Академический Документы
Профессиональный Документы
Культура Документы
For photocopying of material in this volume, please pay a copying fee through the Copyright Clearance
Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to photocopy
is not required from the publisher.
ISBN 978-981-4566-03-2
Printed in Singapore
vii
viii Preface
larger number of spanning trees will have greater reliability for such p. In
the study of graph theory, most of the results regarding the number of
spanning trees have only been proven for simple graphs, so herein, we
investigate the problem for the extended class of multigraphs. It should be
noted that, while extensions to multigraphs make the optimal solution
readily apparent in many problems, it is not the case for the spanning tree
problem.
ix
x Contents
An Introduction to Relevant
Graph Theory and Matrix Theory
Though we assume that the reader of this work is well versed in Graph
Theory, in this section we provide the primary graph theoretic definitions
and operations that are used in the body of the succeeding chapters. For
any other terminology and notation not provided here we refer the reader
to Chartrand, Lesniak and Zhang [Chartrand, 2011].
A multigraph is a pair M = (V , m) , where m is a nonnegative integer-
valued function defined on the collection of all two-element subsets
of V, denoted V(2) . In the case there are several multigraphs under
1
2 Spanning Tree Results for Graphs and Multigraphs
m ({u, v} ) ∈{0,1} , then M is a graph, and m−1 ({1}) = E is called the edge
set. In this case, we use G instead of M and employ the alternate notation
G = (V , E ) .
Remark 0.1 The set of multigraphs includes the set of graphs. Thus any
result that is stated for multigraphs also holds for graphs. On the other
hand, results that are stated for graphs are only applicable to graphs and
do not hold for multigraphs.
M G
Edges are incident at its end-nodes, and two nodes which have an
edge between them are called adjacent. Node u has degree deg(u) equal
to the number of edges having that node as an end-node, i.e.
deg(u ) = ∑ m ({u, v}) .
v ≠u
deg ( v1 ) and deg ( v2 ) . Thus when summing the degrees of the nodes of
M, each edge is counted twice. □
( )
aij where aij = m {vi , v j } . Note if A is the adjacency matrix of M,
n
then ∑a
j =1
ij = deg ( vi ) .
0 1 0 0 2
1 0 1 1 1
A = 0 1 0 3 1 .
0 1 3 0 0
2 1 1 0 0
M T
nonnegative integer-valued function defined on V V u , u | u V .
alternate notation D = (V , A) .
As in the case of multigraphs, directed multigraphs have geometric
representations as well. Each element of V is depicted by a point, and
two points u and v are joined by m ( ( u , v ) ) directed curves, directed from
outdeg(u ) = ∑ m ( ( u, v ) ) .
v ≠u
M D
Figure 0.5
10 Spanning Tree Results for Graphs and Multigraphs
K4 K5
v1 v2
v4 v3
Figure 0.7: A clique inside a graph. The clique edges are thicker.
An Introduction to Relevant Graph Theory and Matrix Theory 11
bipartite graph by K p,q . When p =1, we refer to the graph K1, q as a star.
graph is regular, then each part has the same order. If Vi = ni for
Figure 0.8 depicts the complete bipartite graph K 2,3 , the star K1,4 and
the complete 3-partite graph K 2,2,3 .
( i ± n ) mod n
j for each j, 1 ≤ j ≤ k . The n1 , n2 ,..., n k denote the “jump”
0 0
7 1 7 1
6 2 6 2
5 3 5 3
4 4
graph (V ,V(2) − E ) , i.e. G has the same node set as G and uv is an edge
u
x
v
a) A multigraph M
b) c)
Figure 0.11
An Introduction to Relevant Graph Theory and Matrix Theory 15
n
k ≤ . A special case of edge deletion that we will encounter is K n
2
minus a matching, denoted K n − kK 2 . In Figure 0.12 we depict
K5 − 2K2 . Note, the two edges deleted are arbitrary as long as they are
not incident on a node, as all such graphs are isomorphic.
Figure 0.12: K 5 − 2 K 2 . The deleted edges are indicated by the dotted edges.
v
a) A multigraph M
u u
w w
v v
b) M +2{u ,v} c) M +{u , w}
Figure 0.13
Figure 0.14: K 5+ x .
An Introduction to Relevant Graph Theory and Matrix Theory 17
a) A multigraph M
b) M | x c) M | y
v1
v3
a)
u1 v1
u2
v2 v3
b)
c) K 2 × K3
Figure 0.16: The join and the product of two graphs.
An Introduction to Relevant Graph Theory and Matrix Theory 19
⋮
xr
matrix, Or×s denotes the r × s matrix zero matrix with all entries 0, 0r
the zero vector with all entries 0, and 1r the vector with all entries 1. If
the size is understood we suppress the subscripts. We note that all
matrices and vectors are bold-faced.
20 Spanning Tree Results for Graphs and Multigraphs
k ≤ l , then det(PQ) =
1 2 3 ... k j1 j2 j3 ... jk
∑ det P
j3 ...
det Q
jk 1 2 3 ...
,
k
1≤ j1 < j2 <...< jk ≤l j1 j2
i1 i2 i3 ⋯ i p
where the minor C
j3 ⋯ j p
denotes the submatrix of C
j1 j2
obtained by deleting all rows except i1 , i2 ,…, i p and all columns except
l l l
∑
p1τ1 qτ11 ∑ p1τ 2 qτ 2 2 ⋯ ∑ p τ qτ 1 k kk
τ1 =1 τ2 =1 k =1 τ
l l l
∑
p2τ qτ 1
PQ = τ1 =1 1 1
∑ p τ qτ
τ
2 2 2 2 ⋯
τ k =1
∑
p2τ k qτ k k
.
2 =1
⋮ ⋮ ⋱ ⋮
l l l
pkτ k qτ k k
∑
p q
τ =1 kτ1 τ1 1 ∑ p τ qτ k 2 2 2 ⋯ ∑
1 τ 2 =1 τ k =1
But
22 Spanning Tree Results for Graphs and Multigraphs
p1 j1 q j11 p1 j2 q j2 2 ⋯ p1 jk q jk k
Iσ p2 j q j 1 p2 j2 q j2 2 ⋯ p2 jk q jk k
( −1) det 1 1 ,
⋮ ⋮ ⋱ ⋮
pkj1 q j11 pkj2 q j2 2 ⋯ pkjk q jk k
det(PQ) = ∑ ∑τ τ
1≤ j1 < j2 <...< jk ≤l σ ( j1 , j2 ,..., jk ) = ( 1 , 2 ,...,τ k )
(−1) Iσ q j11q j2 2 ⋅⋅⋅ q jk k i
1 2 ⋅⋅⋅ k
det P .
j1 j2 ⋅⋅⋅ jk
1 2 ⋅⋅⋅ k
det Q = ∑ ( −1) Iσ q j11q j2 2 ⋅ ⋅ ⋅ q jk k . □
j1 j2 ⋅⋅⋅ jk σ
4 1
2 −1 0
Example 0.5 Let P = and Q = 2 −1 , then applying
3 2 2 3 2
Binèt-Cauchy:
An Introduction to Relevant Graph Theory and Matrix Theory 23
2 1 4 1 1 0 2 1
det( PQ ) det det det det
3 2 2 1 2 2 3 2
2 0 4 1
+ det det 7 6 2 7 4 5 36 .
3 2 3 2
6 3 6 3
Multiplying the matrices, PQ = and det 36.
22 5 22 5
span x1 , x 2 ,…, x j is the set consisting of all linear combinations of the
vectors, i.e.
span x1 , x 2 ,…, x j c1x1 c2 x 2 … c j x j | c1 , c2 , , c j .
x1 y1
x y n
2
and y = , the dot product x ⋅ y = ∑x y
2
x= . We will also use
⋮ ⋮ i =1
i i
xn yn
the notation xT y . Two vectors x and y are orthogonal if x ⋅ y = 0 .
{y ∈ℝ n
}
| y ⋅ x = 0, for all x ∈ W . The Euclidean norm or just norm of
a vector, x e
or simply x , is given by x e = x ⋅ x . A set of vectors
1, i = k
pairwise orthogonal, i.e. xi ⋅ x k = . Any set of orthonormal
0, i ≠ k
vectors is necessarily linearly independent. An orthonormal basis is a
basis whose vectors form an orthonormal set.
Let A be an n × n matrix. An eigenvalue of A is a scalar λ such that
Ax = λ x for some non-zero vector x. Any non-zero vector x such that
Ax = λ x is an eigenvector associated with the eigenvalue λ . The
characteristic polynomial of A, PA ( λ ) = det ( λI − A ) . Eigenvalues are
k
exists unique real k × k matrices B and C such that if x = ∑ α i x i ,
i =1
β1 α1
⋮ = B ⋮
k k
Ax = ∑ β i x i , and A x = ∑ χi xi ,
T
then and
i =1 i =1
β k α k
χ1 α1
⋮ = C ⋮ . It is easy to see that BT = C so that if A is symmetric
χ k α k
x T Ax x T Ax
λ j ≥ min T
and λn − j +1 ≤ max T .
0 ≠ x∈ L j x x 0 ≠ x∈L j x x
i.e., Axi = λi xi , and let Lˆ j = span {x1 , x 2 ,…, x j } for j = 1, 2,…, n . Fix j,
( )
and observe that, since dim ( L j ) + dim Lˆ j = ( n − j + 1) + j > n , there
An Introduction to Relevant Graph Theory and Matrix Theory 27
n
exists a nonzero vector x0 ∈ ℝ belonging to both of these subspaces. In
j
particular, since x0 ∈ Lˆ j , it follows that x 0 = ∑ α k x k and thus,
k =1
T
j j j
x 0T Ax0 = ∑ α k x k A ∑α k x k = ∑ λk α k .
2
k =1 k =1 k =1
Hence,
j j
xT0 Ax 0
λ1 ∑ α k ≤ xT0 Ax 0 ≤ λ j ∑ α k , or λ1 ≤
2 2
≤ λj .
k =1 k =1 xT0 x0
xT Ax xT Ax
min ≤ T ≤ λj .
0 ≠ x∈L j xT x x0 x 0
{
span x1 , x2 ,..., x j . } □
{
min ( Ax , x ) = λ j . Indeed, taking L j = span x j , x j +1 ,..., x n , we obtain
0 ≠ x∈ L j
}
such a subspace. □
xT ( A + B )x x T Bx
α n = max ≥ max = β n , and
Lk xT x Lk xT x
xT ( A + B ) x
α k = max min ≥
Lk 0 ≠ x∈Lk xT x
xT Bx
max min = min max xT Bx = β k
Lk 0 ≠ x∈Lk xT x y j =1
e
x j =1, xT y j = 0
e
j = k +1,... n j = k +1,... n
λ − mii ≤ λ − mii ≤ ∑ mij . Since these inequalities are true for each
j ≠i
min mii −
i
∑m ij ≤ λ min ≤ λ max ≤ max mii +
i
∑m ij .
j ≠i j ≠i
Of course, if we are dealing with positive semi-definite matrices, then
λ = λ . This leads to the following corollary:
λmin ≥ min mii − ∑ mij and λmax ≥ max mii − ∑ mij .
i i
j ≠i j ≠i
Example 0.6
1 0 3 1 0 3
2 2 −1 1 −1 2 −1
1
1 0 3
2 −1 2 3 2 2 3 2
1 3 ⊗ 2 −1 1 = =
2 3 2 1 0 3 1 0 3
1 2 −1 1 3 2 −1
1
2 3 2 2 3 2
32 Spanning Tree Results for Graphs and Multigraphs
2 0 6 −1 0 −3
4 −2 2 −2 1 −1
4 6 4 −2 −3 −2
1 0 3 3 0 9
2 −1 1 6 −3 3
2 3 2 6 9 6
and
1 0 3 1 0 3
2 −1 2 −1
1 3 ⊕ 2 −1 1 = 1 3 ⊗ I 3 + I 2 ⊗ 2 −1 1 =
2 3 2 2 3 2
1 0 3 1 0 3
1 2 −1 1 0 2 −1 1
2I 3 −1I 3 2 3 2 2 3 2
1I + =
3 3I3 1 0 3 1 0 3
0 2 −1 1 1 2 −1 1
2 3 2 2 3 2
2 0 0 −1 0 0 1 0 3 0 0 0
0 2 0 0 −1 0 2 −1 1 0 0 0
0 0 2 0 0 −1 2 3 2 0 0 0
+ =
1 0 0 3 0 0 0 0 0 1 0 3
0 1 0 0 3 0 0 0 0 2 −1 1
0 0 1 0 0 3 0 0 0 2 3 2
An Introduction to Relevant Graph Theory and Matrix Theory 33
3 0 3 −1 0 0
2 1 1 0 −1 0
2 3 4 0 0 −1
.
1 0 0 4 0 3
0 1 0 2 2 1
0 0 1 2 3 5
Proof Since the entries of B and C are complex, there exist invertible
matrices P and Q such that B1 = PBP−1 and C1 = QCQ−1 are upper
triangular. Of course, B1i = PBi P−1 and C1j = QC j Q−1 are also upper
triangular, and therefore so is each matrix αij B1i ⊗ C1j ( ) as well as the
sum ∑α ( B
i, j
ij
i
1 )
⊗ C1j . Now it follows from Theorem 0.12 and Corollary
0.13 that
(B ⊗ C ) = P
i
1 1
j −1
( )
Bi P ⊗ Q −1C j Q = ( P ⊗ Q) −1 Bi ⊗ C j = ( P ⊗ Q )
∑α ( B
−1
ij
i
)
⊗ C j ( P ⊗ Q ) = ( P ⊗ Q ) φ ( B; C )( P ⊗ Q ) .
i, j
An Introduction to Relevant Graph Theory and Matrix Theory 35
∑ ς p (1 − p )
e −i
R ( M , p) = i
i
.
i = n −1
term, and since ς n −1 is the number of spanning trees, graphs with a larger
number of spanning trees will have greater reliability for such p.
37
38 Spanning Tree Results for Graphs and Multigraphs
Example 1.1 The directed multigraph depicted in Figure 1.1 has node-
arc incidence matrix
e1 e2 e3 e4 e5 e6 e7
v1 −1 1 0 0 0 0 −1
v2 1 −1 −1 0 0 0 0
( )
S M =
v3 0 0 1 −1 1 1 0
v4 0 0 0 1 −1 −1 1
Calculating the Number of Spanning Trees: The Algebraic Approach 39
e1
v1 v2
e2
e7 e4 e3
e5
v4 e6 v3
Since each arc has exactly two end-nodes, the sum of the row vectors of
S is 0Te . Clearly, any single row can be deleted from S without losing
any information about M . The matrix resulting from the deletion of the
ith row is called a reduced node-arc incidence matrix, and is denoted
( )
S r (i ) M , i.e., in terms of the notation of Chapter 0, this is the submatrix
1 2 3 ⋯ i −1 i +1 ⋯ n
S
1 2 ⋯ i − 1 i i + 1 ⋯ e
or alternatively, if Ei denotes the matrix that results from deleting row i
( )
of the unit matrix, S r ( i ) M = Ei S. Note that it is often convenient to
renumber the nodes of the graph so that node n is deleted and thus node
k of the renumbered nodes corresponds to row k of S r ( n ) .
k
Then ∑α s
i =1
i ji = 0 where α i = 1 if the jth arc is in the cycle, and α i = 0
∑α s
i =1
i ji =0.
( )
First assume the directed tree T has two nodes. Then S r (i ) T is an 1× 1
matrix with entry ±1 ; hence the determinant is ±1 . Now assume that any
directed tree on n − 1 nodes has corresponding reduced node-arc
incidence matrix determinant equal to ±1 . Now consider any reduced
node-arc incidence matrix of a directed tree T on n nodes. As T has at
least two pendant nodes, S r (i ) T( ) has at least one row having a single
v. Then expansion of ( ( ))
det S r (i ) T along that row yields
( )
consider the orientation M and S M given in Example 1.1.
42 Spanning Tree Results for Graphs and Multigraphs
v1 v2
v4 v3
Figure 1.3
T
Then H ( M ) = S M ( )( S(M ) ) =
−1 1 0 0
1 −1 0 0
−1 1 0 0 0 0 −1
0 − 1 1 0
1 −1 −1 0 0 0 0 0 0 −1 1 =
0 0 1 −1 1 1 0
0 0 1 −1
0 0 0 1 −1 −1 1 0 0 1 −1
−1 0 0 1
3 −2 0 −1
−2 3 −1 0
.
0 −1 4 −3
−1 0 −3 4
deg(vi ) if i = j
hij = .
− m(vi , v j ) if i ≠ j
Calculating the Number of Spanning Trees: The Algebraic Approach 43
3 2 0 1 3 0 0 0 0 2 0 1
2 3 1 0 0 3 0 0 2 0 1 0
H DA.
0 1 4 3 0 0 4 0 0 1 0 3
1 0 3 4 0 0 0 4 1 0 3 0
(f) ( H ( M )) ij 0 , whenever i j .
n
(g) trace H M H M ii 2e .
i 1
44 Spanning Tree Results for Graphs and Multigraphs
Proof (a) Follows trivially from the definitions and by Proposition 0.7.
(b) We have x T Hx = xT SS T x = x e ≥ 0. By Proposition 0.7, the
eigenvalues are all nonnegative.
(c) Follows trivially from the fact H = D − A .
(d) Since H1 = 0 by (c) we have that 0 is an eigenvalue, and H
is singular.
(e) and (f) follow trivially from the fact H = D − A .
(g) Follows from the fact that H = D − A and the First Theorem of
Multigraph Theory. □
Remark 1.1 From Proposition 1.2 (a), (b) and (e) we get the
eigenvalues of H ( M ) can be ordered
0 = λ1 ( H ( M ) ) ≤ λ2 ( H ( M ) ) ≤ … ≤ λn ( H ( M ) ) .
place of λi ( H ( M ) ) .
column j from matrix H , and the (i, j)-minor of H is det(Hij ) . The (i, j)-
i+ j
cofactor is given by Cij ( H ) = ( −1) ( )
det H ij . The adjugate matrix of
Proof First observe that Theorem 1.1 together with the Binet-Cauchy
Theorem yields t ( M ) = Cii ( H ) [Brooks, 1940]. Next, observe that by
Theorem 1.2, the rows of S sum to zero and H is symmetric,
thus H1 = 0 and 1T H = 0 , and so rank( H ) is strictly less than n. We
want to show that adj( H ) = α J , where J is the n × n matrix of all ones
and α is a scalar. If H is n × n , with rank less than n − 1 , then adj( H ) =
0, and the result is true, i.e. α = 0 . Thus, the case that remains is when the
rank is n − 1 or equivalently, the null space of H has dimension 1. Now
H adj( H ) = adj( H ) H = 0I , so each column of adj( H ) is an eigenvector
corresponding to the eigenvalue 0 and so adj ( H ) = α1 1, α 2 1,..., α n 1 for
α1 1T
symmetric, and thus, α11,α 2 1,..., α n 1 = ⋮ . Hence, all α i are
α 1T
n
equal, and adj( H ) = α J. □
46 Spanning Tree Results for Graphs and Multigraphs
Remark 1.2 A careful examination of the proof that all cofactors are
equal clearly indicates that all cofactors of a matrix are equal whenever
its rows and columns sum to zero, regardless of whether the matrix is
symmetric. Conversely, it is easily shown from the fact L adj ( L ) =
T
(
Remark 1.3 Consider H = SS T . Let H r ( n ) = S r ( n ) S r ( n ) ) denote the
Proposition 1.4
(a) If 0 = λ1 (G ) ≤ λ2 (G ) ≤ ... ≤ λn (G ) and 0 = λ1 (G ) ≤ λ2 (G ) ≤ ... ≤ λn (G )
( )
H G x = nx − H(G )x = (n − λ )x .
( )
H (G )x + H G x = H( K n )x = (nI n − J n )x = nx − J n x .
( )
H(G ) x + H G x = nx , and H (G ) x = ( n − λ ) x follows immediately.
( )
i.e., 0 = λ2 G = n − λn ( G ) .
(c) We defer the proof of this result until Section 1.6, where Proposition
1.19 provides a more general result. □
48 Spanning Tree Results for Graphs and Multigraphs
n n
n −1
PH′ ( M ) (0) = ∏ ( −λi (M ) ) = ( −1) ∏ ( λi (M ) ).
i=2 i=2
v1 v2
v4 v3
M
Figure 1.4
4 −2 −1 −1
−2 3 −1 0
Its Laplacian is H ( M ) = . Using the (1,1)-cofactor
−1 −1 5 −3
−1 0 −3 4
3 −1 0
det −1 5 −3 = 29 . The eigenvalues of H ( M ) are
1+1
t ( M ) = ( −1)
0 −3 4
2.676595722i5.642973632i7.681330643
t(M ) = = 29 .
4
50 Spanning Tree Results for Graphs and Multigraphs
Remark 1.5 The presence of more than one eigenvalue equal to zero
implies a disconnected multigraph. In fact it is easily shown by
examining the block form of the Laplacian that the number of zero
eigenvalues is the same as the number of components. Thus, the formula
is valid for disconnected multigraphs as well.
We are now in position to find the number of spanning trees for several
different families of graphs.
Proposition 1.6
(a) (Cayley’s Theorem) [Cayley, 1889] The number of spanning
trees of K n is n n− 2 .
(b) The number of spanning trees of a graph that is regular of
degree r and having adjacency matrix eigenvalues
n
1
α1 ≥ α 2 ≥ ... ≥ α n is
n ∏ ( r − α (G )) .
i=2
i
n −1 1 n −1
PH ( K n ) ( λ ) = λ ( λ − n ) . By Theorem 1.5, t ( K n ) = ( n ) = nn−2 .
n
Calculating the Number of Spanning Trees: The Algebraic Approach 51
Our next task is to determine the number of spanning trees for the
regular graph K n × K m using Proposition 1.6(b). To do this, we first
require a result for the eigenvalues of the adjacency matrix of the product
of two graphs based on the eigenvalues of the individual graphs.
(
lexicographically (i.e., label of ui , v j ) is smaller than that of ( u , v ) if
k ℓ
A(G1 × G 2 ) = A ( G1 ) ⊕ A ( G2 ) .
⋮ ⋮ ⋱ ⋮
Bn1 Bn 2 ⋯ Bnn
(
corresponding to u β , v j ) is
{ }
0 if i ≠ j or if i = j but u χ , u β ∉ E1
.
{
1 if i = j and u χ , uβ ∈ E1 }
In conclusion, A ( G1 ) ⊕ A ( G2 ) is the adjacency matrix of G1 × G2 . □
( )
summands, we require the eigenvalues of A K p . But the eigenvalues
A ( K n × K m ) are:
t ( Kn × K m )
1 ( n −1)( m −1) m −1 n −1
=
mn
( m + n − 2 − ( −2) ) ( m + n − 2 − (n − 2) ) ( m + n − 2 − (m − 2) )
0 1 1 0 0 0 1 1
1 0 1 1 0 0 0 1
1 1 0 1 1 0 0 0
0 1 1 0 1 1 0 0
.
0 0 1 1 0 1 1 0
0 0 0 1 1 0 1 1
1 0 0 0 1 1 0 1
1 1 0 0 0 1 1 0
54 Spanning Tree Results for Graphs and Multigraphs
Its rows are successive shifts to the right of the row vector
[0 1 1 0 0 0 1 1] . This is true in general for any circulant
graph, i.e. the adjacency matrix is obtained by successively shifting the
first row vector to the right by one place. More formally, row k of the
circulant matrix A generated by the vector a = a1 ,..., an is
( )
ε k −1 a0,..., an −1 , where ε represents the shift operator, i.e.,
ε ( b1 ,..., bn ) = ( bn , b1 ,..., bn −1 ) . Furthermore, by setting En equal to the
0 1 0 0 ⋯ 0
0 0 1 0 ⋯ 0
0 0 0 1 ⋯ 0
En =
⋮ ⋮ ⋮ ⋮ ⋱ ⋮
0 0 0 0 ⋯ 1
1 0 0 0 ⋯ 0
∑a E
i
we obtain A = i
i
n where Ein = ( E n ) . Those nonzero circulant
i =1
( )
A C82 = E8 + E82 + E86 + E87 . To compute the eigenvalues of a circulant
matrix,
det ( λ I n − En ) =
Calculating the Number of Spanning Trees: The Algebraic Approach 55
λ −1 0 0 ⋯ 0
0 λ −1 0 ⋯ 0
0 0 λ −1 ⋯ 0 n −1 1+ n n −1 n
det = λ ⋅ λ − 1( −1) ( −1) = λ − 1
⋮ ⋮ ⋮ ⋮ ⋱ ⋮
0 0 0 0 ⋯ −1
−1 0 0 0 ⋯ λ
given by
n − ni
k i 2π j i i 2π j k 2π ni j ni
∑ e n + e n =2 ∑ cos
n for j = 0,1,… , n − 1 .
i =1 i =1
n
On the other hand, if nk = , the eigenvalues of
2
n
n k
A Cn n1 , n2 ,..., nk −1 , =
2 ∑(
i =1
E nni + E nn − ni )+ E n2
are given by
k
2π ni j
∑ cos
j
2 + ( −1) , j = 0,1,..., n − 1 .
i =1 n
(
t Cn ( n1 ,..., nk ) = )
56 Spanning Tree Results for Graphs and Multigraphs
2n −1 n −1 k
2π ni j
∏ ∑ cos
n
k − if nk <
n j =1 i =1
n 2
n
−1
n .
2n −1 2 k
4π ni m 2 k
2 ( 2m − 1) π ni
∏ ∑ ∏ ∑
n
k − cos k − 1 − cos if nk =
n
m =1
n n 2
i =1 m =1 i =1
The next result, which follows from Theorem 1.5, gives the number of
spanning trees for G in terms of the Laplacian matrix of the complement
G of G.
n
1
λi ( G ) = n − λn −i ( G ) for i = 1,..., n − 1. Thus, t (G ) =
n ∏ ( n − λ (G )).
i=2
i
n n
1
Now, PH ( G ) ( λ ) = λ ∏ ( λ − λi ( G ) ) so that
i =2
P
n H(G )
(n) = ∏ ( n − λ (G )) ,
i =2
i
k
C1 , C2 ,..., Ck , then PH ( G ) ( λ ) = ∏ P ( ) (λ ).
i =1
H Ci
We record this refinement
Gχ , then t ( G ) = n n − χ − 2 PH □
( Gχ ) (
n).
Proposition 1.12 The formulas for the number of spanning trees for
some special types of graphs are as follows:
k
2
(a) K n minus a matching (Figure 1.5): t ( K n − kK 2 ) = n n − 2 1 − .
n
(b) k-partite graphs (Figure 1.6)
(i) (
complete bipartite graph: t K p , q = q p −1 p q −1 ; )
(ii) regular complete k-partite graph:
(r −1)k
t K r ,r ,...,r = r rk − 2 k k − 2 ( k − 1) ;
k terms
58 Spanning Tree Results for Graphs and Multigraphs
(c)
(i) complete graph of order n minus a path of order p, n > p .
(Figure 1.7):
( )
t K n − Pp =
2
p +1
2
p +1
n n − p −2 n − 2 + n − 2 − 1 − n − 2 − n − 2 − 1 if n > 4
2 2
2
2 2
2 n − 2 − 1 .
2
3− p
4 ⋅ p if n = 4
( )
t Kn − C p =
2
p
n−2 2
p
n − p −2 n − 2 n−2 n−2 p +1
n
2 + 2 − 1 + − 2 − 1 + 2 ( − 1) n
if > 4
2
p
2 − 2 ( − 1 ) if n = 4 and p = 3, 4
4
disjoint. Two special cases are the complete bipartite case and the
complete k-partite case where all parts have the same order. In the
first case, k = 2 and n = χ . Setting χ1 = p and χ 2 = q , we see that
( )
t K p , q = q p −1 p q −1 .
(r −1)k
t K r ,r ,..., r = r rk − 2 k k − 2 ( k − 1) .
k terms
( )
appropriately labeled, H G has the block form given below:
( p + 1) I p − J p −I p
( )
H G =
−I p ( p + 1) I p − J p
.
Of course,
( λ − ( p + 1) ) I p + J p Ip
λI2 p − H ( G ) = .
Ip ( λ − ( p + 1) ) I p + J p
60 Spanning Tree Results for Graphs and Multigraphs
A B
Since any matrix with square blocks A, B, C, and D has
C D
determinant equal to det (AD-BC) if A and C commute, we may
write:
( ) ) = det ( ( ( λ − ( p + 1) ) I )
2
(
det λ I 2 p − H G p + Jp ) − Ip =
( )
But det λ I p − J p = λ p −1 ( λ − p ) , so that
p −1
a a
(
det aI p − bJ p ) =b
b
p
p −1
b − p = a ( a − bp ) .
( )
det aI p − bJ p = a p −1 ( a − bp ) .
a) K 2,3
b) K 2,2,2
c) K3,3 − 3K 2
1 −1 0 0 ⋯ 0
−1 2 −1 0 ⋯ 0
0 −1 2 −1 ⋱ 0
( )
H Pp = .
⋮ ⋱ ⋱ ⋱ ⋱ 0
0 ⋯ 0 −1 2 −1
0 ⋯ 0 0 −1 1
x −1 0 0 ⋯ 0
−1 x + 1 −1 0 ⋯ 0
0 −1 x + 1 −1 ⋱ 0
.
⋮ ⋱ ⋱ ⋱ ⋱ 0
0 ⋯ 0 −1 x + 1 −1
0 ⋯ 0 0 −1 x
After subsequent expansion about the first and last rows (respectively),
denote by g p ( y ) the determinant of the p × p matrix
y −1 0 0 ⋯ 0
−1 y −1 0 ⋯ 0
0 − 1 y −1 ⋱ 0
.
⋮ ⋱ ⋱ ⋱ ⋱ 0
0 ⋯ 0 −1 y −1
0 ⋯ 0 0 −1 y
g p −1 (1 + x) − (1 + x) g p − 2 (1 + x) + g p −3 (1 + x) = 0 .
(1 − x ) yields
(1 − x) g p (1 + x) + ( x 2 − 1) g p −1 (1 + x) + (1 − x) g p − 2 (1 + x) = 0 .
x 2 g p −1 ( x + 1) − 2 xg p − 2 ( x + 1) + g p − 3 ( x + 1) − ( x − 1) g p ( x + 1) = 0,
f p ( x ) = x 2 g p − 2 ( x + 1) − 2 xg p −3 ( x + 1) + g p − 4 ( x + 1) =
( x − 1) g p −1 ( x + 1) . (*)
U p + 2 ( z ) − 2 zU p +1 ( z ) + U p ( z ) = 0, where U 0 ( z ) = 1 and U1 ( z ) = 2 z ,
y
then g p ( y ) = U p . It is known [Seshu, 1961] that
2
p +1 p +1
U ( z) =
p
( z+ z2 − 1 ) −(z − z2 − 1 )
2 z2 − 1
y p
and g p ( y ) = U p , we obtain, f p ( x ) = 0 if x = 1 , 4 p ( −1) if x = −3 ,
2
and
2
p
x +1 2
p
x −1 x + 1 x +1 x + 1
+ −1 − − −1
2 2 2 2
2
x +1
2 − 1
2
(
polynomial det λ I p − H Pp ( ) ) = ( −1) p
f p (1 − λ ) = 0 if λ = 0, 4p if λ = 4 ,
p +1 2
p
2−λ 2
p
( −1) 2 − λ 2−λ
λ 2 − λ
2 + 2 −1 − 2 − 2 −1 .
2
2−λ
2 −1
2
2
p
2
p
n n − p −1
n − 2 n−2 n − 2 n − 2
(n n − p −1
) 2 2 + 2 − 1 − 2 − 2 − 1
n−2
2 − 1
2
if n > 4 .
(ii) We first consider the Laplacian matrix of the cycle,
2 −1 0 0 ⋯ −1
− 1 2 −1 0 ⋯ 0
0 −1 2 −1 ⋱ 0
H Cp = ( ) .
⋮ ⋱ ⋱ ⋱ ⋱ 0
0 ⋯ 0 −1 2 −1
−1 0 ⋯ 0 −1 2
We observe, in a similar fashion to the proof in (c)(i), that this matrix has
the form
y −1 0 0 ⋯ −1
− 1 y −1 0 ⋯ 0
0 − 1 y −1 ⋱ 0
⋮ ⋱ ⋱ ⋱ ⋱ 0
0 ⋯ 0 −1 y −1
−1 0 ⋯ 0 −1 y
and its determinant, hp ( y ) , can be written hp ( y) = y g p −1 ( y)
− 2 g p − 2 ( y) − 2 , where g p ( y) is the determinant of the matrix described
in the proof of (c)(i). Similar calculations lead to
( ( ) ) = ( −1)
det λ I p − H C p
p
hp (2 − λ ) .
Calculating the Number of Spanning Trees: The Algebraic Approach 65
2
p
2
p
n− p−2 n − 2 n − 2 n − 2 n − 2 p +1
n 2 + 2 − 1 + 2 − 2 − 1 + 2 ( −1)
if n > 4 .
The following example checks the validity of the formulas for the
number of spanning trees found in this section. We do not include K n
here, as Cayley’s formula is well known.
Example 1.4 We consider the graphs depicted in Figures 0.16c, 0.9, and
1.5-1.8. For each we find t ( G ) by first applying the formulas found in
Corollary 1.8, Proposition 1.9 and Proposition 1.12, and second by
computing the (1,1)-cofactor of the Laplacian of G.
66 Spanning Tree Results for Graphs and Multigraphs
3 −1 −1 −1 0 0
−1 3 −1 0 −1 0
−1 −1 3 0 0 −1
One Laplacian matrix for K 2 × K 3 is which
−1 0 0 3 −1 −1
0 −1 0 −1 3 −1
0 0 −1 −1 −1 3
has (1,1)-cofactor 75 as well, confirming the result of the formula.
(
By Proposition 1.9: t Cn ( n1 ,..., nk ) = )
2n −1 n −1 k
2π ni j
∏ k − ∑ cos
n
if nk <
n j =1 i =1
n 2
n
−1
n .
2n −1 2 k
4π ni m 2 k
2 ( 2m − 1) π ni
∏ ∑ ∏ ∑
n
k − cos k − 1 − cos if nk =
n m =1
n m =1 n 2
i =1 i =1
n
For C8 (1,2 ) (Figure 0.7) k = 2, n = 8, n1 = 1, and n2 = 2. Thus, nk <
2
so we use the top formula, and
Calculating the Number of Spanning Trees: The Algebraic Approach 67
28−1 8 −1
2
2π ni j
t ( C8 (1, 2 ) ) =
8 ∏ 2 −
j =1
∑ cos
i =1 8
=
2 7 7
2π n1 j 2π n2 j
8 ∏ 2 − cos 8 + cos
8
=
j =1
27 2π (1)(1) 2π ( 2 )(1)
2 − cos + cos ×
8 8 8
2π (1)( 2 ) 2π ( 2 )( 2 )
2 − cos + cos ×
8 8
2π (1)( 3) 2π ( 2 )( 3 )
2 − cos + cos ×
8 8
2π (1)( 4 ) 2π ( 2 )( 4 )
2 − cos + cos ×
8 8
2π (1)( 5 ) 2π ( 2 )( 5 )
2 − cos + cos ×
8 8
2π (1)( 6 ) 2π ( 2 )( 6 )
2 − cos + cos ×
8 8
2π (1)( 7 ) 2π ( 2 )( 7 )
2 − cos + cos = 3528.
8 8
4 −1 − 1 0 0 0 −1 −1
−1 4 −1 −1 0 0 0 −1
−1 −1 4 −1 −1 0 0 0
0 −1 − 1 4 −1 −1 0 0
0 0 −1 −1 4 −1 − 1 0
0 0 0 −1 −1 4 −1 −1
−1 0 0 0 −1 −1 4 −1
−1 −1 0 0 0 −1 −1 4
2
2 4
t ( K 6 − 2 K 2 ) = 6 6 − 2 1 − = 6 4 = 576.
6 9
4 −1 −1 −1
−1 0
−1 4 −1 −1
0 −1
−1 −1 5 −1
−1 −1
One Laplacian matrix for K 6 − 2 K 2 is
−1 0 −1 4 −1
−1
0 −1 −1 −1
−1 4
−1 −1 −1 −1 −1 5
which has (1,1)-cofactor 576 as well, confirming the result of the
formula.
Calculating the Number of Spanning Trees: The Algebraic Approach 69
( )
By Proposition 1.12 b i): t K p , q = q p −1 p q −1 .
( )
For K 2,3 (Figure 1.6 a) p = 2 and q = 3, so t K 2,3 = 32−123−1 = 12.
3 0 −1 −1 −1
0 3 −1 −1 −1
One Laplacian matrix for K 2,3 is −1 −1 2 0 0 which has
−1 −1 0 2 0
−1 −1 0 0 2
(1,1)-cofactor 12 as well, confirming the result of the formula.
(r −1)k
By Proposition 1.12 b ii): t K = r rk − 2 k k −2 ( k − 1) .
r ,r ,...r
k terms
(2 −1)3
( )
t K 2,2,2 = 22(3) − 233− 2 ( 3 − 1) = 24 ⋅ 3 ⋅ 23 = 384.
4 0 −1 −1 −1 −1
0 4 −1 −1 −1 −1
−1 − 1 4 0 −1 −1
One Laplacian matrix for K 2,2,2 is which
−1 − 1 0 4 −1 −1
−1 −1 −1 −1 4 0
−1 −1 −1 −1 0 4
70 Spanning Tree Results for Graphs and Multigraphs
Example 1.4 f – Complete bipartite with equal part sizes minus a full
matching K p , p − pK 2 .
p −1
( )
By Proposition 1.12 b iii): t K p , p − pK 2 = ( p − 1) p p − 2 ( p − 2 ) .
3 −1
( )
t K3,3 − 3K 2 = ( 3 − 1) ⋅ 33− 2 ⋅ ( 3 − 2 ) = 6.
2 0 0 0 −1 −1
0 2 0 −1 0 −1
0 0 2 −1 −1 0
One Laplacian matrix for K3,3 − 3K 2 is
0 −1 −1 2 0 0
−1 0 −1 0 2 0
−1 −1 0 0 0 2
which has (1,1)-cofactor 6 as well, confirming the result of the formula.
2
p +1
2
p +1
nn− p −2 n − 2 + n − 2 − 1 − n − 2 − n − 2 − 1 if n > 4
2 2 2 2
2
= 2 n − 2 − 1
(
t K n − Pp )
.
2
3− p
4 ⋅ p if n = 4
t ( K 5 − P3 ) =
2
3 +1
5 2 2
3 +1
55 −3 − 2 5 − 2 5−2 − 5 − 2
2 + 2 − 1 − 2 − 2 − 1 = 21.
5−2
2
2 − 1
2
3 −1 −1 0 −1
−1 3 0 −1 −1
One Laplacian matrix for K 5 − P3 is −1 0 2 0 −1 which has
0 −1 0 2 −1
−1 −1 −1 −1 4
(1,1)-cofactor 21 as well, confirming the result of the formula.
2
p
n−2 2
p
n − p −2 n − 2 n−2 n−2
+ 2 ( −1) if n > 4
p +1
n + − 1 + − − 1
2 2 2 2
p
2 − 2 ( −1)
if n = 4 and p = 3, 4
4
so t ( K5 − C3 ) =
2
3
2
3
5 5 − 3− 2 5 − 2 + 5 − 2 − 1 − 5 − 2 − 5 − 2 − 1 = 47.
2 2 2 2
72 Spanning Tree Results for Graphs and Multigraphs
2 −1 0 −1 0
−1 4 −1 −1−1
One Laplacian matrix for K 5 − C3 is 0 −1 2 −1 which has
0
0 −1 0 −1 2
−1 −1 −1 −1 4
(1,1)-cofactor 47 as well, confirming the result of the formula.
PL ( λ )( λ − gn ) = PL + gJ ( λ ) λ.
Calculating the Number of Spanning Trees: The Algebraic Approach 73
1 −l12 −l13 ⋯
−l1n
1 λ − l −l23 ⋯
−l2 n n
λ det
⋮ ⋮
22
⋮ ⋱ ⋮ i =1
∑
= λ C (λI − L ).
i1
1 −ln 2 −ln 3 ⋯ λ − lnn
Similarly, consider PL + gJ ( λ ) =
n
1 1
Therefore, ∑ C ( λI − L ) =
i =1
i1
λ
PL ( λ ) = P
λ − gn L + gJ
(λ) . □
PH ( λ )( λ − n ) = PL ( λ )( λ − n ) = PL + J ( λ ) λ = PB ( λ ) λ □
1
particular, if k = 2, t ( G ) = n−2
t ( G1 ) t ( G2 ) .
n
Calculating the Number of Spanning Trees: The Algebraic Approach 75
H (C 1 ) −1 ⋯ −1
−1 H (C 2 ) −1 ⋮
Proof Realize that H ( G ) = so that
⋮ −1 ⋱ −1
−1 ⋯ −1 H (C k )
B (C1 ) 0 ⋯ 0
0 B (C 2 ) 0 ⋮
B (G ) = and
⋮ 0 ⋱ 0
0 ⋯ 0 B (C k )
k
1 1
t (G ) =
n 2
det ( B ( G ) ) = 2
n
∏ det ( B (C ) ).
i =1
i
B (C i ) 0
Now, B ( Gi ) =
0
( )
, so det B ( Gi ) = n n − ni det B (C i ) and
nI n − ni
and ( )
k
1
t ( Gi ) =
n 2 ( ) (
det B ( Gi ) = n n − ni − 2 det B (C i ) . Thus, ) ∏ t (G ) =
i =1
i
k k
∏ ( )
n n − ni − 2 det B (C i ) = n kn − n − 2 k ∏ det ( B (C ) ) =n( i
k −1) n − 2 k
n 2t ( G ) .
i =1 i =1
k
1
Therefore, t ( G ) = ( k −1) n − 2( k −1) ∏ t ( G ). i □
n i =1
det ( L + gJ ) = gn2C ( L ) .
76 Spanning Tree Results for Graphs and Multigraphs
Proof By Remark 1.2 (following the proof of Theorem 1.3), L has row
and column sum zero. Thus, the addition of all columns of L + gJ to the
first yields:
gn l12 + g l13 + g ⋯ l1n + g
gn l + g l + g ⋯ l2 n + g
det
22 23
=
⋮ ⋮ ⋮ ⋱ ⋮
gn ln 2 + g ln 3 + g ⋯ lnn + g
det ( B ) = n 2 C ( H ) = n 2 ( t ( G ) ) .
Calculating the Number of Spanning Trees: The Algebraic Approach 77
1.5 Multigraphs
eigenvalues of H .
As the minimum eigenvalue of H is 0, the maximum eigenvalue of
B (and thus of H as well) is n. This fact does not generalize to
multigraphs, as demonstrated by the multigraph given in Figure 1.9.
3 −1 −1 −1 0
−1 3 −1 −1 0
Its Laplacian matrix is H = −1 −1 3 0 −1 which has
−1 −1 0 4 − 2
0 0 −1 −2 3
1 1
PH ( λ ) = Pˆ ( λ ) ,
λ λ − gn B
80 Spanning Tree Results for Graphs and Multigraphs
λ2 ( M ) ≥ 2δ − g ( n − 2 ) ,
∑h
j ≠i
ij + g = g ( n − 1) − di , since hij + g = g − α ij , where α ij is the
( )
λ0 Bˆ ≥ bˆii − ∑ bˆij . Thus,
j ≠i
{
λ1 ( M ) ≥ min i di + g − ( g ( n − 1) − di ) = }
min i {2 di − g ( n − 2 )} ≥ 2δ − g ( n − 2 ) .
Calculating the Number of Spanning Trees: The Algebraic Approach 81
2δ − g ( n − 2 ) = 2 g ( a − 1 + k ) − g ( k + 2a − 2 ) = gk = gκ . □
gK a
+g
gK k
+g
gK a
Figure 1.12: The graph that realizes the bounds. Note that there are g edges
between each node in gK k and each node in both gK a .
∆ ∆ ∆
PH ( M ) ( λ ) = λ
∏ ( λ − m ) − ∑ m ∏ ( λ − m ) ,
i i j
i =1 i =1 i≠ j
∆
i =1
∑
mi − m1 ⋅ ⋅ − m∆
−m m1 0 ⋅ 0
H= 1 .
− m2 0 m2 0 ⋅
⋅ ⋅ ⋅ ⋅ ⋅
− m∆ 0 ⋅ 0 m∆
∆
∆ ∆
(m ) 2
PH ( M ) ( λ ) = ∏ ( λ − mi ) λ − ∑ mi − ∑ (λ −i m )
i =1 i =1 i =1 i
because
a x1 x2 ⋅ xn
x y1 0 ⋅ 0
1 n n n
∑( x ) ∏ y .
2
det x2 0 y2 0 ⋅ =a ∏ yi − i j
i =1 i =1 i≠ j
⋅ ⋅ ⋅ ⋅ ⋅
xn 0 ⋅ 0 yn
∆ ∆
mi ∆ ∆ ∆
∏ ( λ − mi ) λ − λ ∑
= λ
( λ − mi ) ∏ ( λ − mi ) − ∑ mi ∏ λ − m j .
( )
i =1 i =1 i =1 i =1 i≠ j
□
A multistar with underlying graph K1,q whose q leaf nodes all have
matrix for the multstar and A the Laplacian matrix for the rest of the
graph. Then λn ( M ) ≥ (1 + ∆ ) f . □
( )
g, then λi M ( ) = ng − λn −i + 2 for 2 ≤ n are the eigenvalues of
g
( )
H M ( g ) = g ( nI − J ) − H ( M ) . Furthermore, λn = ng if and only if
M (g) is disconnected.
(
Proof For ease of notation, let H = H ( M ) and H = H M ( g ) . Let λ )
be a nonzero eigenvalue of H , and x an associated eigenvector, i.e.,
Hx = λ x. Then 1T H = 0 implies 1T x = 0. Thus, it follows from
Hx + Hx = g ( nI − J ) x that Hx = ( gn − λ ) x. Likewise, H1+ Η1 =
Hx = λ2 x. Now,
(
from below: λn ( M ) ≥ g n − ∆ M ( g )( )), where ∆ is the maximum
( ) ( ) (
Proof λn ( M ) = ng − λ2 M ( g ) ≥ ng − g ∆ M ( g ) = g n − ∆ M ( g ) ( )). □
There are some results for the Laplacian matrix that rely on some
techniques of matrix theory. The Laplacian matrix, besides being real
and obviously symmetric, is positive semi-definite, which means that
88 Spanning Tree Results for Graphs and Multigraphs
1
The reasoning is as follows: let , x 2 , x 3 ,…, x n be an orthonormal
n
n
basis of eigenvectors, and consider x = ∑α x .
i =2
i i Now xT Px =
n
≥ λ 2 x . Also, xT Px = λ 2 . Consider the matrix Pɶ = P −
2
∑λ α
i =2
i
2
i
( ) (
c22 xT Px − λ2 xT ( I n − n −1J n ) x ≥ c22 λ2 x − λ2 x
2 2
) = 0.
Therefore, P is non-negative definite, and thus,
1
min i mii − λ2 1 − ≥ 0. □
n
n
Corollary 1.27 For any multigraph M, λ2 ( M ) ≤ δ.
n −1
n n
Proof Clearly, λ2 ( M ) ≤ min hii = δ. □
n −1 n −1
n
Corollary 1.28 For any multigraph M, λn ( M ) ≥ ∆.
n −1
Proof Since ∆ ( M ) = ( n − 1) g − δ ( M ) ,
(
δ M (g) = ) ( n − 1) g − ∆ ( M ) .
But
n n n
( )
λ2 M ( g ) ≤
n −1
(
δ M (g) = )
n −1
( ( n − 1) g − ∆ ( M ) ) = ng −
n −1
∆ ( M ).
Hence,
n n
( )
ng − λn ( M ) = λ2 M ( ) ≤ ng −
g
n −1
∆ ( M ) , or λn ( M ) ≥
n −1
∆ ( M ).
It can be said that K n has the greatest number of spanning trees of any
graph in the same class. As a matter of fact, K n is the only graph in its
class, so the result is obvious. However, it is also true that K n has the
greatest number of spanning trees among all multigraphs in its class, and
that it is unique in this regard. This was originally established by
[Kelmans, 1974]. We include a proof for completeness. First, because
90 Spanning Tree Results for Graphs and Multigraphs
1 n 1
(Proposition 1.6(a)) follows: t ( K n ) = ∏ λi = n n −1 = n n − 2 . Since
n i =2 n
the diagonal entries of the Laplacian matrix are the degrees of the nodes,
n
trace ( H ) = 2e = ∑ λi , which is the same for all multigraphs in a given
i =1
The following theorem extends this result. After obtaining this result
we learned that apparently it was also proven by Kelmans [Kelmans,
1967]. However, since the original paper has never been translated into
English, we shall provide a proof here, which first appeared in [Heinig,
2013]:
Proof Indeed,
Bˆ ( gK n ) = H ( gK n ) + gJ =
g ( n − 1) −g ⋯ −g g g ⋯ g
−g g ( n − 1) ⋯ − g g g ⋯ g
+ =
⋮ ⋮ ⋱ ⋮ ⋮ ⋮ ⋱ ⋮
− g ⋯ −g g ( n − 1) g g ⋯ g
gn 0 ⋯ 0
0 gn 0 ⋮
.
⋮ ⋮ ⋱ ⋮
0 ⋯ 0 gn
n
The eigenvalues have product ( gn ) . To prove uniqueness, suppose that
93
94 Spanning Tree Results for Graphs and Multigraphs
Problem Pt :
Maximize t ( M ) subject to
M ∈ Ω ( n, e ) .
Problem Ph :
1 n
Maximize ∏ λi subject to
n i=2
1. {0, λ1 , λ2 ,… , λn } is the spectrum of an n × n symmetric matrix
n
3. ∑h
i =1
ii = 2e .
Note that this relaxation contains instances which are not feasible for
Pt because, if H = hij is the Laplacian of a multigraph M then the
main diagonal terms must also meet the condition: if hkk = max hii then
1≤ i ≤ n
Multigraphs with the Maximum Number of Spanning Trees: An Analytic Approach 95
hkk ≤ ∑ hii [Hakimi, 1962]. The reader will also observe that both Pt and
i≠k
Problem Ps :
1 n
Maximize ∏ λi subject to
n i =2
1. λi ≥ 0 for all i = 2,3,…, n
n
2. ∑λ
i=2
i = 2e ,
as
n
∑h + ∑h
i =1
2
ii
i≠ j
2
ij =C
96 Spanning Tree Results for Graphs and Multigraphs
for Ph ,Cm then a solution for Pt is obtained. As it happens there are two
such cases that we shall present.
To begin with we give necessary and sufficient conditions on C and e
to insure that Ps ,C is well-posed, i.e. has feasible solutions and an
optimal objective value.
4e 2
≤ C ≤ 4e 2 . Furthermore, the set of feasible solutions contains an n-
n −1
4e 2
tuple with all positive entries if and only if ≤ C < 4e 2 .
n −1
Proof First suppose that λ2 , λ3 ,…, λn satisfies the constraints. Then, with
i =2 i< j
Multigraphs with the Maximum Number of Spanning Trees: An Analytic Approach 97
4e 2
so ≤ C ≤ 4e 2 . Moreover, if one of the λi ' s is positive, 4e 2 > C .
n −1
Conversely, suppose the inequalities hold. In the event that C = 4e2 , the
n − 1 -tuple ( 2e,0,0,… ,0 ) satisfies the constraints. Next suppose that
4e 2 2e
≤ C < 4e 2 . Realize that λ = 1 terminates on the plane
n −1 n −1
2e 2e 2e
λ ⋅ 1 = 2e . Now (2e, 0, 0,… ,0 ) − 1 ⋅ 1 = 0 so x = 1
n −1 n −1 n −1
4e 2
and y = C − u , where u is the unit vector in the direction of
n −1
2e
( 2e,0,0,…,0) − 1 , are perpendicular. We claim that λ = x + y is a
n −1
vector with positive components that satisfy the constraints of Ps ,C .
Indeed,
λ ⋅ 1 = x ⋅ 1 + y ⋅ 1 = x ⋅ 1 = 2e .
Also
2
2 2e 4e 2
λ = + C − =C.
n −1 n −1
Next
( 2e,0,… ,0 ) −
2e 2e 1 1 1 1
,… , = 2e 1 − ,− ,… , − = 2e 1 −
n −1 n −1 n −1 n −1 n −1 n −1
1 1 1 1
so that u = 1 − ,− ,… , − . Therefore
1 n −1 n −1 n − 1
1−
n −1
2e 4e 2 1
λ2 = + C− 1− >0
n −1 n −1 n −1
and, for i > 3
98 Spanning Tree Results for Graphs and Multigraphs
2e 4e 12
1
λi = − C− > 0.
n −1 n −1 1 n −1
1−
n −1
1 n
Of course the continuous function ∏ λi achieves a maximum value
n i=2
on the compact set defined by the constraints. □
4e 2
Remark 2.1 If ≤ C ≤ 4e2 holds, then Problem Ps ,C has an optimal
n −1
solution, because the objective function is continuous on the nonempty
compact set defined by the constraints.
4e 2
whenever ≤ C < 4e2 , replacement of constraint (1) in Ps ,C by the
n −1
requirement that all λi be strictly positive, yields a problem having the
solution. We designate this problem as Problem Ps,C , p :
Problem Ps ,C , p
n
1
Maximize
n ∏λ i =2
i subject to
1. λi > 0, i = 2,3,..., n
n
2. ∑λ
i =2
i = 2e
Multigraphs with the Maximum Number of Spanning Trees: An Analytic Approach 99
n
3. ∑ (λ )
i=2
i
2
=C .
4e 2
impossible, because C > . Therefore, rank ( ℑ ) = 2 .
n −1
n
In addition, f ( λ ) = ∏λ
i =2
i has continuous partials in a neighborhood of
n
n
n
LL((λλ,,α11, α 2 ) = ∏ λi ++αα11 ∑ λλi −−22ee ++αα2 2 ∑ λi2λ− −C C
i =2 i=2 i =2
then there exists α1* , α 2* such that L has a critical point at ( λ * , α1* , α 2* ) .
∂L
In particular,
∂λi
= ∏λ
j ≠i
j + α1 + 2α 2 λi = 0 for i = 2,..., n must be
satisfied by ( λ * , α1* , α 2* ) . Suppose there exists λi1 * < λi2 * < λi3 * (i.e., three
distinct entries). Then
∂L *
1. = ∏λ j + α1 + 2α 2 λi*1 = 0 ,
∂λi*1 j ≠ i1
∂L *
2. = ∏λ j + α1 + 2α 2 λi*2 = 0 ,
∂λi*2 j ≠ i2
∂L *
3. = ∏λ j + α1 + 2α 2 λi*3 = 0 .
∂λi*3 j ≠ i3
1 *
Subtracting the first two equations leads to α 2 =
2 ∏λ
j ≠ i1 ,i2
j , while
1 *
subtracting the last two equations yields α 2 =
2 ∏λ
j ≠i2 ,i3
j , and putting
these together implies that λi*1 = λi*3 , which contradicts λi*1 < λi*3 . Therefore,
Proposition 2.3
(i) The collection of m-tuples λ in which there are only two distinct
entries and which satisfy
1. λ ⋅ 1ɶ = 2e
Multigraphs with the Maximum Number of Spanning Trees: An Analytic Approach 101
2 4e 2
2. λ = C , λi > 0 (where where < C < 4e 2 )
n −1
is finite in number. The possible multiplicities for the larger entry L are
1, 2,… , n − 2 . For each such multiplicity k,
( n − 1)( n − 1 − k ) 4e 2
2e + C −
k n −1
L(k ,2e, C ) = and
n −1
(n − i)k 4e 2
2e − C −
n −1− k n −1
S (k , 2e, C ) =
n −1
for the larger and smaller entries L ( k , 2e, C ) and S ( k , 2e, C ) ,
respectively, or simply, just L and S.
n
1
(ii) The objective function
n −1 ∏λ
i=2
i achieves its maximum only at
λ * = ( L(1, 2e, C ), S (1, 2e, C ),..., S (1, 2e, C )). This location is unique up to
permutation of the entries.
2e = kL + ( n − 1 − k ) S
Proof (i) Consider the system:
C = kL2 + (n − 1 − k ) S 2 .
2e − kL
Substituting S = from the first equation into the second leads
n −1 − k
to ( n − 1) kL2 − 4ekL + [4e2 + C (k − n + 1)] = 0 and application of the
quadratic formula and simplification leads to the formula for L and S.
102 Spanning Tree Results for Graphs and Multigraphs
2e = kL + ( n − 1 − k ) S
(ii) Consider the system as before:
C = kL2 + (n − 1 − k ) S 2 .
∂S S−L
=
∂k 2(n − 1 − k )
Taking partials and solving yields
∂L S − L
= .
∂k 2k
Because there are two distinct entries, k cannot equal zero, and so 1 is
the smallest possible value for k. Given P = S n −1− k Lk , logarithmic
dP L L S
differentiation leads to the formula = P ln − + .
dk S 2S 2 L
L x 1
The substitution x = into f ( x ) = ln( x ) − + yields the quantity in
S 2 2x
∂P
the bracketed part of . Since f (1) = 0 and f ′ < 1 for x > 1, the
∂k
value within the bracket is always negative, and therefore, P is a strictly
decreasing function of k. Hence, the maximum value for P is attained
only when k = 1. □
n
1
Proposition 2.4 The optimal value of the objective function
n ∏λ
i =2
i is
(n − 1)(n − 1 − k ) 4e 2
2e + C −
k n −1
L= =
n −1
C (n − 1)( n − 1 − k ) 4e (n − 1 − k )
2
2e +
k k
.
n −1
2
2e C ( n − 1 − k ) 4e 2 ( n − 1 − k )
So, L − n − 1 = (n − 1)k − , and the partial
k (n − 1) 2
∂L ( n − 1)( n − 1 − k )
derivative with respect to C is = > 0. Similarly,
∂C 2k ( L(n − 1) − 2e)
∂L 4e 2 − C
= <0.
∂k 2 2e
( )
k ( n − 1) 2 L −
n − 1
104 Spanning Tree Results for Graphs and Multigraphs
(ii) The proof of the first part is the same as that for Proposition
∂L ( n − 1) ( n − 1 − k )
(3.3(ii)). For the second part, consider =
∂C 2k ( L(n − 1) − 2e )
and
∂S
=
( (n − 1)k ) . The partial derivative
∂C 2 ( n − 1 − k )( S ( n − 1) − 2e )
∂P dL dS
= Lk −1S n − k − 2 kS + L (n − 1 − k ) can be shown to be
∂C dC dC
negative, as the quantity in the brackets reduces to
S (n − 1) ( n − 1 − k )( S (n − 1) − 2e ) + Lk (n − 1) ( L(n − 1) − 2e )
, which has
2 ( L( n − 1) − 2e )( S ( n − 1) − 2e )
k
1. ∑x
i =1
i =I
there are only finitely many feasible solutions so that the minimum
k
∑x
i =1
2
i exists, any optimal solution can have at most two distinct entries.
Recall that trace of a matrix is the sum of its eigenvalues, and the
eigenvalues of the square of a matrix are the squares of the individual
eigenvalues. We use these observations and Lemma 2.6 to prove the next
two spanning tree results:
Theorem 2.7 The graph K n+ x , i.e. Kn with one edge doubled, has the
greatest number of spanning trees among all multigraphs in its class.
Proof First, we note that there are no graphs in this particular class. In
order to prove the theorem, we require the eigenvalues for the Laplacian
matrix of K n+ x .
106 Spanning Tree Results for Graphs and Multigraphs
λ = ( 0, n, n,..., n, n + 2 ) .
Proof of Claim: Without loss of generality assume the edge x is added
between v1 and v2 . The characteristic polynomial of
n + 1 −1 0 0 ...
−1 n + 1 0 0 ...
0 0 n 0 ...
(
B K n+ x ) = . . . . .
. . . . .
. . . . .
0 0 0 ... n
C = trace[( H ( K n+ x )) 2 ] = (n − 2) n 2 + ( n + 2)2 =
n 3 − 2 n 2 + n 2 + 4 n + 4 = n3 − n 2 + 4 n + 4 ,
4e 2 (n 2 − n + 2)2 4
it follows that = = n3 − n 2 + 4 n + <C. Hence,
n −1 n −1 n −1
by Proposition 2.3(ii), it follows that λ = (n + 2, n,..., n) is the location of
the solution to Problem Ps,C , p .
nn nn
AA
== + 1,
2e2=e =
22
(
+ 1,C C= =trace (
traceH 2 ( M( M ))
,and
) ) , and
and ΩΩn,n, + 1+ 1, , (i.e.,
MM∈ ∈
22
forcing a multigraph situation) the one with the one with the largest
Multigraphs with the Maximum Number of Spanning Trees: An Analytic Approach 107
(
M = K n+ x . Indeed, trace H 2 ( M ) ( ) ) = ∑ h + ∑ (h )
ii ij
2
. It is easy to see
i =1 i≠ j
n
that only the multigraph M ∈ Ω n, + 1 which has two distinct
2
values of hii that differ by one is M = K n+ x .
n
n
Now ∑h
i =1
ii = n2 − n + 2 for all M ∈ Ω n, + 1
2
and
∑ h = ∑ ( −h ) .
i =1
ii
i≠ j
ij Since K n+ x has two distinct values of hij , i ≠ j ,
i≠ j
ij
2
over all
n
M ∈ Ω n, + 1 . □
2
Proof As in Theorem 2.7, it suffices to show that trace(H ( K p , p )) 2 < trace(H ( M ))2
trace(H ( K ))2 < trace(H ( M ))2 for every multigraph M in its class. However, since
∑ ∑ ∑b
2
Consider trace( B (G ))2 = bii 2 + bij . Since ii = 2q 2 + 2q, and
i =1 i≠ j i =1
Ω(2q, q 2 ) .
Next, we claim that no other graph G in Ω (2q, q 2 ) can have the same
contradiction G = K q ,q follows. □
Remark 2.2 Cheng mentioned but did not prove uniqueness of the
eigenvalues for these results.
May 2, 2013 14:6 BC: 8831 - Probability and Statistical Theory PST˙ws
Threshold Graphs
∑ ς p (1 − p )i e −i
R (G ) = i . For small values of p, the reliability
i = n −1
polynomial will be dominated by the ς n−1 term, and since ς n−1 is the
number of spanning trees, graphs with a larger number of spanning
trees will have greater reliability for such p. Computing the ATR of a
probabilistic graph has been demonstrated to be NP-hard [Ball,
1982]. Thus, it is advantageous to bound from below the ATR of a
network represented by a graph in a particular class Ω ( n, e ) (i.e., all
graphs on n nodes and e edges). It has been conjectured that the best
such bound is provided by a threshold graph [Petingi, 1996]. If a
graph minimizes the number of spanning trees, it will minimize
reliability for small values of p. Let N ( u ) denote the deleted
neighborhood of node u, i.e. the set of all nodes adjacent to u. A
graph G is a threshold graph if, for all pairs of nodes u and v in G,
111
112 Spanning Tree Results for Graphs and Multigraphs
Figure 3.1: A threshold graph - K5 is the clique, and the two independent
nodes have degrees 2 and 3, respectively.
Using this convention, Figure 3.2 depicts the same threshold graph as
that given in Figure 3.1. We will use this depiction in the sequel.
Figure 3.2: A threshold graph - K5 is the clique (nodes in the rectangular box),
and the two independent nodes have degrees 2 and 3, respectively.
Because of the way the independent nodes and the edges incident on
them are depicted we refer to the independent set of nodes as cone
points.
In the following discussion, we prove the validity of a formula for
the characteristic polynomial of the Laplacian matrix of a threshold
graph in terms of its degree sequence. In preparation for the proof,
we require two lemmas. Recall PH ( M ) ( λ ) is the characteristic
polynomial of H ( M ) .
(λ ) = λ det λ n −1 − 11 G = λ PH11 (G ) (λ ) = PH (G − u ) (λ ) . □
114 Spanning Tree Results for Graphs and Multigraphs
PH (G −u ) (λ ) =
k k
d1 −1 d j +1 − d j
λ (λ − (n − 1)) ∏ (λ − ( d
i =1
i − 1)) ∏ (λ − (n − 1) + j)
j =1
so that
k k
d1 −1 d j +1 − d j
PH(G −u ) (λ − 1) = (λ − 1)(λ − n) ∏ (λ − d )∏ (λ − n + j)
i =1
i
j =1
.
so that PH(G −u ) (λ − 1) =
n k
d j +1 − d j
(λ − 1)ℓ +1 (λ − (n − ℓ)) dℓ+1 −1 ∏ ( λ − di ) ∏ (λ − (n − ℓ) + j − ℓ) .
i = ℓ +1 j = ℓ +1
k k
λ (λ − n)1 (λ − 1)ℓ (λ − (n − ℓ))dℓ+1 −1 ∏ ( λ − di ) ∏ (λ − (n − ℓ) + j − ℓ)d
i = ℓ +1 j = ℓ +1
j +1 − d j
.
k k
Observe that (λ − 1)ℓ ∏ ( λ − d i ) = ∏ (λ − d i ). Also, d j +1 − d j = 0 for
i = ℓ +1 i =1
j = 1,..., ℓ − 1 and dℓ +1 − dℓ = dℓ +1 − 1, so
k k
d j +1 − d j d j +1 − d j
(λ − (n − ℓ)) dℓ+1 −1 ∏ ( λ − ( n − ℓ) + j − ℓ) = ∏ (λ − n + j ) .
j = ℓ +1 j =1
k k
d d j +1 − d j
Hence, PH (G ) (λ ) = λ (λ − n) 1 ∏i =1
(λ − di ) ∏ (λ − n + j)
j =1
, the required
conclusion.
k n −k −2
PH ( G − u ) (λ ) = λ k PH (K n−( k +1) ) (λ ) = λ λ ( λ − (n − k − 1) ) .
k
d j +1 − d j
∏ (λ − n + j ) = (λ − n + k )dk +1 −1 = (λ − n + k )n − k −1−1 ,
j =1
Remark 3.1 The material concerning the above formula has been
submitted for publication.
118 Spanning Tree Results for Graphs and Multigraphs
Example 3.2 For the threshold graph in Figure 3.2, we reorder the
degree sequence d = 6,6,5,4,4,3,2 , i.e., d1 = 6, d 2 = 6, d 3 = 5, d 4 = 4,
λ4 = d 4* = 5, λ5 = d 5* = 3, λ6 = d6* = 2, and λ7 = d 7* = 0 .
(n ) (n ) (n )
(*) Let G be a connected threshold graph, and let d = (v1 1 , v2 2 ,..., vs s )
kˆ s −1
nkˆ +1 −1
t (G ) = ∏ (vi ) ni ∏ (v i + 1) ni ( vkˆ +1 + 1) (vs + 1) ns −1 ,
i =1 i = kˆ + 2
s − 1
where kˆ = [Hammer, 1996].
2
kˆ s
nkˆ+1 −1
PH (G ) (λ ) = ∏ (λ − vi ) ni ∏ (λ − v i − 1) ni (λ − vkˆ +1 − 1) =
i =1 i = kˆ + 2
2 6
ni ni
∏ (λ − v ) ∏ (λ − v − 1)
i i (λ − v2+1 − 1) n2+1 −1 =
i =1 i =4
s − 1
sequence, with kˆ = .
2
Threshold Graphs 121
kˆ s −1
nkˆ +1 −1
t (G ) = ∏ (vi ) ni ∏ (v i + 1) ni (vkˆ +1 + 1) (vs + 1) ns −1 .
i =1 i = kˆ + 2
k s −1
nkˆ+1 −1
t (G ) = ∏ (vi ) ni ∏ (v i + 1) ni (vkˆ +1 ) (vs + 1) ns −1.
i =1 i = kˆ + 2
k −1
t (Q ( n , k )) = qk n qk −1 ( n − k ) n − q1 − k −1 ∏ q (n − k + i)
i
qi − qi +1
i =1
[Bogdanowicz, 1985].
2 6
λ ∏ (λ − vi )n ∏ (λ − vi − 1)n (λ − v3 )n −1 =
i i 3
i =1 i =4
2
λ ( λ − 2 )( λ − 3)( λ − 6 )( λ − 7 )( λ − 8 ) ( λ − 4 ) .
i =1
k −1
= q3 8q3 −1 (8 − 3) 8 − q1 − 3 −1
∏ q (8 − 3 + i)
i
qi − qi +1
i =1
= 2 ⋅ 8 ⋅ 4 ⋅ 6 ⋅ 3 ⋅ 7 = 8064.
The “lollipop” graph, illustrated in Figure 3.4 has the same ATR and
number of spanning trees as the threshold graph depicted in Figure 3.5.
For our purposes here the placement of the bridges (a bridge is an edge
whose removal disconnects the graph) as either a path joined to the
remainder of the graph or single edges joined to the graph does not
matter as all those edges must appear in every connected spanning
subgraph, in particular every spanning tree.
conjectured to form a lower bound on the ATR for all connected graphs
in its class.
⋯ ⋯
Figure 3.4: Lollipop - Kn−k with 1 cone of degree c and a Pk −1 appended to the
cone.
⋯ ⋯
Figure 3.5: Threshold graph having Kn−k with 1 cone of degree c and
k − 1 cones of degree 1.
k −1 c 1 n − k −1− c
λ ( λ − n )( λ − 1) ( λ − c )( λ − n + (k − 1) ) − ( λ − n + k ) .
The formula follows by applying Theorem 1.5. □
( n − 1)( n − 2 ) n
+ 1 < e ≤ (**).
2 2
⋯ ⋯
m ({w, y} ) ≥ m ({ y, v} ) .
w y w y
v v
w y w y
u1 v u1 v
u2 u2
Proof (of Lemma 3.10) The proof proceeds by nested induction, first on
the number of nodes in M − w − v, i.e., the yi ' s , and then on the number
simple exercise to prove that u1′l1′ ≥ u1l1 when u1′ + l1′ = u1 + l1 and
Assume the lemma is true when there are yi ' s . To prove the lemma for
w
u1 u2 u3 u s h
y1 y2 y3 ….. y s
l1 l2 l3 ls
We now apply the factoring theorem to x1 and x1 . Their deletion makes
the sum of the edges incident on y1 equal to p in both graphs. Hence, by
x1 yields the multigraphs of the lemma, one with l1 h edges joining w
and v and s yi ' s , namely, y2 , y3 ,..., ys 1 in M | x1 and the second with the
conclude that the number of spanning trees of M̂ c with l1c l1 edges from
wc and vc removed is at least as great as t Mˆ . Thus, t (Mˆ c) t t (Mˆ ). If
w
u2 u3 us
y2 y3 ….. ys
h+ l1 l2 l3 ls
v
Figure 3.9: The multigraph M | x1 . Edge labels indicate the multiplicity
of each multiple edge.
wc
u2c u3c u sc
vc
Figure 3.10: The multigraph M c | x1c . Edge labels indicate the multiplicity
of each multiple edge.
132 Spanning Tree Results for Graphs and Multigraphs
1 if xi is contracted
ai 0 if xi is deleted
if x has become a loop .
i
t ( M ) t ( M | x1 ) t ( M x1 ).
x1 ,..., xk . We know t ( M ) = ∑ t (M
a∈ A
a ), t (M
′) = ∑ t (M ′ ) ,
a∈ A
a where A
( n − 1)( n − 2 )
e≥ + 1, that admits a unique balloon graph B and another
2
threshold graph T having at least two cone nodes. Denote by R(G) the all
terminal reliability of any graph. Then R ( B ) ≤ R (T ) for all values of p,
i.e., the balloon is a uniform lower bound on ATR for all threshold
graphs in its class. [Petingi, 1996]
∑d
i =1
i = 6 ≤ 6 = 1(1 − 1) + (1 + 1 + 1 + 1 + 1 + 1) ;
2
∑d
i =1
i = 11 ≤ 12 = 2 ( 2 − 1) + ( 2 + 2 + 2 + 2 + 2 ) ;
3
∑d
i =1
i = 16 ≤ 18 = 3 ( 3 − 1) + ( 3 + 3 + 3 + 3 ) ;
∑d
i =1
i = 21 ≤ 23 = 4 ( 4 − 1) + ( 4 + 4 + 3) ;
5
∑d
i =1
i = 25 ≤ 27 = 5 ( 5 − 1) + ( 4 + 3) ;
6
∑d
i =1
i = 29 ≤ 33 = 6 ( 6 − 1) + ( 3) ;
136 Spanning Tree Results for Graphs and Multigraphs
∑d
i =1
i = 7 ≤ 6 = 1(1 − 1) + (1 + 1 + 1 + 1 + 1 + 1) ;
this sequence fails the first inequality and thus is not graphical.
Figure 3.11: A split graph. We denote this graph 1-IPS(2,3,0), whose clique is of
order 6, and observe that both sides of the 6th Erdos and Gallai inequality are
equal to 36.
Threshold Graphs 137
1
( cd + b ) ( cd + b + 1) ( ) W1 ⋅ W2 ⋅ Z1 ⋅ Z 2 , where W1 =
b −1 c d −1
n
1
2 ( 2 2
( cd + b ) + d + 1 + ( cd + b ) + 2 ( cd + b )( d + 1) + ( d + 1) − 4dn , )
1
( 2 2
W2 = ( cd + b ) + d + 1 − ( cd + b ) + 2 ( cd + b )( d + 1) + ( d + 1) − 4dn ,
2 )
1
( 2
Z1 = ( cd + b ) + d + 1 + ( cd + b ) − 2 ( cd + b )( d − 1) + ( d + 1) ,
2
2
)
and Z 2 =
1
2((cd + b) + d + 1 − (cd + b) − 2 (cd + b)( d − 1) + (d + 1) ).
2 2
(Figure 3.12)
(3) [Fuller, 2014] If x >1 and b = 0, then t(x-IPS(c,d,0) =
c
( d −1) 1
1 cd x c 1−
x
c
−1
c
−1
( x + d ) x + d ( P + Q) x ( P − Q) x ,
cd x
c+
x
2
1 cd 1 cd cd 2
where P = x + d + , and Q = x+d + −4 .
2 x 2 x x
(Figure 3.13)
Proof Parts (1) and (2) of Theorem 3.14 were proven in the referenced
publications, while part (3) has been submitted for publication. All of the
proofs follow from elementary methods involving matrix row operations.
To give a flavor for these proofs, we provide a sketch of the proof of (2).
Consider the Temperley’s B-Matrix, B, as defined in Chapter 1. Based
on the structure of the x-IPS(c, d, b) graphs when b > 0, we choose
to arrange the matrix such that the nodes representing each row
(column) are in decreasing order of degree. Hence, B − λI is of the
Q Q4
form 1 . Submatrix Q 1 is a diagonal ( cd + b ) × ( cd + b )
Q 2 Q 3
Threshold Graphs 139
cd + b + 1 − λ 0 0 ⋯ 0 0
0 ⋱ 0 ⋯ 0 0
0 0 cd + b + 1 − λ 0 ⋮ ⋮
Q1 = .
⋮ ⋮ 0 cd + b − λ 0 ⋮
0 0 ⋯ 0 ⋱ 0
0 ⋯ ⋯ ⋯ 0 cd + b − λ
0 0 ⋯0 1 1 ⋯1 ⋯ 1 1 ⋯1 1 ⋯ 1
1 0 0 ⋯ 0 1 1 ⋯1 1 1 ⋯1 ⋮ 1 ⋮
Q2 = .
⋮ 1 ⋱ 1 1 ⋯1 1 ⋮ 1
1 ⋯ 1 0 0 ⋯ 0 1 1 1
d + 1 − λ 1 ⋯ 1
1 d +1− λ 1 ⋮
Q3 = .
⋮ 1 ⋱ 1
1 1 1 d +1− λ
140 Spanning Tree Results for Graphs and Multigraphs
Finally, Q4 = Q2T .
1
cd + b + 1 − λ by − and adding to the appropriate rows of
cd + b + 1 − λ
Q 2 . Then we proceed to the rightmost c × b part of Q 2 , multiplying the
1
rows of Q 1 that have diagonal entry cd + b − λ by − and
cd + b − λ
adding to the appropriate rows of Q 2 . After this is completed, the matrix
has Q1 and Q 4 unchanged, the matrix Q 2 is all zeros, and Q 3 has all
main diagonal entries the same, say, A, and all off- diagonal the terms the
same, say, B, where
d ( c − 1) b
A = (d +1− λ ) − − , and
( cd + b + 1) − λ ( cd + b ) − λ
d (c − 2) b
B =1− − .
( + + ) − ( + b) − λ
cd b 1 λ cd
A B B ⋯ B
B A B ⋯ B
Since Q 3 = B B ⋱ B B , the algorithmic application of row
⋮ ⋮ B A B
B B ⋯ B A
cd b ( A + B )( A − B ) ( A + 2 B )( A − B )
( cd + b + 1 − λ ) ( cd + b − λ ) ( A )
A ( A + B)
( A + 3B )( A − B ) A + ( c − 2 ) B ( A − B ) A + ( c − 1) B ( A − B )
( ) ( )
⋯ .
( A + 2B ) (
A + ( c − 3) B
)
A + (c − 2) B(
)
After cancellation, we have
c −1
( cd + b + 1 − λ ) ( ) ( cd + b − λ ) ( λ 2 − ( cd + b + 1) λ + d ( cd + b ) )
c d −1 b −1
( λ 2 − ( cd + b + d + 1) λ + d ( cd + b + c ) ) ( cd + b + c − λ ) .
The roots of this polynomial are multiplied together, after dividing out
the n 2 (as in Theorem 1.13), t(1-IPS(c,d,b))=
1 b −1 c ( d −1)
× ( cd + b ) × ( cd + b + 1) × W1 × W2 × Z1 × Z 2 ,
n
1
( 2 2
where W1 = ( cd + b ) + d + 1 + ( cd + b ) + 2 ( cd + b )( d + 1) + ( d + 1) − 4dn ,
2 )
(
1 2 2
W2 = ( cd + b ) + d + 1 − ( cd + b ) + 2 ( cd + b )( d + 1) + ( d + 1) − 4dn ,
2 )
1
( 2 2
Z1 = ( cd + b ) + d + 1 + ( cd + b ) − 2 ( cd + b )( d − 1) + ( d + 1) ,
2 )
1
( 2 2
and Z 2 = ( cd + b ) + d + 1 − ( cd + b ) − 2 ( cd + b )( d − 1) + ( d + 1) .
2
□ ) □
1 n−2c
( d + 1)( n − c + 1) i
n
c-1
[d (c + 1) + 1] + (c − 1) 2 d 2 + 2(c + 1)d + 1]
i
2
c −1
[d (c + 1) + 1] − (c − 1) 2 d 2 + 2(c + 1)d + 1]
.
2
For 1-IPS(2,3,0) , c = 2 , d = 3 and n = 8, so
1
( 3 + 1)( 8 − 2 + 1) − ( ) i
8 2 2
t (1 − IPS (2,3,0) ) =
8
2-1
[3(2 + 1) + 1] + (2 − 1)2 32 + 2(2 + 1)3 + 1]
i
2
2-1
[3(2 + 1) + 1] − (2 − 1)2 32 + 2(2 + 1)3 + 1]
= 21609.
2
One Laplacian matrix for 1-IPS(2,3,0) is
6 −1 −1 −1 −1 −1 −1 0
−1 6 −1 −1 −1 −1 −1 0
−1 −1 6 −1 −1 −1 −1 0
−1 −1 −1 6 −1 −1 0 −1
−1 −1 −1 −1 6 −1 0 −1
−1 −1 −1 −1 −1 6 0 −1
−1 −1 −1 0 0 0 3 0
0 0 0 −1 −1 −1 0 3
which has (1,1)-cofactor 21609 as well, confirming the result of the
formula.
By Theorem 3.14(2),
1
t (1 − IPS ( c, d , b )) = (cd + b)b −1 (cd + b + 1)c(d −1) i
n
1
2 (
(cd + b) + d + 1 + (cd + b) + 2 (cd + b)( d + 1) + ( d + 1) − 4dn )i
2 2
2(
(cd + b) + d + 1 − (cd + b) + 2 (cd + b)( d + 1) + ( d + 1) − 4dn )i
1 2 2
2(
(cd + b) + d + 1 + (cd + b) − 2 (cd + b)( d − 1) + ( d + 1) )i
1 2 2
2(
(cd + b ) + d + 1 − (cd + b ) − 2 (cd + b)( d − 1) + ( d + 1) ).
1 2 2
1 1−1 2( 3−1)
t (1 − IPS (2,3,1) ) =
9
( ( 2 )( 3) + 1) ( ( 2 )( 3) + 1 + 1) i
1 2 2
11 +
2
( ( 2 )( 3) + 1) + 2 ( ( 2 )( 3) + 1) ( 3 + 1) + ( 3 + 1) − 4 ( 3)( 9 ) i
1 2 2
11 −
2
( ( 2 )( 3) + 1) + 2 ( ( 2 )( 3) + 1) ( 3 + 1) + ( 3 + 1) − 4 ( 3)( 9 ) i
1 2 2
11 +
2
( ( 2 )( 3) + 1) − 2 ( ( 2 )( 3) + 1) ( 3 − 1) + ( 3 + 1) i
1 2 2
2
11 − ( ( 2 )( 3) + 1) − 2 ( ( 2 )( 3 ) + 1) ( 3 − 1) + ( 3 + 1) =
1 4 11 + 13 11 − 13 11 + 37 11 − 37
⋅8 ⋅ ⋅ ⋅ ⋅ = 258048.
9 2 2 2 2
7 −1 −1 −1 −1 −1 −1 −1 0
−1 7 −1 −1 −1 −1 −1 −1 0
−1 −1 7 −1 −1 −1 −1 −1 0
−1 −1 −1 7 −1 −1 −1 0 −1
−1 −1 −1 −1 7 −1 −1 0 −1
−1 −1 −1 −1 −1 7 −1 0 −1
−1 −1 −1 −1 −1 −1 6 0 0
−1 −1 −1 0 0 0 0 3 0
0 3
0 0 −1 −1 −1 0 0
which has (1,1)-cofactor 258048 as well, confirming the result of the
formula.
4
( 3−1) 1
1 ( 4 )( 3) 2 4 1−
⋅ ( 2 + 3) 2 + ⋅3 2
i
4+
( 4 )( 3 ) 2
2
4
−1
2 2 2
1 2 + 3 + ( 4 )( 3 ) + 1 2 + 3 + ( 4 )( 3) − 4 ( 4 )( 3) i
2 2 2
2
2
4
−1
2 2 2
1 2 + 3 + ( 4 )( 3 ) − 1 2 + 3 + ( 4 )( 3 ) − 4 ( 4 )( 3 ) =
2 2 2
2
2
331776.
7 −1 −1 −1 −1 −1 −1 −1 0 0
−1 7 −1 −1 −1 −1 −1 −1 0 0
−1 −1 7 −1 −1 −1 −1 −1 0 0
−1 −1 −1 7 −1 −1 0 0 −1 −1
−1 −1 −1 −1 7 −1 0 0 −1 −1
−1 −1 −1 −1 −1 7 0 0 −1 −1
−1 −1 −1 0 0 0 3 0 0 0
−1 −1 −1 0 0 0 0 3 0 0
0 0 0 −1 −1 −1 0 0 3 0
0 0 0 −1 −1 −1 0 0 0 3
147
148 Spanning Tree Results for Graphs and Multigraphs
The end nodes of the lone double edge can exist in any of the following
situations:
1. exactly one end node is a fnp;
2. both end nodes are fnp's;
3. neither end node is a fnp.
Case 1: The multiple edge joins a fnp w and a point v which is not a fnp.
Case 3: The multiple edge is incident on two nodes each not of full
neighborhood.
Case 3 is much more involved, but is solved for many instances of
multigraphs in the class. To get a sense of the problem, we require some
further algebraic development. Given two vectors of nonnegative
integers x and y , both of which are of length n, and whose elements are
k k
nonincreasing, x is majorized by y , denoted xx y , if
i 1
xi y
i 1
i
n n
for each 1 k n 1 and i 1
xi y
i 1
i [Marshall, 1979]. In other words,
Example 4.1 Consider the following three vectors x = (15,10,10,5,3,3) , y = 15,12,11, 4,3,1 and z = 1
x = 15,10,10,5,3,3 , y = (15,12,11, 4,3,1) and z = (16,12,7, 4,4,3) . The table below lists the
partial sums.
k
x = (15,10,10,5,3,3) y = (15,12,11,4,3,1) z = (16,12,7, 4, 4,3)
∑i=1
k=1 15 15 16
k=2 25 27 28
k=3 35 38 35
k=4 40 42 39
k=5 43 45 43
k=6 46 46 46
k k 6 6
Clearly, x ≺ y , as ∑
i =1
xi ≤ ∑ i =1
yi for all k ≤ 5 , and ∑
i =1
xi = ∑y.
i =1
i As
6 6
indicated ∏x
i =1
i = 67500 ≥ 23760 = ∏y .
i =1
i We note that there is no
∑
i =1
xi = 25 < 28 = ∑
i =1
zi , but ∑
i =1
xi = 40 > 39 = ∑z
i =1
i .
Approaches to the Multigraph Problem 151
ˆ =
A y
n × n matrix A T . In addition, let λ1 ≤ λ2 ≤ ... ≤ λn −1 represent
y a
the eigenvalues of A and λˆ1 ≤ λˆ2 ≤ ... ≤ λˆn represent the eigenvalues of
152 Spanning Tree Results for Graphs and Multigraphs
ˆ .
interlace those of A
x
arbitrary ( n − j + 1) -dimensional subspace of ℝ n and
and xˆ = . By
z
Theorem 0.5 (Courant-Fischer),
ˆˆ
xˆ T Ax ˆˆ
xˆ T Ax xT Ax
λˆ j = max min ≤ max min = max min = λ j −1.
Lj 0 ≠ x∈L j xˆ T xˆ Ln − j +1 0 ≠ x∈Ln− j +1 xˆ T xˆ Ln − j +1 0 ≠ x∈Ln − j +1 x T x
of A , λ = ( λn , λn −1 ,..., λ1 ) , i.e., d ≻ λ .
j j
we have ∑
i =1
dn−i ≥ ∑ λˆ
i =1
n −i , j = 1,..., n − 1. By Lemma 4.6, we have
j j
λ1 ≤ λˆ1 ≤ λ2 ≤ λˆ2 ... ≤ λn −1 ≤ λˆn −1 ≤ λn , so ∑ λˆn −i ≥ ∑ λn −i , j = 1,..., n − 1.
i =1 i =1
j j
Therefore, ∑i =1
d n −i +1 ≥ ∑λ
i =1
n −i +1 , j = 1,..., n − 1, and, by Theorem 0.11,
n n
have ∑ d = ∑ λ , i.e., the trace is the sum of the eigenvalues.
i i □
i =1 i =1
∏i =1
di (B( M )) ≥ ∏ λ (B(M )) and
i =1
i
n n
∏ ( i =1
di B ( K n − kK 2 ) ≥ ) ∏λ (B(K
i =1
i n − kK 2 ) . )
Let M be such that
d ( B ( M ) ) ≻ λ B ( K n − kK 2 ) .
( )
Then
n n
∏
i =1
di (B(M )) ≤ ∏ λ (B ( K
i =1
i n − kK 2 ))
and thus such a multigraph has fewer spanning trees. Using majorization,
many multigraphs can be shown to have fewer spanning trees than
K n − kK 2 in their respective classes.
∑ d i ( B ( M ) ) > k (n − 2) = ∑ λi ( B ( K n − kK 2 ) ) .
i = n − k +1 i = n − k +1
i.e., the sum of the last k elements of the vector of diagonals of the
multigraph’s B-matrix is greater than the sum of the corresponding
elements of the vector of K n − kK 2 B-matrix eigenvalues.
matrix of the multigraph and let λi refer to the eigenvalues for the unique
n n
K n − kK 2 in its class. If ∑
i = n − k +1
d i > k ( n − 2) = ∑
i = n − k +1
λi , then
Approaches to the Multigraph Problem 155
n n n−k n−k n n
∑
i =1
di − ∑
i = n − k +1
di = ∑
i =1
di < ∑ λi = ∑ λi −
i =1 i =1
∑
i = n − k +1
λi .
j −1 j −1
smallest such index. Then d j < λ j , or else ∑ di < ∑ λi , which would
i =1 i =1
and so,
n j n j n n
∑ d = ∑ d + ∑ d <∑ λ + ∑ λ = ∑ λ ,
i =1
i
i =1
i
i = j +1
i
i =1
i
i = j +1
i
i =1
i
∑
i = j +1
di > ∑
i = j +1
λi ,with ∑
i = j +1
di < ∑λ ,
i = j +1
i
it follows that
n n
∑ di < ∑ λi = k (n − 2) . □
i = n − k +1 i = n − k +1
Corollary 4.4 If the vector of diagonals does not majorize the vector of
eigenvalues, then d1 ≤ 2n − k − 2, and d n ≥ n − k .
so that
j j
∑ i =1
di ≥ d1 + ( j − 1)(n − 1) = d1 + jn − j − n + 1 and ∑λ
i =1
i = jn .
j j
Therefore, since ∑i =1
di < ∑λ
i =1
i , we have
∑
i = n − k +1
d i > k ( n − 2) = ∑
i = n − k +1
λi , and d n − k +1 ≤ n − 1 .
Thus,
n
∑
i = n − k +1
d i < (k − 1)(n − 1) + d n so (k − 1)(n − 1) + d n > k (n − 2)
implying d n > n − k − 1 . □
159
160 Spanning Tree Results for Graphs and Multigraphs
n
Proof Obviously, k ≤ . Now, K n − kK 2 has as its complement k
2
n
copies of K 2 , and if k < , n − 2k copies of K1 . Thus, the
2
eigenvalues of the complement of K n − kK 2 are n − k zeros, and k
Remark 5.1 The proofs of the spanning tree formulas for complete
bipartite and regular complete k-partite graphs rely heavily on the fact
that the complements of these graphs are disjoint complete graphs. Thus,
these graphs are also Laplacian integral.
part (c)(ii); while C3 ,C4 , and C6 all have integer eigenvalues, this is
not the case for all cycles. For example, it can be shown by direct
application of Theorem 1.5 that C8 has eigenvalue spectrum
c
( d −1) 1
1 cd x c 1−
x
t ( x − IPS (c, d ,0) = ( x + d ) x + d
x
i
cd
c+
x
c
−1
2 x
1 x + d + cd + 1 x + d + cd − 4 cd
2
i
2 x 2 x x
162 Spanning Tree Results for Graphs and Multigraphs
c
−1
1 2 2 x
x + d + cd − 1 x + d + cd − 4 cd .
2 x 2 x x
The characteristic polynomial giving the result is
c
( d −1)
( λ − ( x + d ) ) cdx + x − λ
c x
c−
λ (d − λ ) x i
c
−1
2 cd cd 2 x
λ − x + d + λ − ,
x x
which has roots
cd c
λ = 0, ( x + d ) , x + with multiplicity ( d − 1) , and d ,
x x
1 2
cd 2
cd 1 cd
x + d + + x + d + − 4
2 x 2 x x
and
1 2
cd 2
cd 1 cd
x + d + − x + d + − 4
2 x 2 x x
1
each with multiplicity c 1 − .
x
We note that, if the discriminant is an odd perfect square, the graph
will be Laplacian Integral. For example, the graph 3-IPS(6,2,0) has
Laplacian spectrum 0,2,2,2,2,5,7,7, obtained by substituting x = 3,
x c==3,6c = and
6 and d = 2 . Some other examples of Laplacian integral
x-IPS(c,d,0) graphs are 3-IPS(12,2,0), and 2-IPS(4,3,0) . [Fuller, 2010]
a split graph, but can be transformed to one by the deletion of one edge.
We will denote by AS(n,2) an almost split graph on n nodes, with a
clique of order n − 2 , two adjacent nodes that are also adjacent to
roughly half the clique nodes, with one clique node in their common
neighborhood. The degrees of the non-clique nodes will be the same in
n +1
the case where n is odd (i.e., their degrees will each be ), and if n
2
is even, the degrees of these nodes will differ by one (i.e., one will be
n n
of degree and the other will be of degree + 1 ). When n is odd,
2 2
n +1 n +1
AS(n,2) has degree sequence , , n − 2,..., n − 2, n − 1. When n is
2 2 n 3 terms
−
n n
even, AS(n,2) has degree sequence , + 1, n − 2,..., n − 2, n − 1. It is
2 2 n 3 terms
−
n2 − 3n + 6
easy to show that such graphs will always have edges,
2
regardless of the parity of n. Figures 5.1 and 5.2 depict two examples of
AS(n,2).
Recall that a split graph satisfies the mth Erdos and Gallai inequality
with equality [Hammer, 1981], where m represents the largest index i in
a degree sequence such that d i ≥ i − 1. For the graph in Figure 5.1 (a),
the degree sequence is 10, 9, 9, 9, 9, 9, 9, 9, 9, 6, 6; m = 9, and
9
∑d
i =1
i = 82 ≤ 84 = 9 ( 9 − 1) + ( 6 + 6 ) , so the m = 9th Erdos and Gallai
n +1 n +1
0, , , n − 1,..., n − 1, n, n, all of which are integers.
2 2 n 5 terms
−
n n
(b) The eigenvalues for AS ( n, 2 ) , n even, are 0, , + 1, n − 1,..., n − 1, n, n,
2 2 n 5 terms −
n−3
0, 1,...1, + 1. Thus, the eigenvalues for these three
n −3
2
-1 terms
2
n +1 n +1
0, , , n − 1,..., n − 1, n, n.
2 2 n 5 terms −
(b) The proof for n even follows similarly. The only difference is
that, instead of two identical stars, the two stars in the
complement are K n −2 and K n−4 .
1, 1,
2 2
2
n +1 n −5
Remark 5.2 The graph AS ( n, 2 ) , n even, has ( n − 1) n
2
2
n n−5
spanning trees, while AS ( n, 2 ) , n odd, has ( n − 1) n spanning
2
trees.
3
Figure 5.3: The multigraph MT7;2 2 which has degree sequence 2, 2, 4, 4, 4, 8, 8.
eigenvalues as follows:
(a) (even case: v1 = n − n1 − 1 )
v1n1 , v2 n2 .
Laplacian Integral Graphs and Multigraphs 167
v1n1 , v2 n2 , v3n3 .
The proof of Proposition 5.3 follows from a sequence of elementary
matrix row operations, similar to the proof of Theorem 3.14(2).
n2 −1 n3 −1
v1n1 −1 ( v2 + 1) ( v3 + µ ) spanning trees, while MT µ n;v n1 , in the odd
1
n2 −1
case, has v1n1 −1 ( v2 + µ ) spanning trees.
May 2, 2013 14:6 BC: 8831 - Probability and Statistical Theory PST˙ws
[Ball, 1982] M.O. Ball and S.J. Provan, Bounds on the reliability polynomial for
shellable independence systems, SIAM J. Alg. Disc. Meth. 3, 166-181, 1982.
[Ball, 1983] M. Ball and J.S. Provan, The complexity of counting cuts and of computing
the probability that a graph is connected, SIAM J. on Computing 12, 777-788,1983.
[Bapat, 2010] R. B. Bapat. Graphs and Matrices. Springer: London, Hindustan Book
Agency, New Dehli, India, 2010.
[Bleiler, 2007] S. Bleiler, and J.T. Saccoman. A Correction in the Formula for the
Number of Spanning Trees in Threshold Graphs. Australasian Journal of
Combinatorics 37, 205-213, 2007.
[Bogdanowitz, 1985] Z. Bogdanowicz, Spanning trees in undirected simple graphs,
Doctoral Dissertation, Stevens Tech, Computer Science Department, 1985.
[Bogdanowitz, 2009] Z. Bogdanowicz, Undirected simple connected graphs with
minimum number of spanning trees, Discrete Math 309 (no. 10), 3074-3082, 2009.
[Boesch, 1984] F.T. Boesch and C.L. Suffel, A survey of the algebraic approach to the
study of spanning trees, Technical Report, Stevens Tech, Computer Science
Department, 8401, 1984.
[Brooks, 1940] R.L. Brooks, C.A.B. Smith, A.H. Stone and W.T. Tutte, Dissection of a
Rectangle into Squares, Duke Mathematics Journal 7, 312-340, 1940.
[Brown, 1993]. J.J. Brown, C. Colbourn, and J.S. Devitt, Network transformations and
bounding network reliability, Networks 23, 1- 17, 1993.
[Cayley, 1889] A. Cayley, A theorem on trees, Quart. J. Math 23,376-378: 69, 1889.
[Cheng, 1981] C. Cheng, Maximizing the total number of spanning trees in a graph: two
related problems in graph theory and optimum design theory, Journal of
combinatorial Theory B 31,1981.
[Chartrand, 2011] G. Chartrand, L. Lesniak, and P. Zhang, Graphs and Digraphs (5th ed),
CRC Press, Boca Raton, FL, 2011.
169
170 Bibliography
[Erdos, 1960] P. Erdos and T. Gallai, Graphs with prescribed degrees of vertices
(Hungarian), Mat. Lapok 11, 264-274, 1960.
[Fiedler, 1973] M. Fieldler, Algebraic Connectivity of Graphs, Czech. Math Journal 23,
298-305, 1973.
[Fuller, 2010] J. Fuller. Spanning trees of extended ideal proper split graphs. Honors
Thesis, Seton Hall University, Department of Mathematics and Computer Science,
2010.
[Hakimi, 1962] Hakimi, S. L. On realizability of a set of integers as degrees of the
vertices of a linear graph. I. J. Soc. Indust. Appl. Math. 10, 496–506, 1962.
[Hammer, 1978] P.L. Hammer, T. Ibaraki, and B. Simeone. Degree sequences of
threshold graphs. Congressus Numeratium, Vol. 21, 329-355, 1978.
[Hammer, 1981] P.L. Hammer and B. Simeone. The splittance of a graph. Combinatoria
Vol. 1, 275-284, 1981.
[Heinig, 2012] M. Heinig and J.T. Saccoman, Laplacian integral multigraphs,
Congressus Numerantium 212, 131-143, 2012.
[Kelmans, 1967] A.K. Kelmans, Properties of characteristic polynomial of a graph,
Kibernetiku-na sluzbu kommunizmu 4, Energija, Moskva-Leningrad, 27-41, 1962.
[Kelmans, 1974] A.K. Kelmans and V. M. Chelnokov, A certain polynomial of graph
and graphs with an extremal number of trees, Journal of Combinatorial Theory B 16,
197-214, 1974.
[Kelmans 1981]. On graphs with randomly deleted edges, Acta Math Acad. Sci Hungary
37, 77-88, 1981.
[Kirchhoff, 1847] Über die ausflösung der gleichungen, auf welche man bei der
untersuchung der linearen vertheilung galvanischer ströme geführt wird. (“On the
solution of the equations obtained from the investigation of the linear distribution of
galvanic currents”). Annalen der Physik und Chemie 72,497-508, 1847.
[Lancaster, 1985] P. Lancaster and M. Tismenetsky, Theory of Matrices, Academic
Press, Orlando, FL, 1985.
[Marshall, 1979]. A.W. Marshall and L. Olkin, Inequalities: The Theory of Majorization
and its Applications, Academic Press, New York, 1979.
[Merris, 1994] R. Merris, Degree maximal graphs are Laplacian integral,
Linear Algebra Appl. 199, 381–389, 1994.
[Michewicz, 2007] J. Michewicz and J.T. Saccoman. A formula for the number of
spanning trees of certain non-threshold split graphs. Congressus Numerantium, Vol.
184, 85-96, 2007.
Bibliography 171
[Moore, 2013] K. Moore and J.T. Saccoman. Spanning trees of almost ideal proper split
graphs. Congressus Numerantium, Vol. 216, 129-141, 2013.
[Moskowitz, 1958] F. Moskowitz, The analysis of redundancy networks, AIEE
Transactions on Commumication Electronics 39, 627-632, 1958.
[Petingi, 1996] L. Petingi, J.T. Saccoman. and L. Schoppmann, Uniformly least reliable
graphs, Networks 27(2), 125-131, 1996.
[Seshu, 1961] S. Seshu and M. Reed, Linear Graphs and Electrical Networks, Addison-
Wesley, Reading, MA, 1961.
[Shier, 1974] D. Shier, Maximizing the spanning trees in a graph with n nodes with n
nodes and m edges, Journal of Research of the National Bureau of Standards-B 78B,
193-196, 1974.
[Schoppmann, 1990] L. Schoppmann, Network bounds for edge reliability, Doctoral
dissertation, Stevens Tech, Mathematics Department, 1990.
[Satyanarayana, 1992] A. Satyanarayana, L. Schoppmann, and C.L. Suffel, A reliabilty-
improving graph transformation with applications to network reliability, Networks 22,
209-216, 1992.
[Temperley, 1964] H.N.V. Temperley, On the mutual cancellation of cluster integrals in
Mayer’s fugacity series, Proc. Phys. Soc. 83, 3–16, 1964.
May 2, 2013 14:6 BC: 8831 - Probability and Statistical Theory PST˙ws
Acylic, 6 graph, 10, 11, 52, 57, 58, 59, 66, 77,
Addition of edges, 15-16 91
Adjacency matrix, 4, 41 k-partite, 11, 59, 69, 160
Adjacent Component, 6
from (a node in a directed graph), 8 Cone points, 113, 114, 115, 121, 122,
to (a node in a directed graph), 8 136, 140, 166
Adjacent Connected multigraph, 6, 7
edges, 4 Contraction, 17, 127, 129, 130
nodes, 4 Cospectral, 90
Adjugate matrix, 44 Courant-Fischer Max-Min Theorem,
Algebraic complement, 77 26-27
All Terminal Reliability (ATR), 37, Cycle, 6, 40, 58, 64, 71, 160
111, 126, 134
Arc, 8 δ (G ) (minimum degree of a graph
Arc-set, 8
G), 46, 48
Balloon graph, 125, 134 ∆ (G ) (maximum degree of a graph
Basis, 24 G), 81, 82, 83, 87
Binet-Cauchy Theorem, 21-23 Degree of a node, 4, 5, 43, 83, 87, 90
Bipartite graph, 11 Degree matrix, 43
Degree sequence, 113, 115, 118, 119,
Cauchy-Schwarz Inequality, 97 136
Cayley’s theorem, 50 Deleted neighborhood, 111
Characteristic polynomial, 48, 56, 58, Deletion
63, 72, 74, 81, 82, 106, 111-115 of nodes, 13-14, 57
Chebyshev polynomial, 63 of edges, 14, 15, 65
Circulant
graph, 12, 53-57, 66-68 Determinant, 20, 21, 22, 30, 40, 41, 49,
matrix, 54 62, 63, 64
Class, 3 Dimension of a subspace (dim), 23, 25,
Clique, 10 27
Cofactor, 44-46, 72, 75 Directed cycle, 38
Complement, 13, 46, 56-57, 59, 74, 77, Directed Graph (digraph), 7
78, 134, 54, 55, 72, 75 Directed multigraph, 7
Complete Disconnected multigraph, 6, 46, 47, 50,
bipartite graph, 11, 57, 58, 64, 67, 78, 85
68, 108 Dominant Matrix Theorem, 30-31
173
174 Index