Professional Documents
Culture Documents
Textbook Regular Graphs A Spectral Approach 1St Edition Zoran Stanic Ebook All Chapter PDF
Textbook Regular Graphs A Spectral Approach 1St Edition Zoran Stanic Ebook All Chapter PDF
Textbook Regular Graphs A Spectral Approach 1St Edition Zoran Stanic Ebook All Chapter PDF
https://textbookfull.com/product/regular-graphs-a-spectral-
approach-1st-edition-zoran-stanic-2/
https://textbookfull.com/product/infrared-spectral-
interpretation-a-systematic-approach-first-edition-smith/
https://textbookfull.com/product/clinical-imaging-of-spinal-
trauma-a-case-based-approach-1st-edition-zoran-rumboldt-editor/
https://textbookfull.com/product/probability-on-graphs-random-
processes-on-graphs-and-lattices-second-edition-grimmett/
JavaScript Regular Expressions 1st Edition Groner
Loiane Manricks Gabriel
https://textbookfull.com/product/javascript-regular-
expressions-1st-edition-groner-loiane-manricks-gabriel/
https://textbookfull.com/product/probability-on-graphs-random-
processes-on-graphs-and-lattices-geoffrey-grimmett/
https://textbookfull.com/product/completely-regular-semigroup-
varieties-a-comprehensive-study-with-modern-insights-2024th-
edition-petrich/
https://textbookfull.com/product/generating-random-networks-and-
graphs-1st-edition-annibale/
https://textbookfull.com/product/linear-discrete-time-systems-
zoran-m-buchevats/
Zoran Stanić
Regular Graphs
De Gruyter Series in Discrete
Mathematics and Applications
Edited by
Michael Drmota, Vienna, Austria
János Pach, Lausanne, Switzerland
Martin Skutella, Berlin, Germany
Volume 4
Zoran Stanić
Regular Graphs
A Spectral Approach
Mathematics Subject Classification 2010
05C50, 05C12, 05C76, 05C81, 05B05, 05B20, 65F15, 68R10, 94B05
Author
Prof. Dr. Zoran Stanić
University of Belgrade
Faculty of Mathematics
Studentski trg 16
11000 Belgrade
Serbia
zstanic@math.rs
ISBN 978-3-11-035128-6
e-ISBN (PDF) 978-3-11-035134-7
e-ISBN (EPUB) 978-3-11-038336-2
Set-ISBN 978-3-11-035135-4
ISSN 2195-5557
www.degruyter.com
Preface
This book has been written to be of use to scientists working in the theory of graph
spectra. It may also attract the attention of graduate students dealing with the same
subject area as well as all those who use graph spectra in their research, in particular,
computer scientists, chemists, physicists or electrical engineers.
In the entire book we investigate exactly one class of graphs (regular) by using
exactly one approach (spectral), that is we study eigenvalues and eigenvectors of a
graph matrix and their interconnections with the structure and other invariants. Reg-
ular graphs appear in numerous sources. In particular, they can probably be encoun-
tered in all the books concerning graph theory. An intriguing phenomenon is that, in
many situations, regular graphs are met as extremal, exceptional or unique graphs
with a given property. More tangibly, an inequality is attained for a regular graph or
a claim holds unless a graph under consideration is regular or it does hold exactly
for regular graphs. In other words, depending on a given problem they fluctuate from
one amplitude to another, which is making them probably the most investigated and
the most important class of graphs, and simultaneously motivating us to consider the
theory of graph spectra through the prism of regular graphs. There is also a relevance
of regular graphs in the study of other discrete structures such as linear spaces or
block designs. A final motivation for this book is a remarkable significance of regu-
lar graphs in branches of computer science, chemistry, physics and other disciplines
that consider processes dealing with graph-like objects with high regularity and many
symmetries.
The rapid development of graph theory in last decades caused an inability to cover
all relevant results even if we restrict ourselves to a singular approach, so we must
say that this material represents a narrowed selection of (in most cases, known) re-
sults. Our intention was to include classical results concerning the spectra of regular
graphs along with recent developments. Occasionally, possible applications are indi-
cated. The terminology and notation are taken from our previous book [290]. This book
was written to be as self-contained as possible, but we assume a decent mathematical
knowledge, especially in algebra and the theory of graph spectra.
Here is a brief outline of the contents. Chapter 1 is preparatory. In Chapter 2 we fo-
cus on basic properties of the spectra of regular graphs. In Chapter 3 we pay attention
to some particular types of regular graph. Graph products, walk-regular graphs, differ-
ent subclasses of distance-regular graphs, regular graphs of line systems and various
block designs occupy the central place of this chapter. In Chapter 4 we consider the
problem of determining regular graphs that satisfy fixed spectral constraints. There,
we mostly deal with regular graphs with bounded least or second largest eigenvalue,
those with integral spectrum or a comparatively small number of distinct eigenval-
ues. Cospectrality of regular graphs is also considered. Chapter 5 deals with a single
subclass of regular graphs called expanders. Expanders have an enormous number
DOI: 10.1515/9783110351347-202
VI | Preface
List of Figures | XI
1 Introduction | 1
1.1 Terminology and notation | 1
1.2 Notes on regular graphs | 5
1.3 Elements of the theory of graph spectra | 7
2 Spectral properties | 15
2.1 Basic properties of the spectrum of a regular graph | 15
2.2 Operations on regular graphs and resulting spectra | 17
2.3 Other kinds of graph spectrum | 22
2.3.1 Laplacian and signless Laplacian spectrum | 22
2.3.2 Seidel spectrum | 26
2.4 Deducing regularity from the spectrum | 27
2.5 Spectral distances of regular graphs | 28
2.6 Inequalities for eigenvalues | 33
2.6.1 Second largest eigenvalue | 33
2.6.2 Toughness of regular graphs | 36
2.6.3 More inequalities | 37
2.7 Extremal regular graphs | 39
2.8 Eigenvalue estimation | 43
2.9 Eigenvalue distribution | 46
5 Expanders | 173
5.1 Combinatorial approach | 174
5.2 Algebraic approach | 179
5.3 Random walks | 184
5.4 Constructions | 187
5.4.1 Covers of graphs | 188
5.4.2 Zig-zag product | 189
5.4.3 Examples of Cayley expanders | 191
5.5 Codes from graphs | 194
5.5.1 A probabilistic algorithm | 195
Contents | IX
References | 217
Index | 233
List of Figures
Fig. 1.1 Two illustrations of the 4-dimensional cube. | 2
Fig. 1.2 Cubic graph 2C3 and its 4-regular line graph. | 4
Fig. 1.3 A cubic graph with 8 vertices. | 6
Fig. 1.4 A pair of cospectral regular graphs with 10 vertices. | 10
Fig. 3.1 C6 (top), its bipartite double (left) and extended bipartite double (right). | 51
Fig. 3.2 A walk-regular graph with 13 vertices. | 52
Fig. 3.3 8-crossed prism. | 54
Fig. 3.4 Halved 4-dimensional cube. | 62
Fig. 3.5 Shrikhande graph. | 63
Fig. 3.6 Taylor graph with (r, s) = (5, 2) (icosahedron), Johnson graph J(4, 2) (octahedron) and
Hamming graph H(3, 2) (lattice Line(K3,3 )). | 64
Fig. 3.7 Hoffman graph. | 69
Fig. 3.8 13-Paley graph and Clebsch graph. | 76
Fig. 3.9 Generalized quadrangle GQ(2, 4). | 83
Fig. 3.10 Desargues graph and its cospectral mate. | 87
Fig. 3.11 Two unicyclic star complements for λ2 = 1 with unique maximal extensions (for
Exercise 3.7.11). | 103
Fig. 4.1 Cubic and 4-regular graphs with λ2 ≤ 1 (for Theorem 4.1.3). | 122
Fig. 4.2 Integral bipartite cubic graphs (for Theorem 4.3.2). | 128
Fig. 4.3 Integral non-bipartite cubic graphs (for Theorem 4.3.2). | 128
Fig. 4.4 Integral bipartite 4-regular graphs (for Theorem 4.3.3). | 130
Fig. 4.5 Integral non-bipartite 4-regular graphs (for Theorem 4.3.3). | 131
Fig. 4.6 Q-integral bipartite (3, 4)-semiregular graphs (for Theorem 4.3.11). | 138
Fig. 4.7 A 5-regular graph with a unique spectrum (for Theorem 4.3.15). | 141
Fig. 4.8 Heawood graph and its bipartite complement. | 149
Fig. 4.9 Franklin, Möbius–Kantor, F26A and Dyck graph. | 162
Fig. 4.10 Graph cospectral with Line(K3,6 ). | 167
Fig. 6.1 Three forbidden induced subgraphs (for Exercise 6.4.2). | 214
1 Introduction
Here we give a survey of the main graph-theoretic terminology, notation and neces-
sary results. The presentation is separated into three sections. In the first two we deal
with graph structure and give some observations including statistical data on regular
graphs. In the third section we focus on the adjacency matrix and the corresponding
spectrum.
Since all parts of this chapter can be found in numerous sources, by the assump-
tion that the reader is familiar with most of the concepts presented, we orientate to a
brief but very clear and intuitive exposition. More details are given in the introductory
chapter of our previous book [290].
DOI: 10.1515/9783110351347-001
2 | 1 Introduction
Remark 1.1.1. We suggest to the reader to remember that in this book the parameter k
denotes the number of vertices in a k-walk, while its length is then equal to k − 1.
A path is a walk in which all vertices are mutually distinct. A graph which is itself a
path with n vertices is denoted P n . There is a special case of a path consisting of a
single vertex (i.e., P1 ≅ K1 ). A vertex of minimal degree in P n is called an end of P n .
By inserting an edge between the ends of P n (n ≥ 3), we get a cycle C n . The cycle C3
is known as a triangle, C4 is known as a quadrangle, and so on. Since all paths and
cycles are particular walks, the length of a path or a cycle is equal to the number of its
edges. A graph with n vertices is Hamiltonian if it contains C n as a spanning subgraph.
Any such cycle is referred to as a Hamiltonian cycle.
We say that a graph G is connected if every two (not necessary distinct) vertices are
the ends of at least one path in G. Otherwise, we say that G is disconnected. Maximal
connected induced subgraphs of a disconnected graph are referred to as its compo-
nents. A component consisting of a single vertex is called an isolated vertex of a graph,
while a graph consisting entirely of (at least two) isolated vertices is called totally dis-
connected.
A tree is a connected graph that does not contain any cycle (as a subgraph). Obvi-
ously, the number of edges m of any tree equals n − 1.
The vertex (resp. edge) connectivity c v (resp. c e ) of a connected graph is the mini-
mal number of vertices (resp. edges) whose removal results in a trivial or disconnected
graph.
The distance d(u, v) between (not necessary distinct) vertices u and v belonging to
the same component is the length of the shortest path between u and v. The diameter
of a connected graph G is defined by D = max{d(u, v) : u, v ∈ V(G)} (i.e., it is the
longest distance between two vertices of G). The girth g is the length of the shortest
cycle induced in G.
Fig. 1.2. Cubic graph 2C3 and its 4-regular line graph.
Certain operations
For two graphs G and H we define G ∪ H to be their disjoint union. We also use kG
to denote the disjoint union of k copies of G. The join G∇H is the graph obtained by
inserting an edge between every vertex of G and every vertex of H. This operation
may be applied to a sequence of graphs G1 , G2 , . . . , G k . In that case we usually write
G1 ∇G2 ∇ ⋅ ⋅ ⋅ ∇G k .
The complement of a graph G is the graph G with the same vertex set as G, in
which any two distinct vertices are adjacent if and only if they are non-adjacent in
G. Observe that if G is disconnected, then G must be connected. If G is bipartite with
parts X and Y, then its bipartite complement is the bipartite graph G with the same
parts having the edge between X and Y exactly where G does not.
A multigraph includes the possible existence of loops or multiple edges. All previ-
ous numerical invariants can be defined for multigraphs in a similar way.
1.2 Notes on regular graphs | 5
The line graph Line(G) of a multigraph G is the graph whose vertices are iden-
tified with the edges of G, with two vertices adjacent whenever the corresponding
edges have exactly one common vertex. Clearly, if G is a (simple) graph, then any two
edges cannot have more than one common vertex. We say that G is the root graph
(of Line(G)). An example is illustrated in Figure 1.2.
We say that a petal is added to a multigraph when we add a pendant vertex and
then duplicate the edge incident with it. Given a graph G with vertex set
V = {v1 , v2 , . . . , v n }, let a1 , a2 , . . . , a n be non-negative integers. The generalized
line graph Line(G; a1 , a2 , . . . , a n ) is the graph Line(̂ G), where ̂G is the multigraph
obtained from G by adding a i petals at vertex v i (1 ≤ i ≤ n).
Regular graphs of degree 0 consist of isolated vertices, all components of those of de-
gree 1 are isomorphic to the complete graph K2 , while those of degree 2 are disjoint
unions of cycles.
Observing that the complement of an r-regular graph with n vertices is regular of
degree n − r − 1, we conclude that the complete graph K n and, for n even, the cocktail
party graph CP(n) are obtained by complementing regular graphs of degree 0 and 1, re-
spectively. Similarly, every (n − 3)-regular graph is the complement of a disjoint union
of cycles.
Therefore, r-regular graphs with r ∈ {0, 1, 2, n − 1, n − 2, n − 3} are easy to de-
scribe. By considering the remaining possible values for r, we are stepping into a rich
world of regular graphs having various structural properties. Of course, since for an
r-regular graph it holds nr = 2m, we have that there is no regular graph of odd order
and odd degree. Moreover, concerning bipartite regular graphs that are not totally dis-
connected, we easily conclude that two parts must be of the same size, and so there is
no bipartite regular graph of odd order unless it is totally disconnected.
All r-regular graphs with r ≥ 2n are obtained by complementing those with r < 2n .
In addition, every regular graph whose degree is at least half the number of its vertices
must be connected. Considering small values for n, we deduce that all connected reg-
ular graphs with at most 7 vertices are cycles, cocktail party graphs, complete graphs
or complements of (unions of) cycles. This gives 16 graphs in total.¹ For n = 8, the only
critical case is a determination of cubic graphs. One of them is illustrated in Figure 1.3,
while the remaining are left as an exercise.
Exercise 1.2.1. Show that all connected cubic graphs with 8 vertices are Hamiltonian.
Show next that exactly 4 additional non-isomorphic cubic graphs can be obtained by
1 In the entire book, whenever counting graphs we assume that isomorphic graphs are identified.
6 | 1 Introduction
relocating the 4 edges that do not belong to a Hamiltonian cycle of the graph illustrated
in Figure 1.3.
The 12 remaining connected regular graphs with 8 vertices are easily determined, giv-
ing 17 in total. The situation is similar for n = 9. There are 16 connected 4-regular
graphs, giving 22 in total. The remaining numbers for 16 or fewer vertices are as fol-
lows (regular graphs of a comparatively small order can efficiently be generated by the
publicly available program GENREG described in [228]):
Order 10 11 12 13 14 15 16
# 167 539 18 979 389 436 50 314 796 2 942 198 440 1 698 517 036 411
2(2)
n
,
n!
while according to Bollobás [38] (see also Bender and Canfield [32]), the number of
non-isomorphic regular graphs of degree at least 3 is asymptotically
n−1
(2m)! r2 −1
∑ e− 4 ,
r=3
2m (r!)n m!n!
where m = nr 2 . It immediately follows that the share of regular graphs in the set of all
graphs with a given order is comparatively small. For example, less than 2.7 ⋅ 10−9 %
out of all connected graphs with 16 vertices are regular.
1.3 Elements of the theory of graph spectra | 7
Here we introduce certain basic results concerning the spectrum of the adjacency ma-
trix of a graph.² We start with the following, general, consideration. If the eigenvalues
of a real n × n matrix M are real, then they can be arranged in non-increasing order as
ν1 ≥ ν2 ≥ ⋅ ⋅ ⋅ ≥ ν n .
N1 0
P−1 MP = ( ),
0 N2
where N1 and N2 are square matrices. Otherwise, M is irreducible.
It is well known from the Perron–Frobenius theory for non-negative matrices
(see, for example, [136, Chapter 10] or [231, Chapter 8]) that the largest eigenvalue ν1
of an irreducible real symmetric matrix³ with non-negative entries is simple, while the
coordinates of any eigenvector associated with ν1 are non-zero and of the same sign.
In addition, |ν| ≤ ν1 holds for all remaining eigenvalues ν. Moreover, the largest eigen-
value of any principal submatrix of M is less than ν1 .
Throughout the book we use I, J and O to denote a unit, an all-1 and an all-0 ma-
trix, respectively. We also write j for an all-1 vector, O for an all-0 vector and ei for
the ith vector of the canonical basis of ℝn (1 ≤ i ≤ n). Occasionally, the size of any of
these matrices or the length of vectors will be given in a subscript.
Rayleigh principle
The Rayleigh quotient for a real symmetric matrix M is a scalar
yT My
,
yT y
where y is a non-zero vector in ℝn .
The Rayleigh principle reads as follows. If {x1 , x2 , . . . , xn } is an orthonormal ba-
sis of the eigenvectors of M (that correspond to ν1 , ν2 , . . . , ν n , respectively) and if a
non-zero vector y lies in the space spanned by the vectors xi , xi+1 , . . . , xn , then
yT My
νi ≥ ,
yT y
2 The whole contents can be extended to multigraphs in a very intuitive way. Multigraphs appear
sporadically in Chapter 5. Occasionally, they are abbreviated to ‘graphs’.
3 Here we restrict ourselves to symmetric matrices, although a large part of the Perron–Frobenius
theory covers non-symmetric ones.
8 | 1 Introduction
with equality if and only if My = ν i y. Similarly, if a non-zero vector y lies in the space
spanned by the vectors x1 , x2 , . . . , xi , then
yT My
νi ≤ ,
yT y
with equality if and only if My = ν i y.
k k
ν1 ( ∑ N i ) = ∑ ν1 (N i ).
i=1 i=1
1, if i is incident with e,
r ie = {
0, otherwise.
The adjacency matrix A (or A(G)) of G is defined by A = (a ij ), where
1, if i ∼ j,
a ij = {
0, otherwise.
Remark 1.3.1. In a directed graph, every edge has a direction assigned with it. If such a
graph is simple, then its adjacency matrix is also a (0, 1)-matrix in which the entry (i, j)
is 1 if there is an edge directed from i to j, and 0 otherwise. Directed graphs appear in
only a few pages of this book.
the characteristic polynomial of a graph is invariant under vertex labellings since all
adjacency matrices are mutually similar. The collection of eigenvalues of G (with rep-
etition) is called the spectrum of G. These eigenvalues are denoted
λ1 , λ2 , . . . , λ n .
In addition, we assume that λ i ≥ λ j holds whenever i < j. We simultaneously use
θ1 , θ2 , . . . , θ k
to denote all distinct eigenvalues of G ordered decreasingly. We denote the multiplic-
ity of the eigenvalue θ i by m i . Sporadically, we also use m λ for the multiplicity of an
arbitrary eigenvalue λ. The spectrum of G may be denoted [λ1 , λ2 , . . . , λ n ], but we
rather use the short expression
[θ1 m1 , θ2 m2 , . . . , θ k m k ],
where the multiplicities of distinct eigenvalues are given in the exponents.
Remark 1.3.2. The λ-notation for eigenvalues dominates in this book. The θ-notation
is convenient when we deal with distinct eigenvalues of a graph. It is used in The-
orem 2.1.6, Sections 3.3, 3.4, 4.3–4.5 and Chapter 6. Also, the spectrum of a graph is
always given in this notation. Throughout the book, when we say that a graph has k
distinct eigenvalues, we always mean that it has exactly k distinct eigenvalues.
The largest eigenvalue λ1 is called the spectral radius of G. For a non-negative inte-
ger k, the (i, j)-entry of A k is equal to the number of walks that contain k edges, start
at vertex i and terminate at vertex j. In addition, for k ≥ 0 the kth spectral moment of G
is defined by M k = ∑ni=1 λ ki . Note that M k is equal to the trace tr(A k ). The spectral gap η
is the difference between the two largest eigenvalues, that is η = λ1 − λ2 .
If a graph under consideration contains at least one edge, then its spectrum con-
tains both positive and negative eigenvalues. By Smith [283], a graph has exactly one
positive eigenvalues if and only if it has at least one edge and its non-isolated vertices
form a complete multipartite graph.
Since the adjacency matrix A of a graph G is symmetric, we can determine its mini-
mal polynomial on the basis of the spectrum. It is given by ∏ki=1 (x − θ i ), where θ i ’s are
all distinct eigenvalues of G. The identity ∏ki=1 (A l − θ i I) = O (l ≥ 0) follows directly.
Using this identity, we obtain the result concerning an interplay between the number
of distinct eigenvalues and the diameter.
Theorem 1.3.3 ( cf. [87, Theorem 3.13]). If a connected graph has k distinct eigenval-
ues, then its diameter satisfies the inequality D ≤ k − 1.
We say that a graph is determined by the spectrum if there is no another graph with
the same spectrum. Conversely, non-isomorphic graphs are said to be cospectral if
they share the same spectrum. The smallest (with respect to the order) cospectral
graphs are K1,4 and C4 ∪ K1 with common spectrum [2, 03 , −2]. The smallest con-
nected cospectral graphs have 6 vertices.
It can be confirmed by a computer search that all regular graphs with at most 9
vertices are determined by the spectrum. A pair of cospectral regular graphs with 10
vertices is illustrated in Figure 1.4. Their common spectrum⁴ is
Eigenvectors
Any eigenvalue λ of G satisfies the identity
Ax = λx, (1.1)
for some non-zero vector x ∈ ℝn . Every vector that satisfies this identity is called an
eigenvector of G. A subspace of ℝn spanned by eigenvectors associated with λ is re-
ferred to as the eigenspace of λ. We denote it E(λ). If the vertices of G are labelled
1, 2, . . . , n and x = (x1 , x2 , . . . , x n )T is an eigenvector of G, then it is usually assumed
that the coordinate x i corresponds to the ith vertex, where 1 ≤ i ≤ n. The equation
λx i = ∑ x j (1 ≤ i ≤ n). (1.2)
j∼i
4 In the entire book, irrational eigenvalues are rounded to two decimal places.
1.3 Elements of the theory of graph spectra | 11
follows directly from the identity (1.1). It is known as the eigenvalue equation of G.
In fact, this is a system of n equations involving an eigenvalue and the coordinates of
an associated eigenvector.
Given a complete system of orthonormal eigenvectors x1 , x2 , . . . , xn of A corre-
sponding to the eigenvalues λ1 , λ2 , . . . , λ n , let X = (x1 | x2 | ⋅ ⋅ ⋅ | xn ) and let B be the
diagonal matrix diag(λ1 , λ2 , . . . , λ n ). Then X is orthogonal matrix (i.e., X −1 = X T ) and
it holds
A = XBX T . (1.3)
Let now x1 , x2 , . . . , xi be a complete set of orthonormal eigenvectors that corre-
spond to an eigenvalue λ. If X λ = (x1 | x2 | ⋅ ⋅ ⋅ | xi ), then it follows that
A k = ∑ λ k X λ X λT (k ≥ 0), (1.4)
where the summation goes over all distinct eigenvalues. Moreover, if the orthonormal
basis of ℝn formed by columns of X is constructed by stringing together the orthonor-
mal basis of the eigenspaces of all distinct eigenvalues θ1 , θ2 , . . . , θ k , then the ma-
trix B from (1.3) can be expressed as B = ∑ki=1 θ i E i , where for 1 ≤ i ≤ k, each matrix E i
has a block diagonal form diag(O, . . . , O, I, O, . . . , O). Then the adjacency matrix has
the spectral decomposition
k
A = ∑ θi Pi , (1.5)
i=1
In other words, the inequality |λ| ≤ ∆ holds for any eigenvalue λ of any graph G.
Since the adjacency matrix of a connected graph is irreducible, the next results
are immediate consequences of the Perron–Frobenius theory.
Theorem 1.3.4 (cf. [89, Corollary 1.3.8]). A graph is connected if and only if its spectral
radius is a simple eigenvalue with an eigenvector whose coordinates are non-zero and
of the same sign.
Corollary 1.3.5 (cf. [89, Proposition 1.3.10]). For any vertex i or any edge e of a con-
nected graph G, we have λ1 (G − i) < λ1 (G) and λ1 (G − e) < λ1 (G).
For a graph G with n vertices we have the following inequalities bounding its spectral
radius
12 | 1 Introduction
λ1 (P n ) ≤ λ1 (G) ≤ λ1 (K n ), (1.7)
where for the first one the connectedness is stipulated. Equalities hold if and only if
G ≅ P n or G ≅ K n . The first inequality was obtained by Lovász and Pelikán [216], while
the second is a direct consequence of the last corollary.
Eigenvalue interlacing
Consider the following general result.
Theorem 1.3.6 (cf. [89, Theorem 1.3.11]). For a real n × n matrix N such that N T N = I,
let M be an n × n real symmetric matrix with eigenvalues ν1 ≥ ν2 ≥ ⋅ ⋅ ⋅ ≥ ν n . If the
eigenvalues of N T MN are ξ1 ≥ ξ2 ≥ ⋅ ⋅ ⋅ ≥ ξ n , then
ν n−n +i ≤ ξ i ≤ ν i (1 ≤ i ≤ n ).
Now, if G is a graph with n vertices and H its induced subgraph with n vertices, then
A(H) = N T A(G)N, where N is the block matrix N = (I | O)T . Thus we get the following
result.
Theorem 1.3.7 ( Interlacing Theorem, cf. [89, Corollary 1.3.12]). Let G be a graph with
n vertices and eigenvalues λ1 (G) ≥ λ2 (G) ≥ ⋅ ⋅ ⋅ ≥ λ n (G), and let H be an induced sub-
graph of G with n vertices and eigenvalues λ1 (H) ≥ λ2 (H) ≥ ⋅ ⋅ ⋅ ≥ λ n (H). Then
Theorem 1.3.8 ( cf. [89, Corollary 1.3.13] or [158]). Let M be a real symmetric matrix
with eigenvalues ν1 ≥ ν2 ≥ ⋅ ⋅ ⋅ ≥ ν n . Given a partition {1, 2, . . . , n} = P1 ⊔ P2 ⊔ ⋅ ⋅ ⋅ ⊔ P n
with |P i | = n i > 0, consider the blocking M = (M ij ) where M ij is an n i × n j block. Let e ij
1.3 Elements of the theory of graph spectra | 13
e
be the sum of entries in M ij and set B = ( niji ). Then the eigenvalues of B interlace those
of M.
1
N=(
1
c1 ⋅ ⋅ ⋅ 1 cn ) ,
√n1 √n c2 √n
2 n
Theorem 1.3.9 (Petersdorf and Sachs, cf. [87, Theorem 4.5]). The characteristic poly-
nomial of any divisor of a graph divides its characteristic polynomial.
Theorem 1.3.10 (Cvetković, cf. [90, Theorem 2.4.5]). The spectrum of any divisor of a
graph G includes the main part of the spectrum of G.
Remark 1.3.11. Theorems 1.3.9 and 1.3.10 remain valid for multigraphs [87, Chapter 4].
The spectral radius of a connected graph is always a main eigenvalue since the coor-
dinates of a corresponding eigenvector are non-zero and of the same sign. The main
angles of G are the cosines of angles between the eigenspaces of its eigenvalues and
the vector j.
O N
( ). (1.8)
NT O
14 | 1 Introduction
If G has n1 vertices in the first part and n2 vertices in the second, then the size of N
is n1 × n2 . Let λ be an eigenvalue of A and x be a corresponding eigenvector. By setting
x1 x1
x = ( ) and x = ( )
x2 −x2
where the length of x i is n i (1 ≤ i ≤ 2), we obtain
O N x1 −Nx2 −λx1
Ax = ( )( )=( T )=( ) = −λx .
NT O −x2 N x1 λx2
In other words, the spectrum of G is symmetric with respect to the origin. The converse
is also true. Indeed, if the spectrum of a graph is symmetric with respect to the origin,
then its odd spectral moments are 0, and hence it does not contain odd cycles, which
yields its bipartiteness. Moreover, it can be proved that the assumption of the converse
state may be reduced to λ1 = −λ n (so, only the symmetry between extreme eigenvalues
is required) [89, Theorem 3.2.4]. In this way we arrive at the following formulation.
Theorem 1.3.12 (Cvetković, cf. [89, Theorems 3.2.3 and 3.2.4] or [100]). A graph G is
bipartite if and only if its spectrum is symmetric with respect to the origin. If G is con-
nected, then it is bipartite if and only if λ1 = −λ n .
Remark 1.3.13. The form (1.8) of the adjacency matrix of a biparitite graph will be fre-
quently used throughout the book. We suggest to the reader to remember the role of
the submatrix N in this representation.
2 Spectral properties
Basic properties of the spectra of regular graphs, including the Hoffman polynomial,
are introduced in Section 2.1.
In Section 2.2 we consider certain graph operations (like making the complement,
the join or the line graph) applied to a regular graph and the spectrum of the resulting
graph.
Those who study graphs by means of their spectra usually work with different
theories based on different matrices associated with graphs. In Section 2.3 we intro-
duce the three frequently investigated matrices (precisely, the Laplacian matrix, the
signless Laplacian matrix and the Seidel matrix) and give properties of their spectra.
In a short Section 2.4 we inspect whether the regularity of a graph can be deduced
from the spectrum of an associated matrix or not.
The concept of distances of regular graphs, which is based on different kinds of
the spectrum and includes reproduction of some of our computational results, is given
in Section 2.5.
Inequalities connecting spectral and structural invariants of regular graphs have
received the attention in Section 2.6. There we meet the significance of the second
largest and the least eigenvalue, and also consider the spectral radius of nearly regu-
lar graphs. There is a large amount of known inequalities that are attained exactly for
regular graphs. Some of them are separated in Section 2.7. We include some classical
results (like those of Stanley, Schwenk, Anderson and Morley or Merris) together with
later and very recent developments. An extension of results concerning inequalities
is given in Section 2.8 in which we estimate sets of largest or smallest eigenvalues by
presenting the famous Serre theorem and also a variant of the Alon–Boppana theo-
rem.
In Section 2.9 we study the eigenvalue distribution for regular graphs. In partic-
ular, we consider the Erdős–Rényi random graphs and the corresponding semicircle
eigenvalue distribution in relation to random regular graphs and the McKay distribu-
tion.
If all-1 vector j is an eigenvector of a graph G with n vertices, then for the adjacency
matrix A (of G) we have Aj = λj for some eigenvalue λ. In other words, it holds
d i = λ (1 ≤ i ≤ n), that is G is a λ-regular graph. Conversely, for every r-regular
graph G we deduce Aj = rj. Thus, we may formulate the following theorem.
DOI: 10.1515/9783110351347-002
16 | 2 Spectral properties
Even more, it can easily be concluded that the equality λ1 = ∆ (of (1.6)) is attained if
and only if G is regular. A generalization of the last theorem is left as an exercise for
the reader.
T
Exercise 2.1.2 ([154]). Prove that if (√d1 , √d2 , . . . , √d n ) is an eigenvector of a graph
G, then G is either regular or bipartite semiregular.
Corollary 2.1.3 (cf. [87, Theorem 3.22] or [100]). A graph is r-regular if and only if
n
r = λ1 and nλ1 = ∑ λ2i .
i=1
In other words, regularity can be recognized from the spectrum. The next corollary is
obvious.
Corollary 2.1.4. The number of components of a regular graph is equal to the multiplic-
ity of its spectral radius.
Since in the case of a connected regular graph every eigenvector outside the space
spanned by j is orthogonal to j, we have the following simple observation.
Corollary 2.1.5. Connected graphs with exactly one main eigenvalue are precisely regu-
lar graphs. This eigenvalue is the largest one.
Hoffman polynomial
For any graph G with the adjacency matrix A, its adjacency algebra consists of all
matrices P(A) where P is a polynomial with real coefficients. There is an interesting
property of regular graphs.
Theorem 2.1.6 (Hoffman Polynomial, [166], cf. [89, Theorem 3.5.5]). For a graph G
with the adjacency matrix A there exists a polynomial P such that P(A) = J, where J is
all-1 matrix, if and only if G is connected and regular. If G is r-regular, then we have
k
x − θi
P(x) = n ∏ , (2.1)
i=2
r − θi
Proof. Assume that the corresponding polynomial exists. In this case, J belongs to the
adjacency algebra of G, which implies AJ = JA. If d1 , d2 , . . . , d n are vertex degrees,
then we have AJ = (d1 , d2 , . . . , d n )T jT and JA = j(d1 , d2 , . . . , d n ), giving d1 = d2 =
⋅ ⋅ ⋅ = d n , and so G is regular. If G is disconnected, then there are vertices u and v
belonging to different components of G. If so, then the (u, v)-entry of P(A) is 0 for any
polynomial P, which contradicts the fact that J belongs to the adjacency algebra of G.
Another random document with
no related content on Scribd:
Thus the Pawnees, too, though in actual want themselves, showed
their good will towards us by preparing for us the best they could
afford. One chief placed before us, besides dried buffalo meat, a
small dish of marrow, which by them is considered a great delicacy.
Labor.—The more enlightened a people has become through the
benign influence of religion, and the cultivation of the arts and
sciences, the more the weaker sex attains to that rank in society,
which the Creator intended for them; but the more debased and sunk
in heathenism a Nation lies, the more is woman enslaved, and held
in subjection. So among the Pawnees, too, the men are the lords,
who, while they themselves are not disposed to do any work, and
generally consider it a shame to labor, direct the women, as inferior
beings, how everything must be done. The females, besides
attending to the housework, generally incumbent upon their sex, are
obliged to cut and carry all the fire-wood; prepare the fields for
cultivation with their hoes, plant the corn, weed it, and finally, when
ripe, gather it in; and take care of the horses of their lords, when not
used by them.
The most difficult labor, however, that falls to their lot, is the
erecting of their lodges. As the upper village had been burnt by their
enemies during their absence, the labor of rebuilding of course
devolved upon the poor women, and in the course of about two
months, when we visited them, they had completed nearly 150
lodges. When we observed that the nearest timber to the site of the
village was two or three miles off, and that they have no beasts of
burden, but that all the building material, much of which is very
heavy, was carried so great a distance on the shoulders of the
women, it seemed almost incredible to us. We noticed even girls of
12 to 14 years bearing baskets on their backs, filled with wet turf, to
cover or repair their lodges, which must have been very heavy. Thus
early the females are trained to carry burdens (either at home or
while travelling), which inures them to hardships, but at the same
time deforms them, rendering them low and bent in stature, and
pitiable objects to the beholder. While the men, trained to no work
whatever, are erect, tall and well-proportioned in their form, and
almost feminine in their features, the women are short and low, with
stooping gait, downcast looks, harsh features, and hair dishevelled;
and dirty, ragged, and filthy in their appearance. Such is a picture of
heathenism! Have we not cause to be grateful for the benign
influence of the gospel upon our land, so that we are elevated far
above such a miserable state of heathenism? Unto the Lord be all
the praise!
Though it is generally a feature among heathen nations, that it is
considered a shame for the men to labor, while all the work devolves
upon the females, still the Pawnees seem not altogether indisposed
to be instructed. Mr. Sharpee, who has been acquainted with them
and other Indians in these parts for nearly 20 years, told us that the
Pawnee men were unlike those of other tribes, since they were
willing to lay hold of a plough or any other farming utensil, and only
the instruction was wanting to make them industrious. Mr. Allis, too,
who has been acquainted with them for the last 20 years, and has
lived among them a great portion of this period, informed us that
when the missionaries were formerly living among them, several of
the Pawnees had already commenced the use of the plough. The
good beginning was, however, interrupted by their missionaries
having to leave them.
Dress.—The dress of the men consists in nothing but a small
piece of cloth, fastened around the middle. This, we observed, was
generally the only covering they had on, when in their lodges. When
abroad, they had a buffalo robe or blanket wrapped around them.
The females are more decently clothed than the men. Their dress
consists of leggings, generally made of scarlet cloth, over which a
piece of blue or brown woolen cloth is tied around the waist, the
ends lapping over each other and extending a little below the knees.
The upper dress consists of a shirt, or short gown, made of calico,
reaching to the middle. Over the whole a blanket or robe is slung,
above which, on the back, may frequently be discovered the black
sparkling eyes of a little “papoose” peering forth.
Intoxicating Drink.—It is somewhat remarkable that the
Pawnees are not so fond of “fire water” as other tribes, who are fast
dwindling away from the demoralizing and debasing effects of
intoxicating drink. Nothing, perhaps, has aided more in diminishing
the population of different Indian tribes, than spirituous liquors. To
the missionary it is always gratifying and encouraging to observe,
when a tribe takes a bold stand against the introduction of this
destructive poison. The Pawnees allow none to be brought into their
villages.
Burying Their Dead.—On the highest mounds in the prairie, we
often observed little hillocks of earth, which we were informed were
the places of sepulture of their chiefs and others of their tribe. A tall
bush was frequently stuck in the ground, to designate the spot. We
were also informed, that among the Pawnees and other wild tribes
the cruel practice prevails, when they are on their hunt, and their old
people get so feeble, that they can no longer accompany them, of
leaving them behind, and even burying them alive.—In allusion to
this fact Mr. Allis, pointing to their old and feeble chief, addressed the
other chiefs as follows: “I have learned that you intend to put aside
your old chief, because he is now feeble and sickly, and perhaps you
will even think of burying him, when he can no longer follow you; but
I want you to take care of him. He has been like a father to you.
When he was yet strong, you know his kettle was always over the
fire with plenty. He has always been a friend of the whites, and you
ought to esteem him for the services he has rendered your nation,
and not to bury him before he is dead.”
More might be added, to show the wretchedness of their present
state, but we should fear to become too tedious. Sufficient, however,
has been said to prove that these poor people need the assistance
of missionaries to better them in their outward condition. But, when
we consider their spiritual destitution—their ignorance, darkness and
superstition—
1.D. The copyright laws of the place where you are located also
govern what you can do with this work. Copyright laws in most
countries are in a constant state of change. If you are outside
the United States, check the laws of your country in addition to
the terms of this agreement before downloading, copying,
displaying, performing, distributing or creating derivative works
based on this work or any other Project Gutenberg™ work. The
Foundation makes no representations concerning the copyright
status of any work in any country other than the United States.
1.E.6. You may convert to and distribute this work in any binary,
compressed, marked up, nonproprietary or proprietary form,
including any word processing or hypertext form. However, if
you provide access to or distribute copies of a Project
Gutenberg™ work in a format other than “Plain Vanilla ASCII” or
other format used in the official version posted on the official
Project Gutenberg™ website (www.gutenberg.org), you must, at
no additional cost, fee or expense to the user, provide a copy, a
means of exporting a copy, or a means of obtaining a copy upon
request, of the work in its original “Plain Vanilla ASCII” or other
form. Any alternate format must include the full Project
Gutenberg™ License as specified in paragraph 1.E.1.
• You pay a royalty fee of 20% of the gross profits you derive from
the use of Project Gutenberg™ works calculated using the
method you already use to calculate your applicable taxes. The
fee is owed to the owner of the Project Gutenberg™ trademark,
but he has agreed to donate royalties under this paragraph to
the Project Gutenberg Literary Archive Foundation. Royalty
payments must be paid within 60 days following each date on
which you prepare (or are legally required to prepare) your
periodic tax returns. Royalty payments should be clearly marked
as such and sent to the Project Gutenberg Literary Archive
Foundation at the address specified in Section 4, “Information
about donations to the Project Gutenberg Literary Archive
Foundation.”
• You comply with all other terms of this agreement for free
distribution of Project Gutenberg™ works.
1.F.
Most people start at our website which has the main PG search
facility: www.gutenberg.org.