Trees

March 2, 2010

Forests and Trees

A graph is said to be acyclic if it does not contain cycle. An acyclic graph is also called a forest. A connected acyclic graph is called a tree. A tree is said to be trivial if it contains a single vertex. Every nontrivial tree has at least two leaves, i.e., two vertices of degree 1. Any two vertices in a tree are connected by a unique path. If T is a tree, then |E(T )| = |V (T )| 1. If F is a forest, then |E(F )| = |V (F )| c(F ), where c(F ) is the number of connected components of F .

Rooted Trees
A tree T with a specied vertex x is called a rooted tree with the root x, denoted T (x). A tree with a root x is referred to an x-tree. A branching is a rooted tree with an orientation such that every vertex but the root has in-degree 1. A branching with a root x is referred to an x-branching. Let D be a digraph and x a vertex of D. Let X be the set of vertices reachable from x in D. Then there exists an x-branching T (x) in D with V (T ) = X.

Spanning Trees
A spanning subgraph H of a graph G is a subgraph such that V (H) = V (G). A spanning tree T of a connected graph G is a tree and is also a spanning subgraph of G. Every connected graph has a spanning tree. If T is a spanning tree of a graph G, then for any edge e E(G) E(T ), there exists a unique cycle Ce in T e. Such a cycle is called a fundamental cycle of G with respect to T . Proof. Let u, v be end-vertices of e. There are two internal-vertex disjoint paths between u and v in T e, i.e., uev and the unique path between u and v in T . The two paths form a cycle Ce in T e. Suppose Ce is a cycle other than Ce in T e. Then Ce Ce is a nontrivial even graph contained T . This is a contradiction for T contains no cycle. If T is a spanning tree of a graph G, then for any edge e E(T ), there exists a unique bond Be of G which is contained in T c e. Such a bond is called a fundamental bond of G with respect to T .

Proof. Remove the edge e from T to obtain two disjoint trees T1 and T2 . Then the edge set between the vertex sets V (T1 ) and V (T2 ) is a bond of G. Suppose Be is a bond of G other than Be and is contained in T c e. Then the symmetric dierence Be Be is an edge cut of G and is contained in T c . Clearly, E(T ) is contained in E(G) Be Be . Since T is connected, so is E(G) Be Be . This a contradict to that Be Be is an edge cut of G. Theorem 3.1. The collection of fundamental cycles of a connected graph G with respect to a spanning tree T forms basis of the cycle space of G. So the cycle space has dimension |E(G)| |V (G)| + 1. Proof. Let {Ce | e E(G T )} be the collection of fundamental cycles of G with respect to T . We rst show that {1Ce | e E(G T )} is linearly independent over F2 . Assume eE(GT ) ae 1Ce = 0 for some coecients ae in F2 . For a particular edge e0 E(G T ), we have e0 Ce0 and e0 Ce for all e E(G T ) such that e = e0 . So the left-hand side of eE(GT ) ae 1Ce = 0 is ae0 . Thus ae0 = 0. This proves the linear independence. To show that {1Ce | e E(G T )} spans the cycle space of G, given a cycle C of G. We claim that 1C = eE(CT ) 1Ce , i.e., 1C +
eE(CT c )

1Ce = 0.

(1)

It is clear that the left-hand side of (1) is zero on all edges of E(T c C). The left-hand side of (1) also cancels to zero on the edges of E(C T c ). Thus the left-hand side of (1) is the characteristic function of an even graph on T . Since T does not contain cycle, it follows that the left-hand side of (1) is zero on all edges of G. Theorem 3.2. The collection of fundamental bonds of a connected graph G with respect to a spanning tree T forms basis of the bond space of G. So the bond space has dimension |V (G)| 1. Proof. Let {Be | e E(T )} be the collection of fundamental bonds of G with respect to T . We rst show that {1Be | e E(T )} is linearly independent over F2 . Assume eE(T ) ae 1Ce = 0 for some coecients ae in F2 . For a particular edge e0 E(T ), we have e0 Be0 and e0 Be for all e E(T ) such that e = e0 . So the left-hand side of eE(T ) ae 1Be = 0 is ae0 . Thus ae0 = 0. This proves the linear independence. To see that {1Ce | e E(G T )} spans the bond space of G, given a bond B of G. We claim that 1B = eE(BT c ) 1Be , i.e., 1B +
eE(BT )

1Be = 0.

(2)

It is clear that the left-hand side of (2) is zero on all edges of E(T B). The left-hand side of (2) also cancels to zero on the edges of E(B T ). Thus the left-hand side of (2) is the characteristic function of an edge cut U on T c . Since T = G T c is connected, thus G U is also connected. This implies that U must be empty. So left-hand side of (2) is identically zero.

Flow Space and Tension Space of an Oriented Graph

Let G = (V, E) be a graph. Recall that an orientation of G is a multi-valued function : V E {0, 1, 1} such that (i) (v, e) = {1} if e is a loop at v and (ii) if e = uv is a non-loop at its end-vertices u, v then (u, e), (v, e) are single-valued and (u, e)(v, e) = 1. A graph with an orientation is called an oriented graph. The incidence matrix of an oriented graph (G, ) is a matrix M = [mve ], where mve = 0 if e is a loop at v and mve = (v, e) if e is a non-loop at v. Let (G, ) be an oriented graph throughout. A real-valued ow of (G, ) is a function f : E R such that the ow-in equals the ow-out at every vertex v, i.e., f (e) =
eE (v) eE + (v)

f (e),

(3)

where E (v) and E + (v) are the sets of edges whose arrows have heads and tails at v respectively. The set of all real-valued ows of (G, ) forms a subspace of the Euclidean RE , called the ow space of (G, ), denoted F (G, ). A real-valued tension of (G, ) is a function g : E R such that for any directed cycle (C, C ) of G, [, C ](e) g(e) = 0.
eE(C)

The set T (G, ) of all real-valued tensions of (G, ) forms a subspace of RE , called the tension space of (G, ). The inner product of RE is dened by f, g := f (e)g(e).
eE

Let Gi (i = 1, 2) be subgraphs of G [1 , 2 ] : E {0, 1, 1}, dened by 1 1 [1 , 2 ](e) := 0

with orientations i . The coupling of (G1 , 1 ) and (G2 , 2 ) is a function if e E(G1 ) E(G2 ) is at v and 1 (v, e) = 2 (v, e), if e E(G1 ) E(G2 ) is at v and 1 (v, e) = 2 (v, e), otherwise.

(4)

If we view i as the extended functions i : E {0, 1, 1} by setting i (v, e) = 0 for edges e E(Gi ) at their end-vertices v, then for any edge e at its end-vertex v, [1 , 2 ](e) = 1 (v, e) 2 (v, e). Let C be a cycle of G. A direction of C is an orientation C of C such that the digraph (C, C ) has a head and a tail at every vertex of C; and (C, C ) is called a directed cycle. The indicator function IC of C is the characteristic function of the edge set E(C). Proposition 4.1. Let (C, C ) be a directed cycle of an oriented graph (G, ). Then the coupling [, C ] is a ow of (G, ). Proof. For any vertex not in the cycle C, it is clear that (3) is satised. For a vertex v V (C), let e and e be edges of C at v such that C (v, e) = 1 and C (v, e ) = 1. It is routing to check [, C ](x)
xE (v) xE + (v)

[, C ](x) = 0

for the cases: (1) (v, e) = C (v, e), (v, e ) = C (v, e ); (2) (v, e) = C (v, e), (v, e ) = C (v, e ); (3) (v, e) = C (v, e), (v, e ) = C (v, e ); and (4) (v, e) = C (v, e), (v, e ) = C (v, e ). Let U = [X, X c ] be an edge cut of G. A direction of U is an orientation U on U such that every arc of the digraph (U, U ) has its tail in X and head in X c ; and (U, U ) is called a directed cut of G. Proposition 4.2. Let (U, U ) be a directed cut of an oriented graph (G, ). Then [, U ] is a tension of (G, ). Proof. For any edge e at its end-vertex v, we have [, C ](e) [, U ](e) = (v, e) C (v, e) (v, e) U (v, e) = [C , U ](e). To see that [, U ] is a tension, it suces to show that [, C ](e) [, U ](e) =
eE(C) eE(C)

[C , U ](e) =
eE(C)E(U )

[C , U ](e)

is zero. Since C is a cycle and U = [X, X c ] is a cut, the number of edges having agreed orientations with respect to C and U equals the number of edges having opposite orientations. Hence the above right-hand side is zero. Remark. The ows [, C ] and tensions [, U ] are orthogonal in the Euclidean space RE . 3

Theorem 4.3. Let F be a forest of an oriented graph (G, ). For each edge e E(F c ), let Ce be the fundamental cycle of G and e a direction of Ce such that orientations and e agree on e. Then the set of ows [, e ], where e E(F c ), forms a basis for the ow space F (G, ). Moreover, for each ow f F (G, ), f=
eE(F c )

f (e) [, e ].

Proof. We rst show that {[, e ] : e E(F c )} is linearly independent. In fact, set ae [, e ] = 0
eE(F c )

for some real numbers ae . Note that for a particular edge e0 E(F c ) and an arbitrary edge e E(F c ), we have [, e ](e0 ) = 1 if e = e0 and [, e ](e0 ) = 0 if e = e0 . Thus the above left-hand side at e0 is ae0 . Hence ae0 = 0. We proved linear independence. Let f be a ow of (G, ). We claim that the function g := f eE(F c ) f (e) [, e ] is identically zero. It is clear that g is a ow of (G, ), for it is a linear combinations of ows. Note that for e0 , e E(F c ), we have [, e ](e0 ) = 1 if e = e0 and [, e ](e0 ) = 0 if e = e0 . It follows that g(e0 ) = 0, i.e., g|E(F c ) 0. Since F is a forest, if F is not trivial (i.e. F contains some edge), then F has a leaf v incident with an edge e of F . For each edge e E(F c ) at a leaf v, we must have g(e) = 0, for g is conservative at v. Continue this procedure; we see that g|E(F ) 0. So g is identically zero. Theorem 4.4. Let F be a forest of an oriented graph (G, ). For each edge e E(F ), let Be be the fundamental bond of G and e a direction of Be such that orientations and e agree on e. Then the set of ows [, e ], where e E(F ), forms a basis for the tension space T (G, ). Moreover, for each tension g T (G, ), g=
eE(F )

g(e) [, e ].

Proof. We rst show that {[, e ] : e E(F c )} is linearly independent. In fact, set ae [, e ] = 0
eE(F )

for some real numbers ae . Note that for a particular edge e0 E(F c ) and an arbitrary edge e E(F c ), we have [, e ](e0 ) = 1 if e = e0 and [, e ](e0 ) = 0 if e = e0 . Thus the above left-hand side at e0 is ae0 . Hence ae0 = 0. We proved linear independence. Let g be a tension of (G, ). We claim that the function h := g eE(F ) g(e) [, e ] is identically zero. It is clear that h is a tension of (G, ), for it is a linear combinations of tensions. Note that for e0 , e E(F ), we have [, e ](e0 ) = 1 if e = e0 and [, e ](e0 ) = 0 if e = e0 . It follows that h(e0 ) = 0, i.e., h|E(F ) 0. Now for each edge x E(F c ), let x be the orientation of the fundamental cycle Cx such that the orientations and x agree on x. Since h|E(F ) 0, then by denition of tension, [, x ](e) h(e) = [, x ](x) h(x) = h(x)
eE(Cx )

is zero. Thus h|E(F c ) 0. Hence h is identically zero. Corollary 4.5. Let (G, ) be an orientated graph. Then the ow space F (G, ) and the tension space T (G, ) are orthogonal complements in RE , and dim F (G, ) = |E(G)| |V (G)| + c(G), dim T (G, ) = |V (G)| c(G).

Cayleys Formula

Theorem 5.1. The number of labeled trees on n vertices is nn2 . Proof. Recall that a labeled branching is an oriented rooted tree such that there is exactly one edge tail at each vertex, except the root. We show that the number of labeled branchings on n vertices is nn1 . Then Cayleys formula follows directly because each labeled tree gives rise to n labeled branchings, one for each choice of the root vertex. Note that each labeled branching on n vertices can be build up, one at a time to add an edge, starting with the empty graph on n labeled vertices. In order to end up with a branching, the subgraph constructed at each stage must be a branching forest (each of its component is a branching). Initially, this branching forest has n components, each consists of a isolated vertex. At each stage, we add a new edge joining a root of one branching to a vertex of another branching; the number of components decreases by one. If there are k components, the number of ways to add a new edge e = uv is n(k 1): each of the n vertices may be the vertex u, there are n choices for u; whereas v must be the root of a branching that does not contain the vertex u, there are k 1 choices for v. The total number of ways of constructing a branching on n vertices in this way is thus
n

n(k 1) = (n 1)! nn1 .

k=2

On the other hand, any individual branching on n vertices is constructed exactly (n 1)! times by this procedure, once for each of the order of n 1 added edges. It follows that the number of labeled branchings is nn1 . Proposition 5.2. Let G be a graph and e be a non-loop edge. Let t(G) denote the number of spanning trees of a labeled graph G. Then t(G) = t(G\e) + t(G/e). Proof. The spanning trees of G that do not contain the edge e are exactly the spanning trees of G\e. The spanning trees T that contain the edge e correspond to the spanning trees T /e of G/e.