You are on page 1of 11

CSE 713: Random Graphs and Applications Lecturer: Hung Q.

Ngo
SUNY at Buffalo, Fall 2003 Scribe: Hung Q. Ngo

Definition 0.1. A bipartite graph G = (I ∪ O; E) is called an (n, d, c)-expander if |I| = |O| = n,

∆(G) = d, and for every X ⊂ I such that |X| ≤ n/2, we have
  
|X|
|Γ(X)| ≥ 1 + c 1 − |X|
n

where Γ(X) is the set of neighbors of all vertices X, ∆(G) denotes the maximum degree of vertices in
G, d is called the density, and c is called the expansion rate of the expander G. It is a strong (n, d, c)-
expander if the above inequality holds for all X ⊆ I. A family of linear expanders of density d and
expansion c is a sequence {Gi }∞ i=1 , where Gi is an (ni , d, c)-expander and ni → ∞, ni+1 /ni → 1 as
i → ∞.

1 Expanders
The main objective of this section is to present several results on the eigenvalue characterization of the
expansion rate of expanders. Moreover, we shall give a partial answer to the question: “how large can
the second smallest eigenvalue of the Laplacian of a graph be?”, and give some pointers to results on
direct constructions of expanders.

1.1 Algebraic graph theory

We first fix notations and terminologies from algebraic graph theory needed for the rest of the section.
The reader is referred to the standard books by Biggs (1993, [10]), Godsil (1993, [15]), Cvetković, Doob,
Let G = (V, E) be a simple graph with n vertices, we will always assume the eigenvalues of G are
λ1 ≥ λ2 ≥ · · · ≥ λn . Those are the eigenvalues of the adjacency matrix A of G. Moreover, λ(G) will be
used to denote λ2 . Let D be the n × n diagonal matrix indexed by vertices of G with (D)vv = deg(v);
then, the matrix L := D − A is called the Laplacian matrix of G. We shall use µ1 ≥ µ2 ≥ · · · ≥ µn to
denote the eigenvalues of L. In contrast to the λ’s, we use µ(G) to denote µn−1 . The reason for this is
that when G is d-regular then λi = d − µn−i+1 .
Let N be the incident matrix of any orientation H of G(V, E). Let L2 (V ) (L2 (E)) be thep space of
real valued functions on V (E), with the usual inner product hf, gi and the usual norm kf k = hf, f i.
Note that L2 (V ) is isomorphic to Rn and the Rayleigh quotient for f is RA (f ) = hLf,f
kf k2
i
. Also note that

hLf, f i = hN T N f, f i = hN f, N f i
X
= (f (u) − f (v))2
(u,v)∈E(H)
X
= (f (u) − f (v))2
u∼v

1
Here u ∼ v is used in lieu of (u, v) ∈ E for convenience. The previous equation implies that L is non-
negative definite, thus µi ≥ 0, ∀i. Moreover, it is not hard to see that µn is always 0 and that µn−1 = 0
iff G is not connected.
The following bounds on µn−1 will be used quite often.

P

X
(f (u) − f (v))2
hLf, f i
µ(G) ≤ = u∼v X
kf k f 2 (v)
v

In fact, a stronger statement holds

hLf, f i
µ(G) = min
f 6=0 kf k
P
with the min runs over all f satisfying v f (v) = 0.
X
Note. (f (u) − f (v))2 is sometime called the Dirichlet sum of G.
u∼v

Proof. Let un = 1/ n be a unit µn -eigenvector of L, then the variational characterization of the eigen-
values (see [16], e.g.) gives

hLf, f i
µn−1 = min n RL (f ) = min n
06=f ∈C 06=f ∈C kf k
f ⊥un f ⊥1
P
The condition f ⊥ 1 is the same as u f (u) = 0.

Proposition 1.2. Suppose G is connected, and f ∈ L2 (V ) is a µ(G)-eigenvector. Let V + := {v ∈

V | f (v) > 0} and V − := V − V + . Let g ∈ L2 (V ) be defined by
(
f (v) if v ∈ V +
g(v) =
0 otherwise.

Then, we have X
(g(u) − g(v))2
u∼v
µ≥ X
g 2 (v)
v

Proof. Note that since G is connected, µ 6= 0, making f 6= 0. Hence, V + 6= ∅. By definition, we have

(Lf )(v) = µf (v), ∀v ∈ V . Thus, X
(Lf )(v)f (v)
v∈V +
µ= X
f 2 (v)
v∈V +

But, X X
f 2 (v) = g 2 (v)
v∈V + v∈V

2
and,
 
X X X
(Lf )(v)f (v) = d(v)f 2 (v) − f (v)f (u)
v∈V + v∈V + u∈Γ(v)
X X
= (f (u) − f (v))2 + f (u)(f (u) − f (v))
uv∈E(V +) uv∈E(V + ,V − )
X
2
≥ (g(u) − g(v))
u∼v

which completes our proof.

1.2 The eigenvalue characterization of expansion rate for regular strong expanders
The results in this section are from Tanner (1984, [28]), Alon and Milman (1985, [6]), and Alon (1986,
[3]). As the title of the section indicated, our goal is to show that the larger the expansion rate of a regular
strong expander G is, the larger µ(G) has to be, and vice versa.
We again need to define several types of graphs closely related to expanders: enlargers, magnifiers,
and bounded concentrators.

Definition 1.3. An (n, d, )-enlarger is a graph G on n vertices with maximum degree d and µn−1 (G) ≥
.

Definition 1.4. An (n, d, c)-magnifier is a graph G = (V, E) on n vertices, ∆(G) = d and for every
X ⊂ V with |X| ≤ n/2, |Γ(X) − X| ≥ c|X| holds. The extended double cover of a graph G =
(V, E) with V = [n] is the bipartite graph H on the sets of inputs X = {x1 , . . . , xn } and outputs
Y = {y1 , . . . , yn } so that (xi , yj ) ∈ E(H) iff i = j or (i, j) ∈ E(G).

Definition 1.5. An (n, θ, d, α, c)-bounded strong concentrator is a bipartite graph G = (I, O; E) with n
inputs, θn outputs, θ < 1, and at most dn edges, such that if X ⊆ I with |X| ≤ αn, then |Γ(X)| ≥ c|X|.
An (n, θ, d, α)-bounded concentrator is an (n, θ, d, α, 1)-bounded strong concentrator.

For the purpose of this section, enlargers were introduced just to shorten the sentence: “let G be a
graph with maximum degree d and µ(G) large enough.” On the other hand, we need magnifiers because
their extended double covers are expanders. Magnifiers are, in a sense, the non-bipartite version of
expanders. It is intuitively clear, and will be made precise later, that bounded concentrators are closely
related to expanders. The proof of the following lemma is straightforward, hence omitted.

Lemma 1.6. The double cover of an (n, d, c)-magnifier is an (n, d + 1, c)-expander.

The following theorem, which is an improved version of Theorem 3.4 in [3], makes our goal precise.

Theorem 1.7. Let G = (I, O; E) be a d-regular bipartite graph, where |I| = |O| = n and µ = µ(G)
(= d − λ(G)). The following hold:
c2
(i) If G is an (n, d, c) strong expander then µ ≥ 576−48c+2c2
.
2d−2
(ii) If µ ≥ , then G is an (n, d, c)-expander, where c = d2
.

We shall prove this theorem using a sequence of lemmas, with the following plan in mind. To show
(i), we will show

G is a strong expander → G is a magnifier → G is an enlarger (i.e. µ(G) is large)

3
Similarly, to prove (ii) we shall show the reverse sequence
G is an enlarger → G is a strong expander
Let us now proceed to show that a strong expander is a magnifier.
c
Lemma 1.8. Let G = (I, O; E) be an (n, d, c)-strong-expander. Then G is a (2n, d, 12−c )-magnifier.
Proof. Be definition, every X1 ⊂ I satisfies
 
|X1 |
|Γ(X1 )| ≥ 1 + c(1 − ) |X1 |. (1)
n
In particular,
|Γ(X1 )| ≥ |X1 | (2)
When |X1 | = n/2, (1) implies c ≤ 2. Moreover, for every X2 ⊆ O, setting X1 = I − Γ(X2 ) in (2)
yields Γ(X2 ) ≥ |X2 |.
We are to show that for every X ⊂ I ∪ O with size at most n, it must be the case that
c
|Γ(X) − X| ≥ |X|
12 − c
Let X1 = X ∩ I and X2 = X ∩ O. The intuition is that when |X1 | is very small comparing to |X2 |, then
there will be a lot of neighbors of X2 lying outside of X1 in I, making |Γ(X) − X| large. On the other
hand, when |X1 | is relatively large comparing to |X2 |, then there will be “many” neighbors of X1 not in
X2 . We now turn this intuition into mathematical rigor.
When |X1 | ≤ |X2 |(1 − 6c ), we have
12 − c
|X| ≤ |X2 |.
6
Hence,
c c
|Γ(X) − X| ≥ |Γ(X2 )| − |X1 | ≥ |X2 | − |X1 | ≥ |X2 | ≥ |X|.
6 12 − c
When |X2 |(1 − 6c ) < |X1 | ≤ n2 , we get
 
1 12 − c
|X| < 1 + |X1 | = |X1 |.
1 − 6c 6−c
This relation, the fact that c ≤ 2, and (1) yield
|Γ(X) − X| ≥ |Γ(X1 )| − |X2 |
c 1
≥ (1 + − )|X1 |
2 1 − 6c
4−c c
= |X1 |
2 6−c
4−c c
> |X|
2 12 − c
c
≥ |X|
12 − c
n
Lastly, when |X1 | ≥ 2 it follows that
c n n c c c
|Γ(X) − X| ≥ |Γ(X1 )| − |X2 | ≥ (1 + ) − = n ≥ |X| ≥ |X|.
2 2 2 4 4 12 − c
Here, we use the fact that c ≤ 2 and that the function f (x) = (1 + c(1 − nx ))x is increasing when
n ≥ 2.

4
Next, to complete part (i) of Theorem 1.7 we show that every magnifier has “large” µ.
Lemma 1.9. Let G = (V, E) be an (n, d, c)-magnifier, then G is an (n, d, )-enlarger where  =
c2
2+2(c+1)2
.
c2
Proof. We apply and use notations of Proposition 1.2. Since we have to show that µ ≥  = 2+2(c+1)2
, it
suffices to show that X
(g(u) − g(v))2
u∼v c2

2 + 2(c + 1)2
X
g 2 (v)
v
This is done by using the maxflow-mincut theorem. Consider a network N with vertex set {s} ] V + ]
V ] {t}, where s is the source, t is the sink, and ] denotes the disjoint union. The edges and their
capacities are defined as follows.
(a) For each u ∈ V + , (s, u) has capacity cap(s, u) = 1 + c.
(
1 if uv ∈ E or u = v
(b) For each pair (u, v) ∈ V + × V , cap(u, v) =
0 otherwise.

(c) For each v ∈ V , cap(v, t) = 1.

We claim that the min-cut of N has capacity (1 + c)|V + |. Consider any cut C. Let X := {u ∈
V + | (s, u) ∈/ C}. As G has magnifying rate c and ΓN (X) = X ∪ ΓG (X), it is easy to see that
|ΓN (X)| ≥ (1 + c)|X|. Moreover, for every v ∈ ΓN (X) there must be at least one edge with capacity
one in C incident to it. All these edges are disjoint, thus the capacity of C is at least (1 + c)(|V + | −
|X|) + |ΓN (X)| ≥ (1 + c)|V + |. Lastly, the cut {(s, u) | u ∈ V + } has capacity exactly (1 + c)|V + |,
proving the claim. By the maxflow-mincut theorem, there exists an orientation Ē of edges in G and a
function h : Ē → R such that
0 ≤ h(u, v) ≤ 1 for all (u, v) ∈ Ē
(
X 1 + c if u ∈ V +
h(u, v) =
0 otherwise.
v:(u,v)∈Ē
X
h(u, v) ≤ 1 for all v ∈ V
u:(u,v)∈Ē

Now, the following is straightforward:

X 2 X
h2 (u, v) g(u) + g(v) h2 (u, v) g 2 (u) + g 2 (v)

≤ 2
(u,v)∈Ē (u,v)∈Ē
X  X X 
2 2 2
= 2 g (v) h (u, v) + h (v, u)
v∈V u:(u,v)∈Ē u:(v,u)∈Ē
X
2 2
≤ 2(1 + (1 + c) ) g (v)
v∈V

and
X X  X X 
2 2 2

h(u, v) g (u) − g (v) = g (u) h(u, v) − h(v, u)
(u,v)∈Ē u∈V v:(u,v)∈Ē v:(v,u)∈Ē
X
2
≥ c g (v)
v∈V

5
Now, combining all inequalities above along with Cauchy-Schwarz inequality we get
X 2
g(u) − g(v)
u∼v
µ ≥ X
g 2 (v)
v
X 2 X 2 2
g(u) − g(v) h (u, v) g(u) + g(v)
u∼v (u,v)∈Ē
= X X 2
g 2 (v) h2 (u, v) g(u) + g(v)
v (u,v)∈Ē
 X 2
2 2
h(u, v)|g (u) − g (v)|
(u,v)∈Ē
≥ X 2
2(1 + (c + 1)2 ) g 2 (v)
v∈V
 X 2
h(u, v)(g 2 (u) − g 2 (v))
1  (u,v)∈Ē 

 
2 + 2(c + 1)2
 X 
 g 2 (v) 
v∈V

c2

2 + 2(c + 1)2

The previous two lemmas trivially imply part (i) of Theorem 1.7. Now we are ready to complete
part (ii) of Theorem 1.7. We first need a lemma from [28]. Let G = (I, O; E) be a bipartite graph
such that |I| = n, |O| = m, and that deg(i) = a, ∀i ∈ I, deg(o) = b, ∀o ∈ O. Let M be an n × m
01-matrix whose rows are indexed by I and whose columns are indexed by O such that mio = 1 if
io ∈ E and mio = 0 otherwise. Let B = M M T , then clearly B is real, symmetric, and non-negative
definite. As usual, we let θ1 ≥ θ2 ≥ · · · ≥ θn be the eigenvalues of B and u1 , u2 , . . . , un be a set of
P orthonormal eigenvectors. Notice that (B)ij is the number of common neighbors of i and
corresponding

j, hence j (B)ij = ab, ∀i. (Each neighbor of i is counted b times in the sum.) This implies 1/ n
is an eigenvector of B with corresponding eigenvalue ab. Suppose θ is any eigenvalue of B and e is a
θ-eigenvector. Let ei be a coordinate of e with highest absolute value, then (Be)i = θei implies
X X
θ|ei | = | (B)ij ej | ≤ |ei | (B)ij = ab|ei |
j j

Consequently, ab is also the largest eigenvalue of B, i.e. θ1 = ab. We may thus assume that u1 = 1/ n.
Lemma 1.10. If θ1 > θ2 , then G is an (n, m/n, a, α, c(α))-bounded strong concentrator, where
a2
c(α) ≥
α(ab − θ2 ) + θ2
Proof. Let β be a positive real number not exceeding α. For any X ⊆ I with |X| = βn, let x ∈ {0, 1}n
be X’s characteristic vector. Clearly xxT = kxk2 = βn. Similarly, let Y = Γ(X) and y be its
characteristic vector. As for any o ∈ O, (xM )o is the number of vertices in X adjacent to o, the sum of
entries in xM is exactly βna. Hence,
2
β 2 n2 a2
P
o∈O (xM )o
X
2 2
kxM k = (xM )o ≥ |Y | = (3)
|Y | |Y |
o∈O

6
Now, write x = γ1 u1 + γ2 u2 + · · · + γn un we get
xB = γ1 θ1 u1 + γ2 θ2 u2 + · · · + γn θn un .
Thus, by orthonormality we have
kxM k2 = (xB)xT = θ1 γ12 + θ2 γ22 + · · · + θn γn2 (4)
Now,

P
xi
γ1 = xuT1 = √i = β n
n
Hence, (4) gives
n
X
2
kxM k = θ1 γ12 + θj γj2
j=2

≤ abβ n + θ2 (kxk2 − γ12 )

2

= β 2 n(ab − θ2 ) + θ2 βn (5)
Lastly, combining (3) and (5) yield
|Γ(X)| |Y | a2
= ≥
|X| βn β(ab − θ2 ) + θ2
As this inequality is true for all β ≤ α, the proof is completed.

Part (ii) of Theorem 1.7 could now be derived as a corollary of this lemma.
Corollary 1.11. If G = (I, O; E) is a d-regular bipartite graph with |I| = |O| = n and µ = µ(G), then
(i) G is an (n, d, c)-strong expander, where
2dµ − µ2
c=
d2
(ii) G is an (n, d, c)-expander, where
2dµ − µ2
c=
d2 − dµ + µ2 /2
Proof. Let M and B be the matrices for G as in Lemma 1.10, and let A be G’s adjacency matrix. Also
let θ1 ≥ · · · ≥ θn be the eigenvalues of B and λ1 ≥ · · · ≥ λ2n be the eigenvalues of A as usual. Since
(B)ij = (M M T )ij counts the number of common neighbors of i and j, while (A2 )ij counts the number
of length-2 paths from i to j, it is obvious that
 
B 0
= A2
0 B
It is standard that as G is bipartite, whenever λ is an eigenvalue of A, then so is −λ. Moreover, λ1 = d
and λ2 = d − µ since G is d-regular. Thus, θ1 = d2 and θ2 = (d − µ)2 . Now, for any X ⊆ I with
α = |X|/n, applying Lemma 1.10 gives
d2
|Γ(X)| ≥  |X|
α d2 − (d − µ)2 + (d − µ)2
(2dµ − µ2 )(1 − α)
 
= 1+ 2 |X|
d − (2dµ − µ2 )(1 − α)
(2dµ − µ2 )
  
|X|
≥ 1+ 2
1− |X|
d n

7
When |X| ≤ n/2, we get

(2dµ − µ2 )
  
|X|
|Γ(X)| ≥ 1+ 2 1− |X|
d − dµ + µ2 /2 n

1.3 On the second eigenvalue of a graph

As we have seen, the expanding rate of a graph increases as its second smallest Laplacian eigenvalue
increases. Thus, it makes sense to study how large µn−1 can be. Through out this section, we shall
consider only d-regular graphs and let µ = µ(G) = µn−1 (G) = d − λ2 (G). Alon and Boppana
(mentioned in [4]) proved that for any fixed d and for any infinite family of graphs G with maximum
degree d, √
lim sup µ(G) ≤ d − 2 d − 1
The bound is sharp when d−1 is a prime congruent to 1 modulo 4, as shown by the explicit constructions
of Lubotzky, Phillips and Sarnak [20], and independently by Margulis [22]. Alon [4] conjectured the
following

Conjecture 1.12 (Alon [4]). As n → ∞, the probability that µ(G) ≤ d − 2 d − 1 − o(1) goes to 1.

In other words, as n gets large, with very high probability we have µ(G) ≤ d − 2 d − 1. The
conjecture is still open as far as we know. Friedman, Kahn and Szemeédi (1989, [13]) showed that

µ(G) ≥ d − 2 d − 1 − log d − o(1)

as n → ∞. Nilli (1991, [26]) got an upper bound that does not have any hidden constant:

Theorem 1.13 (Nilli, 1991 [26]). Let G be a graph in which there are two edges with distance at least
2k + 2, and let d be the maximum degree of G. Then,

√ 2 d−1−1
µ(G) ≤ d − 2 d − 1 +
k+1
Proof. Let (u1 , u2 ) and (w1 , w2 ) be two edges with distance at least 2k + 2. Let L be the Laplacian
matrix of G as usual. Let U0 = {u1 , u2 } and W0 = {w1 , w2 }. For 1 ≤ i ≤ k, let Ui (resp. Wi ) be
the set of vertices of distance i from U0 (resp. W0 ). Clearly U := ∪0≤i≤k Ui has distance at least 2
from W := ∪0≤i≤k Wi , so that there is no edge joining the two unions. Moreover, it is easy to see that
|Ui | ≤ (d − 1)|Ui−1 | and |Wi | ≤ (d − 1)|Wi−1 |.
Let f : V (G) → R be defined as

−i/2
a(d − 1)
 for v ∈ Ui , 0 ≤ i ≤ k
f (v) = b(d − 1) −i/2 for v ∈ Wi , 0 ≤ i ≤ k

0 otherwise.

P
where a > 0, b < 0 are real numbers such that v f (v) = 0. The variational characterization yields

hLf, f i
µ(G) = µn−1 ≤
hf, f i

8
We also have
X X
hf, f i = f 2 (u) + f 2 (w)
u∈U w∈W
k k
X a2 X b2
= |Ui | + |Wi |
(d − 1)i (d − 1)i
i=0 i=0
= A1 + B1
where A1 and B1 are the first and second sum, respectively. Moreover, as we have mentioned: (a) there
are no edges joining U and W ; (b) there are no edges connecting Ui or Wi to V (G)−U ∪W if i ≤ k −1;
and (c) there are at most d − 1 edges joining a vertex of Ui (Wi ) to a vertex of Ui+1 (Wi+1 ), we have
X
hLf, f i = (f (u) − f (v))2
u∼v
X X
= (f (u) − f (v))2 + (f (u) − f (v))2
u∼v u∼v
{u,v}∩U 6=∅ {u,v}∩W 6=∅
k−1
X X k−1
X X
= (f (u) − f (v))2 + (f (u) − f (v))2
i=0 u∼v i=0 u∼v
u∈Ui ,v∈Ui+1 u∈Wi ,v∈Wi+1
X X
+ (f (u))2 + (f (u))2
u∼v u∼v
u∈Uk ,v∈V −U ∪W u∈Wk ,v∈V −U ∪W
k−1  2 !
2
X 1 1 d−1
≤ a |Ui |(d − 1) − + |Uk | +
(d − 1)i/2 (d − 1)(i+1)/2 (d − 1)k
i=0
k−1  2 !
X 1 1 d−1
b2 |Wi |(d − 1) − + |Wk |
(d − 1)i/2 (d − 1)(i+1)/2 (d − 1)k
i=0
k
!
X |Ui | √ √ |Uk |
= a2 (d − 2 d − 1) + (2 d − 1 − 1) +
(d − 1)i (d − 1)k
i=0
k
!
X |Wi | √ √ |Wk |
b2 (d − 2 d − 1) + (2 d − 1 − 1)
(d − 1)i (d − 1)k
i=0
= A2 + B2
with A2 and B2 defined in the obvious way. To this end, we are left to show that

A2 + B2 √ 2 d−1−1
≤d−2 d−1+ =: C
A1 + B1 k+1
|Ui | |Ui+1 |
We shall show that A2 /A1 ≤ C and B2 /B1 ≤ C instead. Notice that (d−1)i
≥ (d−1)i+1
, clearly
k
X a2 |Uk |
A1 = |Ui | i
≥ a2 (k + 1)
(d − 1) (d − 1)k
i=0

hence, √
A2 √ 2 2 d − 1 − 1 |Wk |
=d−2 d−1+a ≤C
A1 A1 (d − 1)k
B2 /B1 ≤ C is proved similarly.

9
Corollary 1.14. Let G, d and k be defined as in the previous theorem, If G is d regular, then

 
1 1
λ2 (G) ≥ 2 d − 1 1 − +
k+1 k+1

1.4 Explicit Constructions of Expanders

Many practical applications require explicit constructions of expander graphs. Explicit constructions
turn out to be a lot more difficult than showing existence. In 1973, Margulis [21] gave the first explicit
construction of (strong) expanders. He explicitly constructed a family of bipartite graphs {Ĝn } for
n = m2 , m = 1, 2, . . . , and show that there is a constant k > 0 such that for each n = m2 , m ∈ N,
Ĝn is an (n, 5, k)-strong expander. This construction was certainly undesirable as the constant k was
not known. Moreover, his proof used deep results from Representation Theory. Angluin (1979, [8])
pointed out that Margulis’ method could be used to construct (n, 3, k 0 )-strong expanders but the constant
k 0 is also not known. Gabber and Galil (1981, [14]) slightly modified Margulis’ construction√ and used
relatively elementary Taylor analysis to show how to construct a family of (m 2 , 5, (2 − 3)/4)-strong

expanders for each m ∈ N. They also constructed a family of (m2 , 7, (2 − 3)/2)-strong expanders.
Let us mention here their first construction. For each m ∈ N, construct an m2 × m2 bipartite graph
Ḡm2 = (Im , Om ; E) where Im = Om = Zm × Zm and each vertex (i, j) ∈ Im is connected to 5
vertices in Om defined by the following permutations:

σ1 (i, j) = (i, j)
σ2 (i, j) = (i, i + j)
σ3 (i, j) = (i, i + j + 1)
σ4 (i, j) = (i + j, j)
σ5 (i, j) = (i + j + 1, j)

Here, the additions are done modulo m. As we have mentioned, there is no known elementary proof
that this family are expanders with the prescribed expansion rate. This construction was later modified
slightly by Jimbo and Maruoka (1987, [17]) and Alon, Galil and Milman (1987, [5]) to obtain better
superconcentrators.
As we have discussed in the introduction, after the eigenvalue characterization of expansion rate, the
main problem was to construct Ramanujan graphs, which are optimal expanders in the eigenvalue sense.
The special case where d − 1 is a prime congruent to 1 modulo 4 was “solved” by Lubotzky, Phillips, and
Sarnak (1988, [20]), and independently by Margulis (1988, [22]). Later, in 1994 Morgenstern [23, 24]
constructed for every prime power q many families of (q + 1)-regular Ramanujan graphs. All these
constructions were Cayley graphs of certain groups.
Other works and information on expanders could be found in [1, 2, 7, 9, 11, 18, 19, 27].

References
[1] M. A JTAI, Recursive construction for 3-regular expanders, Combinatorica, 14 (1994), pp. 379–416.

[2] M. A JTAI , J. KOML ÓS , AND E. S ZEMER ÉDI, Generating expanders from two permutations, in A tribute to Paul Erdős,
Cambridge Univ. Press, Cambridge, 1990, pp. 1–12.

[3] N. A LON, Eigenvalues and expanders, Combinatorica, 6 (1986), pp. 83–96. Theory of computing (Singer Island, Fla.,
1984).

[4] , Eigenvalues, geometric expanders, sorting in rounds, and Ramsey theory, Combinatorica, 6 (1986), pp. 207–219.

[5] N. A LON , Z. G ALIL , AND V. D. M ILMAN, Better expanders and superconcentrators, J. Algorithms, 8 (1987), pp. 337–
347.

10
[6] N. A LON AND V. D. M ILMAN, λ1 , isoperimetric inequalities for graphs, and superconcentrators, J. Combin. Theory
Ser. B, 38 (1985), pp. 73–88.

[7] N. A LON AND Y. ROICHMAN, Random Cayley graphs and expanders, Random Structures Algorithms, 5 (1994), pp. 271–
284.

[8] D. A NGLUIN, A note on a construction of Margulis, Inform. Process. Lett., 8 (1979), pp. 17–19.

[9] F. B IEN, Constructions of telephone networks by group representations, Notices Amer. Math. Soc., 36 (1989), pp. 5–22.

[10] N. B IGGS, Algebraic graph theory, Cambridge University Press, Cambridge, second ed., 1993.

[11] F. R. K. C HUNG, Spectral graph theory, Published for the Conference Board of the Mathematical Sciences, Washington,
DC, 1997.

[12] D. M. C VETKOVI Ć , M. D OOB , AND H. S ACHS, Spectra of graphs, Johann Ambrosius Barth, Heidelberg, third ed.,
1995. Theory and applications.

[13] J. F RIEDMAN , J. K AHN , AND E. S ZEMER ÉDI, On the second eigenvalue in random regular graphs, in Proceedings of
the 21st ACM STOC, 1989, pp. 587–598.

[14] O. G ABBER AND Z. G ALIL, Explicit constructions of linear size superconcentrators, in 20th Annual Symposium on
Foundations of Computer Science (San Juan, Puerto Rico, 1979), IEEE, New York, 1979, pp. 364–370.

[15] C. D. G ODSIL, Algebraic combinatorics, Chapman & Hall, New York, 1993.

[16] R. A. H ORN AND C. R. J OHNSON, Matrix analysis, Cambridge University Press, Cambridge, 1985.

[17] S. J IMBO AND A. M ARUOKA, Expanders obtained from affine transformations, Combinatorica, 7 (1987), pp. 343–355.

[18] N. K AHALE, Eigenvalues and expansion of regular graphs, J. Assoc. Comput. Mach., 42 (1995), pp. 1091–1106.

[19] A. L UBOTZKY, Cayley graphs: eigenvalues, expanders and random walks, in Surveys in combinatorics, 1995 (Stirling),
Cambridge Univ. Press, Cambridge, 1995, pp. 155–189.

[20] A. L UBOTZKY, R. P HILLIPS , AND P. S ARNAK, Ramanujan graphs, Combinatorica, 8 (1988), pp. 261–277.

[21] G. A. M ARGULIS, Explicit constructions of expanders, Problemy Peredači Informacii, 9 (1973), pp. 71–80.

[22] , Explicit group-theoretic constructions of combinatorial schemes and their applications in the construction of
expanders and concentrators, Problemy Peredachi Informatsii, 24 (1988), pp. 51–60.

[23] M. M ORGENSTERN, Existence and explicit constructions of q + 1 regular Ramanujan graphs for every prime power q,
J. Combin. Theory Ser. B, 62 (1994), pp. 44–62.

[24] , Ramanujan diagrams, SIAM J. Discrete Math., 7 (1994), pp. 560–570.

[25] H. Q. N GO AND D.-Z. D U, Notes on the complexity of switching networks, in Advances in Switching Networks, D.-Z.
Du and H. Q. Ngo, eds., Kluwer Academic Publishers, 2001, pp. 307–367.

[26] A. N ILLI, On the second eigenvalue of a graph, Discrete Math., 91 (1991), pp. 207–210.

[27] Y. ROICHMAN, Expansion properties of Cayley graphs of the alternating groups, J. Combin. Theory Ser. A, 79 (1997),
pp. 281–297.

[28] R. M. TANNER, Explicit concentrators from generalized N -gons, SIAM J. Algebraic Discrete Methods, 5 (1984),
pp. 287–293.

11