Eigenvalues and The Laplacian of A Graph
Eigenvalues and The Laplacian of A Graph
_
1 if u = v and d
v
,= 0,
d
u
d
v
if u and v are adjacent,
0 otherwise.
We can write
L = T
1/2
LT
1/2
with the convention T
1
(v, v) = 0 for d
v
= 0. We say v is an isolated vertex if
d
v
= 0. A graph is said to be nontrivial if it contains at least one edge.
1.2. THE LAPLACIAN AND EIGENVALUES 3
The Laplacian can be viewed as an operator on the space of functions g :
V (G) R which satises
Lg(u) =
1
d
u
v
uv
_
g(u)
d
u
g(v)
d
v
_
.
When G is k-regular, it is easy to see that
L = I
1
k
A,
where A is the adjacency matrix of G (i.e., A(x, y) = 1 if x is adjacent to y, and 0
otherwise,) and I is an identity matrix. All matrices here are n n where n is the
number of vertices in G.
For a general graph without isolated vertices, we have
L = T
1/2
LT
1/2
= I T
1/2
AT
1/2
.
We note that L can be written as
L = SS
,
where S is the matrix whose rows are indexed by the vertices and whose columns
are indexed by the edges of G such that each column corresponding to an edge
e = u, v has an entry 1/
d
u
in the row corresponding to u, an entry 1/
d
v
in
the row corresponding to v, and has zero entries elsewhere. (As it turns out, the
choice of signs can be arbitrary as long as one is positive and the other is negative.)
Also, S
C
0
.
Since L is symmetric, its eigenvalues are all real and non-negative. We can
use the variational characterizations of those eigenvalues in terms of the Rayleigh
quotient of L (see, e.g., [165]). Let g denote an arbitrary function which assigns to
each vertex v of G a real value g(v). We can view g as a column vector. Then
g, Lg)
g, g)
=
g, T
1/2
LT
1/2
g)
g, g)
=
f, Lf)
T
1/2
f, T
1/2
f)
=
uv
(f(u) f(v))
2
v
f(v)
2
d
v
(1.1)
4 1. EIGENVALUES AND THE LAPLACIAN OF A GRAPH
where g = T
1/2
f and
uv
denotes the sum over all unordered pairs u, v for which
u and v are adjacent. Here f, g) =
x
f(x)g(x) denotes the standard inner product
in R
n
. The sum
uv
(f(u) f(v))
2
is sometimes called the Dirichlet sum of G and
the ratio on the left-hand side of (1.1) is often called the Rayleigh quotient. (We
note that we can also use the inner product f, g) =
G
=
1
= inf
fT1
uv
(f(u) f(v))
2
v
f(v)
2
d
v
. (1.2)
The corresponding eigenfunction is g = T
1/2
f as in (1.1). It is sometimes convenient
to consider the nontrivial function f achieving (1.2), in which case we call f a
harmonic eigenfunction of L.
The above formulation for
G
corresponds in a natural way to the eigenvalues
of the Laplace-Beltrami operator for Riemannian manifolds:
M
= inf
_
M
[f[
2
_
M
[f[
2
,
where f ranges over functions satisfying
_
M
f = 0.
We remark that the corresponding measure here for each edge is 1 although in the
general case for weighted graphs the measure for an edge is associated with the edge
weight (see Section 1.4). The measure for each vertex is the degree of the vertex.
A more general notion of vertex weights will also be considered in Section 1.4.
1.2. THE LAPLACIAN AND EIGENVALUES 5
We note that (1.2) has several dierent formulations:
1
= inf
f
sup
t
uv
(f(u) f(v))
2
v
(f(v) t)
2
d
v
(1.3)
= inf
f
uv
(f(u) f(v))
2
v
(f(v)
f)
2
d
v
, (1.4)
where
f =
v
f(v)d
v
vol G
,
and vol G denotes the volume of the graph G, given by
vol G =
v
d
v
.
By substituting for
f and using the fact that N
N
i=1
(a
i
a)
2
=
i<j
(a
i
a
j
)
2
for a =
N
i=1
a
i
/N, we have the following expression (which generalizes the one in
[127]):
1
= vol G inf
f
uv
(f(u) f(v))
2
u,v
(f(u) f(v))
2
d
u
d
v
, (1.5)
where
u,v
denotes the sum over all unordered pairs of vertices u, v in G. We can
characterize the other eigenvalues of L in terms of the Rayleigh quotient. The
largest eigenvalue satises:
n1
= sup
f
uv
(f(u) f(v))
2
v
f
2
(v)d
v
. (1.6)
For a general k, we have
k
= inf
f
sup
gP
k1
uv
(f(u) f(v))
2
v
(f(v) g(v))
2
d
v
(1.7)
= inf
f TP
k1
uv
(f(u) f(v))
2
v
f(v)
2
d
v
(1.8)
6 1. EIGENVALUES AND THE LAPLACIAN OF A GRAPH
where P
k1
is the subspace generated by the harmonic eigenfunctions corresponding
to
i
, for i k 1.
The dierent formulations for eigenvalues given above are useful in dierent
settings and they will be used in later chapters. Here are some examples of special
graphs and their eigenvalues.
Example 1.1. For the complete graph K
n
on n vertices, the eigenvalues are 0
and n/(n 1) (with multiplicity n 1).
Example 1.2. For the complete bipartite graph K
m,n
on m + n vertices, the
eigenvalues are 0, 1 (with multiplicity m+n 2), and 2.
Example 1.3. For the star S
n
on n vertices, the eigenvalues are 0, 1 (with
multiplicity n 2), and 2.
Example 1.4. For the path P
n
on n vertices, the eigenvalues are 1 cos
k
n1
for k = 0, 1, . . . , n 1.
Example 1.5. For the cycle C
n
on n vertices, the eigenvalues are 1 cos
2k
n
for k = 0, . . . , n 1.
Example 1.6. For the n-cube Q
n
on 2
n
vertices, the eigenvalues are
2k
n
(with
multiplicity
_
n
k
_
) for k = 0, . . . , n.
More examples can be found in Chapter 6 on explicit constructions.
1.3. Basic facts about the spectrum of a graph
Roughly speaking, half of the main problems of spectral theory lie in deriving
bounds on the distributions of eigenvalues. The other half concern the impact and
consequences of the eigenvalue bounds as well as their applications. In this section,
we start with a few basic facts about eigenvalues. Some simple upper bounds and
lower bounds are stated. For example, we will see that the eigenvalues of any graph
lie between 0 and 2. The problem of narrowing the range of the eigenvalues for
special classes of graphs oers an open-ended challenge. Numerous questions can
be asked either in terms of other graph invariants or under further assumptions
imposed on the graphs. Some of these will be discussed in subsequent chapters.
Lemma 1.7. For a graph G on n vertices, we have
(i):
i
n
with equality holding if and only if G has no isolated vertices.
(ii): For n 2,
1
n
n 1
with equality holding if and only if G is the complete graph on n ver-
tices. Also, for a graph G without isolated vertices, we have
n1
n
n 1
.
1.3. BASIC FACTS ABOUT THE SPECTRUM OF A GRAPH 7
(iii): For a graph which is not a complete graph, we have
1
1.
(iv): If G is connected, then
1
> 0. If
i
= 0 and
i+1
,= 0, then G has
exactly i + 1 connected components.
(v): For all i n 1, we have
i
2,
with
n1
= 2 if and only if a connected component of G is bipartite
and nontrivial.
(vi): The spectrum of a graph is the union of the spectra of its connected
components.
Proof. Item (i) follows from considering the trace of L.
The inequalities in (ii) follow from (i) and
0
= 0.
Suppose G contains two nonadjacent vertices a and b, and consider
f
1
(v) =
_
_
_
d
b
if v = a,
d
a
if v = b,
0 if v ,= a, b.
Item (iii) then follows from (1.2).
If G is connected, the eigenvalue 0 has multiplicity 1 since any harmonic eigen-
function with eigenvalue 0 assumes the same value at each vertex. Thus, (iv) follows
from the fact that the union of two disjoint graphs has as its spectrum the union
of the spectra of the original graphs.
Item (v) follows from equation (1.6) and the fact that
(f(x) f(y))
2
2(f
2
(x) +f
2
(y)).
Therefore
i
sup
f
xy
(f(x) f(y))
2
x
f
2
(x)d
x
2.
Equality holds for i = n 1 when f(x) = f(y) for every edge x, y in G.
Therefore, since f ,= 0, G has a bipartite connected component. On the other hand,
if G has a connected component which is bipartite, we can choose the function f
so as to make
n1
= 2.
Item (vi) follows from the denition.
For bipartite graphs, the following slightly stronger result holds:
Lemma 1.8. The following statements are equivalent:
(i): G is bipartite.
(ii): G has i + 1 connected components and
nj
= 2 for 1 j i.
(iii): For each
i
, the value 2
i
is also an eigenvalue of G.
8 1. EIGENVALUES AND THE LAPLACIAN OF A GRAPH
Proof. It suces to consider a connected graph. Suppose G is a bipartite
graph with vertex set consisting of two parts A and B. For any harmonic eigen-
function f with eigenvalue , we consider the function g dened by
g(x) =
_
f(x) if x A,
f(x) if x B.
It is easy to check that g is a harmonic eigenfunction with eigenvalue 2 .
For a connected graph, we can immediately improve the lower bound of
1
in
Lemma 1.7. For two vertices u and v, the distance between u and v is the number of
edges in a shortest path joining u and v. The diameter of a graph is the maximum
distance between any two vertices of G. Here we will give a simple eigenvalue lower
bound in terms of the diameter of a graph. More discussion on the relationship
between eigenvalues and diameter will be given in Chapter 3.
Lemma 1.9. For a connected graph G with diameter D, we have
1
1
D vol G
.
Proof. Suppose f is a harmonic eigenfunction achieving
1
in (1.2). Let v
0
denote a vertex with [f(v
0
)[ = max
v
[f(v)[. Since
v
f(v)d
v
= 0, there exists a
vertex u
0
satisfying f(u
0
)f(v
0
) < 0. Let P denote a shortest path in G joining u
0
and v
0
. Then by (1.2) we have
1
=
xy
(f(x) f(y))
2
x
f
2
(x)d
x
{x,y}P
(f(x) f(y))
2
vol G f
2
(v
0
)
1
D
(f(v
0
) f(u
0
))
2
vol G f
2
(v
0
)
1
D vol G
by using the Cauchy-Schwarz inequality.
Lemma 1.10. Let f denote a harmonic eigenfunction achieving
G
in (1.2).
Then, for any vertex x V , we have
1
d
x
y
yx
(f(x) f(y)) =
G
f(x).
1.3. BASIC FACTS ABOUT THE SPECTRUM OF A GRAPH 9
Proof. We use a variational argument. For a xed x
0
V , we consider f
such that
f
(y) =
_
_
f(x
0
) +
d
x0
if y = x
0
,
f(y)
vol Gd
x0
otherwise.
We have
x,yV
xy
(f
(x) f
(y))
2
xV
f
2
(x)d
x
=
x,yV
xy
(f(x) f(y))
2
+
y
yx0
2(f(x
0
) f(y))
d
x0
y
y=x0
yy
2(f(y) f(y
))
vol Gd
x0
xV
f
2
(x)d
x
+ 2f(x
0
)
2
vol Gd
x0
y=x0
f(y)d
y
+O(
2
)
=
x,yV
xy
(f(x) f(y))
2
+
2
y
yx0
(f(x
0
) f(y))
d
x0
+
2
y
yx0
(f(x
0
) f(y))
vol Gd
x0
xV
f
2
(x)d
x
+ 2f(x
0
) +
2f(x
0
)d
x0
vol Gd
x0
+O(
2
)
since
xV
f(x)d
x
= 0, and
(f(y) f(y
x,yV
xy
(f
(x) f
(y))
2
xV
f
2
(x)d
x
x,yV
xy
(f(x) f(y))
2
xV
f
2
(x)d
x
.
If we consider what happens to the Rayleigh quotient for f
as 0 from above,
or from below, we can conclude that
1
d
x0
y
yx0
(f(x
0
) f(y)) =
G
f(x
0
)
and the Lemma is proved.
10 1. EIGENVALUES AND THE LAPLACIAN OF A GRAPH
One can also prove the statement in Lemma 1.10 by recalling that f = T
1/2
g,
where Lg =
G
g. Then
T
1
Lf = T
1
(T
1/2
LT
1/2
)(T
1/2
g) = T
1/2
G
g =
G
f,
and examining the entries gives the desired result.
With a little linear algebra, we can improve the bounds on eigenvalues in terms
of the degrees of the vertices.
We consider the trace of (I L)
2
. We have
Tr(I L)
2
=
i
(1
i
)
2
1 + (n 1)
2
, (1.9)
where
= max
i=0
[1
i
[.
On the other hand,
Tr(I L)
2
= Tr(T
1/2
AT
1
AT
1/2
) (1.10)
=
x,y
1
d
x
A(x, y)
1
d
y
A(y, x)
1
d
x
=
x
1
d
x
xy
(
1
d
x
1
d
y
)
2
,
where A is the adjacency matrix. From this, we immediately deduce
Lemma 1.11. For a k-regular graph G on n vertices, we have
max
i=0
[1
i
[
n k
(n 1)k
(1.11)
This follows from the fact that
max
i=0
[1
i
[
2
1
n 1
(Tr(I L)
2
1).
Let d
H
denote the harmonic mean of the d
v
s, i.e.,
1
d
H
=
1
n
v
1
d
v
.
It is tempting to consider generalizing (1.11) with k replaced by d
H
. This, however,
is not true as shown by the following example due to Elizabeth Wilmer.
Example 1.12. Consider the m-petal graph on n = 2m+1 vertices, v
0
, v
1
, . . .,
v
2m
with edges v
0
, v
i
and v
2i1
, v
2i
, for i 1. This graph has eigenvalues
0, 1/2 (with multiplicity m 1), and 3/2 (with multiplicity m + 1). So we have
max
i=0
[1
i
[ = 1/2. However,
n d
H
(n 1)d
H
=
_
m1/2
2m
1
2
as m .
1.4. EIGENVALUES OF WEIGHTED GRAPHS 11
Still, for a general graph, we can use the fact that
xy
(
1
d
x
1
d
y
)
2
xV
(
1
d
x
1
d
H
)
2
d
x
n1
1 +
. (1.12)
Combining (1.9), (1.10) and (1.12), we obtain the following:
Lemma 1.13. For a graph G on n vertices,
= max
i=0
[1
i
[ satises
1 + (n 1)
n
d
H
(1 (1 +
)(
k
d
H
1)),
where k denotes the average degree of G.
There are relatively easy ways to improve the upper bound for
1
. From the
characterization in the preceding section, we can choose any function f : V (G) R,
and its Rayleigh quotient will serve as an upper bound for
1
. Here we describe an
upper bound for
1
(see [208]).
Lemma 1.14. Let G be a graph with diameter D 4, and let k denote the
maximum degree of G. Then
1
1 2
k 1
k
_
1
2
D
_
+
2
D
.
One way to bound eigenvalues from above is to consider contracting the
graph G into a weighted graph H (which will be dened in the next section). Then
the eigenvalues of G can be upper-bounded by the eigenvalues of H or by various
upper bounds on them, which might be easier to obtain. We remark that the proof
of Lemma 1.14 proceeds by basically contracting the graph into a weighted path.
We will prove Lemma 1.14 in the next section.
We note that Lemma 1.14 gives a proof (see [5]) that for any xed k and for
any innite family of regular graphs with degree k,
limsup
1
1 2
k 1
k
.
This bound is the best possible since it is sharp for the Ramanujan graphs (which
will be discussed in Chapter 6). We note that the cleaner version of
1
1
2
k 1/k is not true for certain graphs (e.g., 4-cycles or complete bipartite graphs).
This example also illustrates that the assumption in Lemma 1.14 concerning D 4
is essential.
1.4. Eigenvalues of weighted graphs
Before dening weighted graphs, we will say a few words about two dierent
approaches for giving denitions. We could have started from the very beginning
with weighted graphs, from which simple graphs arise as a special case in which
the weights are 0 or 1. However, the unique characteristics and special strength of
12 1. EIGENVALUES AND THE LAPLACIAN OF A GRAPH
graph theory is its ability to deal with the 0, 1-problems arising in many natural
situations. The clean formulation of a simple graph has conceptual advantages.
Furthermore, as we shall see, all denitions and subsequent theorems for simple
graphs can usually be easily carried out for weighted graphs. A weighted undirected
graph G (possibly with loops) has associated with it a weight function w : V V
R satisfying
w(u, v) = w(v, u)
and
w(u, v) 0.
We note that if u, v , E(G) , then w(u, v) = 0. Unweighted graphs are just the
special case where all the weights are 0 or 1.
In the present context, the degree d
v
of a vertex v is dened to be:
d
v
=
u
w(u, v),
vol G =
v
d
v
.
We generalize the denitions of previous sections, so that
L(u, v) =
_
_
_
d
v
w(v, v) if u = v,
w(u, v) if u and v are adjacent,
0 otherwise.
In particular, for a function f : V R, we have
Lf(x) =
y
xy
(f(x) f(y))w(x, y).
Let T denote the diagonal matrix with the (v, v)-th entry having value d
v
. The
Laplacian of G is dened to be
L = T
1/2
LT
1/2
.
In other words, we have
L(u, v) =
_
_
1
w(v, v)
d
v
if u = v, and d
v
,= 0,
w(u, v)
_
d
u
d
v
if u and v are adjacent,
0 otherwise.
1.4. EIGENVALUES OF WEIGHTED GRAPHS 13
We can still use the same characterizations for the eigenvalues of the generalized
versions of L. For example,
G
:=
1
= inf
gT
1/2
1
g, Lg)
g, g)
(1.13)
= inf
f
P
f(x)dx=0
xV
f(x)Lf(x)
xV
f
2
(x)d
x
= inf
f
P
f(x)dx=0
xy
(f(x) f(y))
2
w(x, y)
xV
f
2
(x)d
x
.
A contraction of a graph G is formed by identifying two distinct vertices, say
u and v, into a single vertex v
are dened as
follows:
w(x, v
, v
G
H
The proof follows from the fact that an eigenfunction which achieves
H
for H
can be lifted to a function dened on V (G) such that all vertices in G that contract
to the same vertex in H share the same value.
Now we return to Lemma 1.14.
SKETCHED PROOF OF LEMMA 1.14:
Let u and v denote two vertices that are at distance D 2t +2 in G. We contract
G into a path H with 2t + 2 edges, with vertices x
0
, x
1
, . . . , x
t
, z, y
t
, . . . , y
2
, y
1
, y
0
such that vertices at distance i from u, 0 i t, are contracted to x
i
, and vertices
at distance j from v, 0 j t, are contracted to y
j
. The remaining vertices
are contracted to z. To establish an upper bound for
1
, it is enough to choose a
suitable function f, dened as follows:
f(x
i
) = a(k 1)
i/2
,
f(y
j
) = b(k 1)
j/2
,
f(z) = 0,
where the constants a and b are chosen so that
x
f(x)d
x
= 0.
14 1. EIGENVALUES AND THE LAPLACIAN OF A GRAPH
It can be checked that the Rayleigh quotient satises
uv
(f(u) f(v))
2
w(u, v)
v
f(v)
2
d
v
1
2
k 1
k
_
1
1
t + 1
_
+
1
t + 1
,
since the ratio is maximized when w(x
i
, x
i+1
) = k(k 1)
i1
= w(y
i
, y
i+1
). This
completes the proof of the lemma.
1.5. Eigenvalues and random walks
In a graph G, a walk is a sequence of vertices (v
0
, v
1
, . . . , v
s
) with
v
i1
, v
i
E(G) for all 1 i s. A random walk is determined by the transition
probabilities P(u, v) = Prob(x
i+1
= v[x
i
= u), which are independent of i. Clearly,
for each vertex u,
v
P(u, v) = 1.
For any initial distribution f : V R with
v
f(v) = 1, the distribution after k
steps is fP
k
(i.e., a matrix multiplication with f viewed as a row vector where P
is the matrix of transition probabilities). The random walk is said to be ergodic if
there is a unique stationary distribution (v) satisfying
lim
s
fP
s
(v) = (v).
It is easy to see that necessary conditions for the ergodicity of P are (i) irre-
ducibility, i.e., for any u, v V , there exists some s such that P
s
(u, v) > 0, and
(ii) aperiodicity, i.e., gcds : P
s
(u, u) > 0 = 1. As it turns out, these are also
sucient conditions. A major problem of interest is to determine the number of
steps s required for P
s
to be close to its stationary distribution, given an arbitrary
initial distribution.
We say a random walk is reversible if
(u)P(u, v) = (v)P(v, u).
An alternative description for a reversible random walk can be given by considering
a weighted connected graph with edge weights satisfying
w(u, v) = w(v, u) = (v)P(v, u)/c
where c can be any constant chosen for the purpose of simplifying the values.
(For example, we can take c to be the average of (v)P(v, u) over all (v, u) with
P(v, u) ,= 0, so that the values for w(v, u) are either 0 or 1 for a simple graph.)
The random walk on a weighted graph has as its transition probabilities
P(u, v) =
w(u, v)
d
u
,
where d
u
=
z
w(u, z) is the (weighted) degree of u. The two conditions for
ergodicity are equivalent to the conditions that the graph be (i) connected and
(ii) non-bipartite. From Lemma 1.7, we see that (i) is equivalent to
1
> 0 and
1.5. EIGENVALUES AND RANDOM WALKS 15
(ii) implies
n1
< 2. As we will see later in (1.14), together (i) and (ii) imply
ergodicity.
We remind the reader that an unweighted graph has w(u, v) equal to either 0
or 1. The usual random walk on an unweighted graph has transition probability
1/d
v
of moving from a vertex v to any one of its neighbors. The transition matrix
P then satises
P(u, v) =
_
1/d
u
if u and v are adjacent,
0 otherwise.
In other words,
fP(v) =
u
uv
1
d
u
f(u)
for any f : V (G) R.
It is easy to check that
P = T
1
A = T
1/2
(I L)T
1/2
,
where A is the adjacency matrix.
In a random walk with an associated weighted connected graph G, the transi-
tion matrix P satises
1TP = 1T
where 1 is the vector with all coordinates 1. Therefore the stationary distribution
is exactly = 1T/vol G. We want to show that when k is large enough, for any
initial distribution f : V R, fP
k
converges to the stationary distribution.
First we consider convergence in the L
2
(or Euclidean) norm. Suppose we write
fT
1/2
=
i
a
i
i
,
where
i
denotes the orthonormal eigenfunction associated with
i
.
Recall that
0
= 1T
1/2
/
vol G
since f, 1) = 1. We then have
|fP
s
| = |fP
s
1T/vol G|
= |fP
s
a
0
0
T
1/2
|
= |fT
1/2
(I L)
s
T
1/2
a
0
0
T
1/2
|
= |
i=0
(1
i
)
s
a
i
i
T
1/2
|
(1
)
s
max
x
d
x
min
y
_
d
y
e
s
max
x
d
x
min
y
_
d
y
(1.14)
16 1. EIGENVALUES AND THE LAPLACIAN OF A GRAPH
where
=
_
1
if 1
1
n1
1
2
n1
otherwise.
So, after s 1/(
log(max
x
d
x
/ min
y
_
d
y
)) steps, the L
2
distance between fP
s
and its stationary distribution is at most .
Although
occurs in the above upper bound for the distance between the
stationary distribution and the s-step distribution, in fact, only
1
is crucial in the
following sense. Note that
is either
1
or 2
n1
. Suppose the latter holds,
i.e.,
n1
1 1
1
. We can consider a modied random walk, called the lazy
walk, on the graph G
k
=
k
/2 1, which follows from equation
(1.13). Therefore,
1
1
1
n1
0,
and the convergence bound in L
2
distance in (1.14) for the modied random walk
becomes
2
1
log(
max
x
d
x
min
y
_
d
y
).
In general, suppose a weighted graph with edge weights w(u, v) has eigenvalues
i
with
n1
1 1
1
. We can then modify the weights by choosing, for some
constant c,
w
(u, v) =
_
w(v, v) +cd
v
if u = v,
w(u, v) otherwise.
(1.15)
The resulting weighted graph has eigenvalues
k
=
k
1 +c
=
2
k
n1
+
1
where
c =
1
+
n1
2
1
1
2
.
Then we have
1
1
=
n1
1 =
n1
n1
+
1
.
Since c 1/2 and we have
k
2
k
/(2 +
k
) 2
k
/3 for
k
1. In particular we
set
=
1
=
2
1
n1
+
1
2
3
1
.
Therefore the modied random walk corresponding to the weight function w
has
an improved bound for the convergence rate in L
2
distance:
1
log
max
x
d
x
min
y
_
d
y
.
We remark that for many applications in sampling, the convergence in L
2
distance seems to be too weak since it does not require convergence at each vertex.
There are several stronger notions of distance, several of which we will mention.
1.5. EIGENVALUES AND RANDOM WALKS 17
A strong notion of convergence that is often used is measured by the relative
pointwise distance (see [228]): After s steps, the relative pointwise distance of P
to the stationary distribution (x) is given by
(s) = max
x,y
[P
s
(y, x) (x)[
(x)
.
Let
x
denote the characteristic function of x dened by:
x
(y) =
_
1 if y = x,
0 otherwise.
Suppose
x
T
1/2
=
i
,
y
T
1/2
=
i
,
where
i
s denote the eigenfunction of the Laplacian L of the weighted graph asso-
ciated with the random walk. In particular,
0
=
d
x
vol G
,
0
=
1
vol G
.
Let A
x
(x)[
(x)
= max
x,y
[
y
T
1/2
(I L)
t
T
1/2
x
(x)[
(x)
max
x,y
i=0
[(1
i
)
t
i
[
d
x
/vol G
t
max
x,y
i=0
[
i
i
[
d
x
/vol G
t
max
x,y
|
x
T
1/2
| |
y
T
1/2
|
d
x
/vol G
t
vol G
min
x,y
_
d
x
d
y
e
t(1
)
vol G
min
x
d
x
where
= max
i=0
[1
i
[. So if we choose t such that
t
1
1
log
vol G
min
x
d
x
,
then, after t steps, we have (t) .
18 1. EIGENVALUES AND THE LAPLACIAN OF A GRAPH
When 1
1
,=
, we can improve the above bound by using a lazy walk as
described in (1.15). The proof is almost identical to the above calculation except
for using the Laplacian of the modied weighted graph associated with the lazy
walk. This can be summarized by the following theorem:
Theorem 1.16. For a weighted graph G, we can choose a modied random
walk P so that the relative pairwise distance (t) is bounded above by:
(t) e
t
vol G
min
x
d
x
e
2t1/(2+1)
vol G
min
x
d
x
.
where =
1
if 2
n1
+
1
and = 2
1
/(
n1
+
1
) otherwise.
Corollary 1.17. For a weighted graph G, we can choose a modied random
walk P so that we have
(t) e
c
if
t
1
_
log
vol G
min
x
d
x
+c
_
where =
1
if 2
n1
+
1
and = 2
1
/(
n1
+
1
) otherwise.
We remark that for any initial distribution f : V R with f, 1) = 1 and
f(x) 0, we have, for any x,
[fP
s
(x) (x)[
(x)
y
f(y)
[P
s
(y, x) (x)[
(x)
y
f(y)(s)
(s).
Another notion of distance for measuring convergence is the so-called total
variation distance, which is half of the L
1
distance:
TV
(s) = max
AV (G)
max
yV (G)
[
xA
(P
s
(y, x) (x)) [
=
1
2
max
yV (G)
xV (G)
[P
s
(y, x) (x)[.
The total variation distance is bounded above by the relative pointwise distance,
since
TV
(s) = max
AV (G)
volA
volG
2
max
yV (G)
[
xA
(P
s
(y, x) (x)) [
max
AV (G)
volA
volG
2
xA
(x)(s)
1
2
(s).
Therefore, any convergence bound using relative pointwise distance implies the
same convergence bound using total variation distance. There is yet another notion
1.5. EIGENVALUES AND RANDOM WALKS 19
of distance, sometimes called -squared distance, denoted by
(s) = max
yV (G)
_
_
xV (G)
(P
s
(y, x) (x))
2
(x)
_
_
1/2
max
yV (G)
xV (G)
[P
s
(y, x) (x)[
= 2
TV
(s),
using the Cauchy-Schwarz inequality.
(s) = max
xV (G)
_
_
yV (G)
(P
s
(x, y) (y))
2
(y)
_
_
1/2
max
xV (G)
(
yV (G)
((s))
2
(y))
1
2
(s).
We note that
(s)
2
x
(x)
y
(P
s
(x, y) (y))
2
(y)
=
x
T
1/2
(P
s
I
0
)T
1
(P
s
I
0
)
T
1/2
x
=
x
((I L)
2s
I
0
)
x
,
where I
0
denotes the projection onto the eigenfunction
0
,
i
denotes the ith or-
thonormal eigenfunction of L and
x
denotes the characteristic function of x. Since
x
=
i
(x)
i
,
we have
(s)
2
x
((I L)
2s
I
0
)
x
(1.16)
=
x
(
i
(x)
i
)((I L)
2s
I
0
)(
i
(x)
i
)
i=0
2
i
(x)(1
i
)
2s
=
i=0
2
i
(x)(1
i
)
2s
=
i=0
(1
i
)
2s
.
Equality in (1.16) holds if, for example, G is vertex-transitive, i.e., there is an
automorphism mapping u to v for any two vertices in G (for more discussion, see
Chapter 7 on symmetrical graphs). Therefore, we conclude:
20 1. EIGENVALUES AND THE LAPLACIAN OF A GRAPH
Theorem 1.18. Suppose G is a vertex transitive graph. Then a random walk
after s steps converges to the uniform distribution under total variation distance or
-squared distance in a number of steps bounded by the sum of (1
i
)
2s
, where
i
ranges over the non-trivial eigenvalues of the Laplacian:
TV
(s)
1
2
(s) =
1
2
(
i=0
(1
i
)
2s
)
1/2
. (1.17)
The above theorem is often derived from the Plancherel formula. Here we have
employed a direct proof. We remark that for some graphs which are not vertex-
transitive, a somewhat weaker version of (1.17) can still be used with additional
work (see [75] and the remarks in Section 4.5). Here we will use Theorem 1.18 to
consider random walks on an n-cube.
Example 1.19. For the n-cube Q
n
, our (lazy) random walk (as dened in
(1.15)) converges to the uniform distribution under the total variation distance,
as estimated as follows: From Example 1.6, the eigenvalues of Q
n
are 2k/n with
multiplicity
_
n
k
_
for k = 0, . . . , n. The adjusted eigenvalues for the weighted graph
corresponding to the lazy walk are
k
= 2
k
/(
n1
+
1
) =
k
n/(n +1). By using
Theorem 1.18 (also see [101]), we have
TV
(s)
1
2
(s)
1
2
(
n
k=1
_
n
k
_
(1
2k
n + 1
)
2s
)
1/2
1
2
(
n
k=1
e
k log n
4ks
n+1
)
1/2
e
c
if s
1
4
nlog n +cn.
We can also compute the rate of convergence of the lazy walk under the rela-
tive pointwise distance. Suppose we denote vertices of Q
n
by subsets of an n-set
1, 2, . . . , n. The orthonormal eigenfunctions are
S
for S 1, 2, . . . , n where
S
(X) =
(1)
|SX|
2
n/2
for any X 1, 2, . . . , n. For a vertex indexed by the subset S, the characteristic
function is denoted by
S
(X) =
_
1 if X = S,
0 otherwise.
Clearly,
X
=
S
(1)
|SX|
2
n/2
S
.
1.5. EIGENVALUES AND RANDOM WALKS 21
Therefore,
[P
s
(X, Y ) (Y )[
(Y )
= [2
n
X
P
s
Y
1[
[2
n
X
P
s
X
1[
=
S=
(1
2[S[
n + 1
)
s
=
n
k=1
_
n
k
_
(1
2k
n + 1
)
s
.
This implies
(s) =
n
k=1
_
n
k
_
(1
2k
n + 1
)
s
k=1
e
k log n
2ks
n+1
e
c
if
s
nlog n
2
+cn.
So, the rate of convergence under relative pointwise distance is about twice
that under the total variation distance for Q
n
.
In general,
TV
(s),
1
. This we will leave as an exercise. We remark that Aldous [4] has shown that
if
TV
(s) , then P
s
(y, x) c
depends only on
.
Notes
For an induced subgraph of a graph, we can dene the Laplacian with boundary
conditions. We will leave the denitions for eigenvalues with Neumann boundary
conditions and Dirichlet boundary conditions for Chapter 8.
The Laplacian for a directed graph is also very interesting. The Laplacian for
a hypergraph has very rich structures. However, in this book we mainly focus on
the Laplacian of a graph since the theory on these generalizations and extensions
is still being developed.
In some cases, the factor log
vol G
minx dx
in the upper bound for (t) can be further
reduced. Recently, P. Diaconis and L. Salo-Coste [103] introduced a discrete ver-
sion of the logarithmic Sobolev inequalities which can reduce this factor further for
certain graphs (for