目录
Chapter 3.直径和特征值(Diameters and eigenvalues)
3.1 图的直径(The diameter of a graph)
3.3 许多子集之间的特征值和距离(Eigenvalues and distances among many subsets)
Chapter 3.直径和特征值(Diameters and eigenvalues)
3.1 图的直径(The diameter of a graph)
In a graph , the distance between two vertices u and v, denoted by
, is defined to be the length of a shortest path joining u and v in
. (It is possible to define the diameter by various more general measures). The diameter of
, denoted by
, is the maximum distance over all pairs of vertices in
. The diameter is one of the key invariants in a graph which is not only of theoretical interest but also has a wide range of applications. When graphs are used as models for communication networks, the diameter corresponds to the delays in passing messages through the network, and therefore plays an important role in performance analysis and cost optimization.
Although the diameter is a combinatorial invariant, it is closely related to eigenvalues. This connection is based on the following simple observation:
Let M denote an matrix with rows and columns indexed by the vertices of
. Suppose
satisfies the property that
if
and
are not adjacent. Furthermore, suppose we can show that for some integer and some polynomial
of degree
, we have
for all and
. Then we can conclude that the diameter
satisfies:
Suppose we take to be the sum of the adjacency matrix and the identity matrix and
the polynomial to be just
The following inequality for regular graphs which are not complete graphs can then be derived (which will be proved in Section 3.2 as a corollary to Theorem 3.1; also see [51]):
3.1
Here, basically only depends on
. For example, we can take
if
. In general, we can slightly improve (3.1) by using the same “spectrum shifting” trick as in Section 1.5 (see Section 3.2). Namely, we define
, and we conclude
We note that for some graphs the above bound gives a pretty good upper bound for the diameter. For example, for k-regular Ramanujan graphs, we have
so we get
which is within a factor of 2 of the best possible bound.
The bound in (3.1) can be further improved by choosing pt to be the Chebyshev polynomial of degree t
. We can then replace the logarithmic function by cosh-1
(see [53] and Theorem 3.3):
The above inequalities can be generalized in several directions. Instead of considering distances between two vertices, we can relate the eigenvalue Ai to distances between two subsets of vertices (see Section 3.2). Furthermore, for any , we can relate the eigenvalue
to distances among
distinct subsets of vertices (see Section 3.3).
We will derive several versions of the diameter-eigenvalue inequalities. From these inequalities, we can deduce a number of isoperimetric inequalities which are closely related to expander graphs which will also be discussed in Chapter 6.
It is worth mentioning that the above discrete methods for bounding eigenvalues can be used to derive new eigenvalue upper bounds for compact smooth Riemannian manifolds This will be discussed in the last section of this chapter.
In contrast to many other more complicated graph invariants, the diameter is easy to compute. The diameter is the least integer such that the matrix
has the property that all entries of
are nonzero. This can be determined by using
iterations of matrix multiplication. Using the current best known bound
for matrix multiplication where
)
this diameter algorithm requires at most steps. The problem of determining distances of all pairs of vertices for an undirected graph can also be done in
time. Seidel [223] gave a simple recursive algorithm by reducing this problem for a graph
to a graph G in which
if
. (For directed graphs, an
algorithm can be found in [10].)
Another related problem is to find shortest paths between all pairs of vertices, which can be easily done in steps (in fact
is enough for a graph on
vertices and
edges). Apparently, we cannot compute all shortest paths explicitly in O
time since some graphs can have
pairs of vertices having shortest paths of length at least
each. However, we can compute a data structure that allows all shortest paths be constructed in time proportional to their lengths. For example, a matrix has its
-entry to be a neighbor of
in a shortest path connecting
and
. Seidel gave a randomized algorithm to compute such a matrix in expected time
.
3.2 特征值和两个子集之间的距离
For two subsets of vertices in
, the distance between
and
, denoted by
, is the minimum distance between a vertex in
and a vertex in
, i.e.,
Let denote the complement of
in
.
Theorem 3.1.
Suppose is not a complete graph. For
and
, we have
3.3
Corollary 3.2.
Suppose G is a regular graph which is not complete. Then
Theorem 3.3.
Suppose G is not a complete graph. For and
, we have
Lemma 3.4.
For all , we have
Corollary 3.5.
For with
, where G is not a complete graph, we have
where
Lemma 3.6.
For and any integer
,
where
Lemma 3.7.
For an integer and
with
, we have
where
Lemma 3.8.
For with
and any integer
,
3.3 许多子集之间的特征值和距离(Eigenvalues and distances among many subsets)
Lemma 3.9.
Let denote
arbitrary vectors in d-dimensional Euclidean space. Then there are two of them, say,
such that
.
Theorem 3.10.
Suppose G is not a complete graph. For we have
if . and
, for i=0,1,⋯,k
Theorem 3.11.
For , we have
if and