the


following identity due to Meir [3, p. 259]. Define a function Q(i, j, k) as

Q(1, 0, k) = 1
Q(i, - 1, k) = O, i ~> 1,
Q(1,j,k)=O, j>~l,
and
Q(i, j, k) = ( i + k - 1) Q ( i - 1, j, k)
+ (i+j-2) Q ( i - 1, j - 1, k), otherwise.

Then,
i--1

Q(i, j, k) = (i + k) i- i. (2)
j-0

The previous section can be viewed as a proof of this identity for the case
k = 0. We now give a combinatorial proof of this identity for integers k ~> 1.
Since Q(i, j, k) is a polynomial in k, by interpolation this proves (2) for
all k.
We need the following generalization of Cayley's formula [ 1 ].

THEOREM 2 (Cayley). Given n labeled nodes of which k are designated as


roots, the number of forests of k rooted trees that can be formed on these
nodes is kn n-~ 1.

To be more specific, we have a set of n labeled nodes, of which k have


been designated as "roots." We are counting acyclic graphs with k com-
ponents, where each component contains exactly one root. We will assume
that the nodes 1.... , k are those designated as roots.
Theorem 2 is easily derived from Clarke's proof of Cayley's formula
[2, 5], which proves an equivalent statement by reverse induction on k
(keeping n fixed). Cayley's formula follows from the case k = 1.
We can label the edges in a rooted forest proper and improper, using the
same definition as in the proof of Theorem 1; note that all edges out of a
root are proper. By essentially the same proof as for Theorem 1, we get the
following.

THEOREM 3. The number of forests of k rooted trees on n labeled nodes


with j improper edges and with roots 1, ..., k is kQ(n - k, j, k).
15 8 NOTE

The p r o o f is b y i n d u c t i o n on n,_leaving k fixed. T h e factor of k arises in


the base case of the i n d u c t i o n , when there are k such forests on k + 1
nodes. T o g e t h e r , T h e o r e m 2 a n d T h e o r e m 3 give a p r o o f of the i d e n t i t y (2).
A r e m a r k a b l e fact a b o u t the function Q(i, j, k) is t h a t in a d d i t i o n to the
recurrence given above, it also satisfies the recurrence

O(i,j,k)=(k-j+l) Q(i-l,j,k+l)+(i+j-2) O(i-l,j-l,k+l). (3)

I d e n t i t y (2) follows directly from this recurrence, which can be p r o v e d b y


induction [3]. Although Clarke's proof of Cayley's formula provides a
combinatorial proof that
i--1 i--2
~, Q ( i , j , k ) = ( i + k ) ~, Q ( i - l , j , k + l ) ,
j=O j=O

we d o n o t k n o w a n y c o m b i n a t o r i a l i n t e r p r e t a t i o n of the recurrence (3); it


w o u l d be interesting to find one.

ACKNOWLEDGMENT

I thank the referee for the suggestion that Ruehr's generating function could be used
to compute the expectation and variance of the number of improper edges and for the
Refs. [4, 6] on inversions in labeled trees.

REFERENCES

1. A. CAYLZY,A theorem on trees, Quart. J. Math. 23 (1889), 376-378.


2. L. E. CLARKE,On Cayley's formula for counting trees, J. London Math. Soc. 33 (1958),
26-28.
3. M. S. KLANKN (Ed.), Problems and solutions, SIAM Rev. 21 (1979), 256-263, Problem
78-6.
4. C. L. MALLOWSAND J. RIORDAN,The inversion enumerator for labeled trees, Bull. Amer.
Math. Soc. 74 (1968), 92-94.
5. J. W. MooN, Various proofs of Cayley's formula for counting trees, in "A Seminar on
Graph Theory" (F. Harary, Ed.), pp. 70-78, Holt, Rinehart, & Winston, New York, 1967.
6. J. W. MooN, The expected number of inversions in a random tree, in "Proceedings,
Louisiana Conf. on Combinatorics, Graph Theory and Computing, Louisiana State Univ.,
Baton Rouge, 1970," pp. 375-382.

You might also like