INTRODUCTION TO RANDOM

GRAPHS

ALAN FRIEZE and MICHAŁ KAROŃSKI

To Carol and Jola

3
 Contents

Preface page vi

PART ONE BASIC MODELS 1

1 Random Graphs 3
1.1 Models and Relationships 3
1.2 Thresholds and Sharp Thresholds 9
1.3 Pseudo-Graphs 16
1.4 Exercises 18
1.5 Notes 18
2 Evolution 20
2.1 Sub-Critical Phase 20
2.2 Super-Critical Phase 33
2.3 Phase Transition 39
2.4 Exercises 45
2.5 Notes 47
3 Vertex Degrees 49
3.1 Degrees of Sparse Random Graphs 49
3.2 Degrees of Dense Random Graphs 54
3.3 Exercises 61
3.4 Notes 62
4 Connectivity 64
4.1 Connectivity 64
4.2 k-connectivity 68
4.3 Exercises 69
4.4 Notes 70

i
ii Contents

5 Small Subgraphs 72
5.1 Thresholds 72
5.2 Asymptotic Distributions 76
5.3 Exercises 78
5.4 Notes 79
6 Spanning Subgraphs 82
6.1 Perfect Matchings 82
6.2 Hamilton Cycles 89
6.3 Long Paths and Cycles in Sparse Random Graphs 94
6.4 Greedy Matching Algorithm 96
6.5 Random Subgraphs of Graphs with Large Minimum
Degree 100
6.6 Spanning Subgraphs 103
6.7 Exercises 105
6.8 Notes 108
7 Extreme Characteristics 111
7.1 Diameter 111
7.2 Largest Independent Sets 117
7.3 Interpolation 121
7.4 Chromatic Number 123
7.5 Eigenvalues 129
7.6 Exercises 135
7.7 Notes 136
8 Extremal Properties 140
8.1 Containers 140
8.2 Ramsey Properties 141
8.3 Turán Properties 143
8.4 Containers and the proof of Theorem 8.1 145
8.5 Exercises 153
8.6 Notes 154
9 Resilience 155
9.1 Perfect Matchings 155
9.2 Hamilton Cycles 156
9.3 The chromatic number 167
9.4 Exercises 168
9.5 Notes 168

PART TWO BASIC MODEL EXTENSIONS 171

 Contents iii

10 Inhomogeneous Graphs 173

10.1 Generalized Binomial Graph 173
10.2 Expected Degree Model 180
10.3 Kronecker Graphs 187
10.4 Exercises 192
10.5 Notes 193
11 Fixed Degree Sequence 197
11.1 Configuration Model 197
11.2 Connectivity of Regular Graphs 208
11.3 Existence of a giant component 211
11.4 Gn,r is asymmetric 216
11.5 Gn,r versus Gn,p 219
11.6 Exercises 230
11.7 Notes 231
12 Intersection Graphs 233
12.1 Binomial Random Intersection Graphs 233
12.2 Random Geometric Graphs 243
12.3 Exercises 252
12.4 Notes 253
13 Digraphs 258
13.1 Strong Connectivity 258
13.2 Hamilton Cycles 266
13.3 Exercises 268
13.4 Notes 270
14 Hypergraphs 271
14.1 Component Size 271
14.2 Hamilton Cycles 276
14.3 Thresholds 280
14.4 Exercises 293
14.5 Notes 296

PART THREE OTHER MODELS 299

15 Trees 301
15.1 Labeled Trees 301
15.2 Recursive Trees 308
15.3 Inhomogeneous Recursive Trees 318
15.4 Exercises 328
iv Contents

15.5 Notes 329

16 Mappings 333
16.1 Permutations 333
16.2 Mappings 336
16.3 Exercises 343
16.4 Notes 344
17 k-out 346
17.1 Connectivity 346
17.2 Perfect Matchings 349
17.3 Hamilton Cycles 358
17.4 Nearest Neighbor Graphs 361
17.5 Exercises 365
17.6 Notes 365
18 Real World Networks 368
18.1 Preferential Attachment Graph 368
18.2 A General Model of Web Graphs 374
18.3 Small World 382
18.4 Exercises 384
18.5 Notes 386
19 Weighted Graphs 388
19.1 Minimum Spanning Tree 388
19.2 Shortest Paths 391
19.3 Minimum Weight Assignment 395
19.4 Exercises 398
19.5 Notes 400
20 Brief notes on uncovered topics 403

PART FOUR TOOLS AND METHODS 415

21 Moments 417
21.1 First and Second Moment Method 417
21.2 Convergence of Moments 420
21.3 Stein–Chen Method 424
22 Inequalities 427
22.1 Binomial Coefficient Approximation 427
22.2 Balls in Boxes 428
22.3 FKG Inequality 429
22.4 Sums of Independent Bounded Random Variables 431
 Contents v

22.5 Sampling Without Replacement 436

22.6 Janson’s Inequality 438
22.7 Martingales. Azuma-Hoeffding Bounds 440
22.8 Talagrand’s Inequality 446
22.9 Dominance 450
23 Differential Equations Method 452
24 Branching Processes 457
25 Entropy 458
25.1 Basic Notions 458
25.2 Shearer’s Lemma 460
References 462
26 Indices 498
Author Index 499
Main Index 505
 Preface

Our purpose in writing this book is to provide a gentle introduction to a subject

that is enjoying a surge in interest. We believe that the subject is fascinating
in its own right, but the increase in interest can be attributed to several factors.
One factor is the realization that networks are “everywhere”. From social net-
works such as Facebook, the World Wide Web and the Internet to the complex
interactions between proteins in the cells of our bodies, we face the challenge
of understanding their structure and development. By and large natural net-
works grow in an unpredictable manner and this is often modeled by a random
construction. Another factor is the realization by Computer Scientists that NP-
hard problems are often easier to solve than their worst-case suggests and that
an analysis of running times on random instances can be informative.

History
Random graphs were used by Erdős [278] to give a probabilistic construction
of a graph with large girth and large chromatic number. It was only later that
Erdős and Rényi began a systematic study of random graphs as objects of
interest in their own right. Early on they defined the random graph Gn,m and
founded the subject. Often neglected in this story is the contribution of Gilbert
[374] who introduced the model Gn,p , but clearly the credit for getting the
subject off the ground goes to Erdős and Rényi. Their seminal series of papers
[279], [281], [282], [283] and in particular [280], on the evolution of random
graphs laid the groundwork for other mathematicians to become involved in
studying properties of random graphs.
In the early eighties the subject was beginning to blossom and it received a
boost from two sources. First was the publication of the landmark book of Béla
Bollobás [132] on random graphs. Around the same time, the Discrete Math-
ematics group in Adam Mickiewicz University began a series of conferences
in 1983. This series continues biennially to this day and is now a conference
attracting more and more participants.
The next important event in the subject was the start of the journal Random
Structures and Algorithms in 1990 followed by Combinatorics, Probability and

vi
 Preface vii

Computing a few years later. These journals provided a dedicated outlet for
work in the area and are flourishing today.

Scope of the book

We have divided the book into four parts. Part one is devoted to giving a de-
tailed description of the main properties of Gn,m and Gn,p . The aim is not
to give best possible results, but instead to give some idea of the tools and
techniques used in the subject, as well to display some of the basic results
of the area. There is sufficient material in part one for a one semester course
at the advanced undergraduate or beginning graduate level. Once one has fin-
ished the content of the first part, one is equipped to continue with material of
the remainder of the book, as well as to tackle some of the advanced mono-
graphs such as Bollobás [132] and the more recent one by Janson, Łuczak and
Ruciński [441].
Each chapter comes with a few exercises. Some are fairly simple and these
are designed to give the reader practice with making some the estimations that
are so prevalent in the subject. In addition each chapter ends with some notes
that lead through references to some of the more advanced important results
that have not been covered.
Part two deals with models of random graphs that naturally extend Gn,m and
Gn,p . Part three deals with other models. Finally, in part four, we describe some
of the main tools used in the area along with proofs of their validity.
Having read this book, the reader should be in a good position to pursue
research in the area and we hope that this book will appeal to anyone interested
in Combinatorics or Applied Probability or Theoretical Computer Science.

Acknowledgement
Several people have helped with the writing of this book and we would like to
acknowledge their help. First there are the students who have sat in on courses
based on early versions of this book and who helped to iron out the many typo’s
etc.
We would next like to thank the following people for reading parts of the
book before final submission: Andrew Beveridge, Deepak Bal, Malgosia Bed-
narska, Patrick Bennett, Mindaugas Blozneliz, Antony Bonato, Boris Bukh,
Fan Chung, Amin Coja-Oghlan, Colin Cooper, Andrzej Dudek, Asaf Ferber,
viii Preface

Nikolas Fountoulakis, Catherine Greenhill, Dan Hefetz, Paul Horn, Hsien–

Kuei Hwang, Jerzy Jaworski, Tony Johansson, Mihyun Kang, Michael Kriv-
elevich, Tomasz Łuczak, Colin McDiarmid, Andrew McDowell, Hosam Mah-
moud, Mike Molloy, Tobias Müller, Rajko Nenadov, Wesley Pegden, Boris
Pittel, Dan Poole, Pawel Prałat, Oliver Riordan, Andrzej Ruciński, Katarzyna
Rybarczyk, Wojtek Samotij, Yilun Shang, Matas Šilekis, Greg Sorkin, Joel
Spencer, Dudley Stark, Angelika Steger, Prasad Tetali, Andrew Thomason,
Linnus Wästlund, Nick Wormald, Stephen Young.
Thanks also to Béla Bollobás for his advice on the structure of the book.

Conventions/Notation
Often in what follows, we will give an expression for a large positive integer. It
might not be obvious that the expression is actually an integer. In which case,
the reader can rest assured that he/she can round up or down and obtained any
required property. We avoid this rounding for convenience and for notational
purposes.
In addition we list the following notation:
Mathematical relations

• f (x) = O(g(x)): | f (x)| ≤ K|g(x)| for some constant K > 0 and all x ∈ R.
• f (x) = Θ(g(x)): f (n) = O(g(x)) and g(x) = O( f (x)).
• f (x) = o(g(x)) as x → a: f (x)/g(x) → 0 as x → a.
• A  B: A/B → 0 as n → ∞.
• A  B: A/B → ∞ as n → ∞.
• A ≈ B: A/B → 1 as some parameter converges to 0 or ∞ or another limit.
• A . B or B & A if A ≤ (1 + o(1))B.
• [n]: This is {1, 2, . . . , n}. In general, if a < b are positive integers, then
[a, b] = {a, a + 1, . . . , b}.
• If S is a set and k is a non-negative integer then Sk denotes the set of k-

element subsets of S. In particular, [n]

k dnotes the set of k-sets of {1, 2, . . . , n}.
S
= kj=0 Sj .

S

Furthermore, ≤k
Graph Notation
• G = (V, E): V = V (G) is the vertex set and E = E(G) is the edge set.
• e(G) = |E(G)| and for S ⊆ V we have eG (S) = | {e ∈ E : e ⊆ S} |.
• N(S) = NG (S) = {w ∈/ S : ∃v ∈ S such that {v, w} ∈ E} and dG (S) = |NG (S)|
for S ⊆ V (G).
 Preface ix

• NG (S, X) = NG (S) ∩ X for X, S ⊆ V .

• dS (x) = | {y ∈ S : {x, y} ∈ E} | for x ∈ V, S ⊆ V .
• For sets X,Y ⊆ V (G) we let NG (X,Y ) = {y ∈ Y : ∃x ∈ X, {x, y} ∈ E(G)} and
eG (X,Y ) = |NG (X,Y )|.
• For K ⊆ V (G) and v ∈ V (G), we let dK (v) dnote the number of neighbors of
v in K. The graph G is hopefully clear in the context in which this is used.
• For a graph H, aut(H) denotes the number of automorphisms of H.

Random Graph Models

• [n]: The set {1, 2, . . . , n}.

• Gn,m : The family of all labeled graphs with vertex set V = [n] = {1, 2, . . . , n}
and exactly m edges.
• Gn,m :A random graph chosen uniformly at random from Gn,m .
• En,m = E(Gn,m ).
• Gn,p : A random graph on vertex set [n] where each possible edge occurs
independently with probability p.
• En,p = E(Gn,p ).
≥k : G
• Gδn,m n,m , conditioned on having minimum degree at least k.
• Gn,n,p : A random bipartite graph with vertex set consisting of two disjoint
copies of [n] where each of the n2 possible edges occurs independently with
probability p.
• Gn,r : A random r-regular graph on vertex set [n].
• Gn,d : The set of graphs with vertex set [n] and degree sequence
d = (d1 , d2 , . . . , dn ).
• Gn,d : A random graph chosen uniformly at random from Gn,d .
• Hn,m;k : A random k-uniform hypergraph on vertex set [n] and m edges of
size k.
• Hn,p;k : A random k-uniform hypergraph on vertex set [n] where each of the
n

k possibles edge occurs independently with probability p.
~
• Gk−out : A random digraph on vertex set [n] where each v ∈ [n] independently
chooses k random out-neighbors.
• Gk−out : The graph obtained from ~Gk−out by ignoring orientation and coa-
lescing multiple edges.

Probability

• P(A): The probability of event A.

• E Z: The expected value of random variable Z.
• h(Z): The entropy of random variable Z.
• Po(λ ): A random variable with the Poisson distribution with mean λ .
x Preface

• N(0, 1): A random variable with the normal distribution, mean 0 and vari-
ance 1.
• Bin(n, p): A random variable with the binomial distribution with parameters
n, the number of trials and p, the probability of success.
• EXP(λ ): A random variable with the exponential distribution, mean λ i.e.
P(EXP(λ ) ≥ x) = e−λ x . We sometimes say rate 1/λ in place of mean λ .
• w.h.p.: A sequence of events An , n = 1, 2, . . . , is said to occur with high
probability (w.h.p.) if limn→∞ P(An ) = 1.
d d
• →: We write Xn → X to say that a random variable Xn converges in distribu-
d
tion to a random variable X, as n → ∞. Occasionally we write Xn → N(0, 1)
d
(resp. Xn → Po(λ )) to mean that X has the corresponding normal (resp. Pois-
son) distribution.
P ART ONE

BASIC MODELS
 1
Random Graphs

Graph theory is a vast subject in which the goals are to relate various graph
properties i.e. proving that Property A implies Property B for various proper-
ties A,B. In some sense, the goals of Random Graph theory are to prove results
of the form “Property A almost always implies Property B”. In many cases
Property A could simply be “Graph G has m edges”. A more interesting exam-
ple would be the following: Property A is “G is an r-regular graph, r ≥ 3” and
Property B is “G is r-connected”. This is proved in Chapter 11.
Before studying questions such as these, we will need to describe the basic
models of a random graph.

1.1 Models and Relationships

The study of random graphs in their own right began in earnest with the sem-
inal paper of Erdős and Rényi [280]. This paper was the first to exhibit the
threshold phenomena that characterize the subject.
Let Gn,m be the family of all labeled graphs
with vertex set V = [n] =
{1, 2, . . . , n} and exactly m edges, 0 ≤ m ≤ n2 . To every graph G ∈ Gn,m , we
assign a probability
 n
−1
P(G) = 2 .
m
Equivalently, we start with an empty graph on the set [n], and insert m edges
n

in such a way that all possible (m2) choices are equally likely. We denote such
a random graph by Gn,m = ([n], En,m ) and call it a uniform random graph.

We now describe a similar model. Fix 0 ≤ p ≤ 1. Then for 0 ≤ m ≤ n2 ,
assign to each graph G with vertex set [n] and m edges a probability
n
P(G) = pm (1 − p)(2)−m ,

where 0 ≤ p ≤ 1.
Equivalently, we start with an empty graph with vertex set

[n] and perform n2 Bernoulli experiments inserting edges independently with
probability p. We call such a random graph, a binomial random graph and
denote it by Gn,p = ([n], En,p ). This was introduced by Gilbert [374]

3
4 Random Graphs

As one may expect there is a close relationship between these two models
of random graphs. We start with a simple observation.

Lemma 1.1 A random graph Gn,p , given that its number of edges is m, is
n

equally likely to be one of the (2) graphs that have m edges.
m

Proof Let G0 be any labeled graph with m edges. Then since

{Gn,p = G0 } ⊆ {|En,p | = m}

we have
P(Gn,p = G0 , |En,p | = m)
P(Gn,p = G0 | |En,p | = m) =
P(|En,p | = m)
P(Gn,p = G0 )
=
P(|En,p | = m)
n
pm (1 − p)(2)−m
= n

(2) pm (1 − p)(n2)−m
m
 n
−1
= 2 .
m

Thus Gn,p conditioned on the event {Gn,p has m edges} is equal in distri-
bution to Gn,m , the graph chosen uniformly at random from all graphs with m
edges.
Obviously, the main difference between those two models of random graphs
is that in Gn,m we choose its number of edges, while in the case of Gn,p the
n

number of edges is the Binomial random variable with the parameters 2 and
p. Intuitively, for large n random graphs Gn,m and Gn,p should behave in a
similar fashion when the number of edges m in Gn,m equals or is “close” to the
expected number of edges of Gn,p , i.e., when

n2 p
 
n
m= p≈ , (1.1)
2 2
or, equivalently, when the edge probability in Gn,p
2m
p≈ . (1.2)
n2
Throughout the book, we will use the notation f ≈ g to indicate that f = (1 +
o(1))g, where the o(1) term will depend on some parameter going to 0 or ∞.
We next introduce a useful “coupling technique” that generates the random
 1.1 Models and Relationships 5

graph Gn,p in two independent steps. We will then describe a similar idea in
relation to Gn,m . Suppose that p1 < p and p2 is defined by the equation
1 − p = (1 − p1 )(1 − p2 ), (1.3)
or, equivalently,
p = p1 + p2 − p1 p2 .
Thus an edge is not included in Gn,p if it is not included in either of Gn,p1 or
Gn,p2 .
It follows that
Gn,p = Gn,p1 ∪ Gn,p2 ,
where the two graphs Gn,p1 , Gn,p2 are independent. So when we write
Gn,p1 ⊆ Gn,p ,
we mean that the two graphs are coupled so that Gn,p is obtained from Gn,p1 by
superimposing it with Gn,p2 and replacing eventual double edges by a single
one.
We can also couple random graphs Gn,m1 and Gn,m2 where m2 ≥ m1 via
Gn,m2 = Gn,m1 ∪ H.
Here H is the random graph on vertex
set [n] that has m = m2 − m1 edges
chosen uniformly at random from [n] 2 \ En,m1 .
Consider now a graph property P defined as a subset of the set of all labeled
n
graphs on vertex set [n], i.e., P ⊆ 2(2) . For example, all connected graphs (on
n vertices), graphs with a Hamiltonian cycle, graphs containing a given sub-
graph, planar graphs, and graphs with a vertex of given degree form a specific
“graph property”.
We will state below two simple observations which show a general relation-
ship between Gn,m and Gn,p in the context of the probabilities of having a given
graph property P. The constant 10 in the next lemma is not best possible, but
in the context of the usage of the lemma, any constant will suffice.
Lemma 1.2 Let P be any graph property and p = m/ n

2 where m = m(n)→ ∞,
n

2 − m → ∞. Then, for large n,

P(Gn,m ∈ P) ≤ 10m1/2 P(Gn,p ∈ P).

Proof By the law of total probability,
(n2)
P(Gn,p ∈ P) = ∑ P(Gn,p ∈ P | |En,p | = k) P(|En,p | = k)
k=0
6 Random Graphs
(n2)
= ∑ P(Gn,k ∈ P) P(|En,p | = k) (1.4)
k=0
≥ P(Gn,m ∈ P) P(|En,p | = m).
To justify (1.4), we write
P(Gn,p ∈ P ∧ |En,p | = k)
P(Gn,p ∈ P | |En,p | = k) =
P(|En,p | = k)
pk (1 − p)N−k
= ∑ N
k N−k
G∈P k p (1 − p)
|E(G)|=k
1
= ∑ N

G∈P k
|E(G)|=k
= P(Gn,k ∈ P).
Next recall that the number of edges |En,p | of a random graph

Gn,p is a random
variable with the Binomial distribution with parameters n2 and p. Applying
Stirling’s Formula:
 k
k √
k! = (1 + o(1)) 2πk, (1.5)
e
and putting N = n2 , we get


 
N m n
P(|En,p | = m) = p (1 − p)(2)−m
m
√
N N 2πN pm (1 − p)N−m
= (1 + o(1)) p (1.6)
mm (N − m)N−m 2π m(N − m)
s
N
= (1 + o(1)) ,
2πm(N − m)
Hence
1
P(|En,p | = m) ≥ √ ,
10 m
so
P(Gn,m ∈ P) ≤ 10m1/2 P(Gn,p ∈ P).

We call a graph property P monotone increasing if G ∈ P implies G + e ∈

P, i.e., adding an edge e to a graph G does not destroy the property. For ex-
ample, connectivity and Hamiltonicity are monotone increasing properties. A
 1.1 Models and Relationships 7

monotone increasing property is non-trivial if the empty graph K̄n ∈ / P and

the complete graph Kn ∈ P.
A graph property is monotone decreasing if G ∈ P implies G − e ∈ P, i.e.,
removing an edge from a graph does not destroy the property. Properties of a
graph not being connected or being planar are examples of monotone decreas-
ing graph properties. Obviously, a graph property P is monotone increasing if
and only if its complement is monotone decreasing. Clearly not all graph prop-
erties are monotone. For example having at least half of the vertices having a
given fixed degree d is not monotone.
From the coupling argument it follows that if P is a monotone increasing
property then, whenever p < p0 or m < m0 ,
P(Gn,p ∈ P) ≤ P(Gn,p0 ∈ P), (1.7)
and
P(Gn,m ∈ P) ≤ P(Gn,m0 ∈ P), (1.8)
respectively.
For monotone increasing graph properties we can get a much better upper
bound on P(Gn,m ∈ P), in terms of P(Gn,p ∈ P), than that given by Lemma
1.2.

Lemma 1.3 Let P be a monotone increasing graph property and p = m

N.
Then, for large n and p = o(1) such that N p, N(1 − p)/(N p)1/2 → ∞,
P(Gn,m ∈ P) ≤ 3 P(Gn,p ∈ P).
Proof Suppose P is monotone increasing and p = Nm , where N = n

2 . Then
N
P(Gn,p ∈ P) = ∑ P(Gn,k ∈ P) P(|En,p | = k)
k=0
N
≥ ∑ P(Gn,k ∈ P) P(|En,p | = k)
k=m

However, by the coupling property we know that for k ≥ m,

P(Gn,k ∈ P) ≥ P(Gn,m ∈ P).
The number of edges |En,p | in Gn,p has the Binomial distribution with param-
eters N, p. Hence
N
P(Gn,p ∈ P) ≥ P(Gn,m ∈ P) ∑ P(|En,p | = k)
k=m
8 Random Graphs
N
= P(Gn,m ∈ P) ∑ uk , (1.9)
k=m

where  
N k
uk = p (1 − p)N−k .
k
Now, using Stirling’s formula,
N N pm (1 − p)N−m 1 + o(1)
um = (1 + o(1)) = .
mm (N − m)N−m (2πm)1/2 (2πm)1/2
Furthermore, if k = m + t where 0 ≤ t ≤ m1/2 then
t
1 − N−m
 
uk+1 (N − k)p t t +1
= = ≥ exp − − ,
uk (k + 1)(1 − p) 1 + t+1
m
N −m−t m
after using Lemma 22.1(a),(b) to obtain the inequality. and our assumptions on
N, p to obtain the second.
It follows that for 0 ≤ t ≤ m1/2 ,
( )
t−1 
1 + o(1) s s+1
um+t ≥ exp − ∑ − ≥
(2πm)1/2 s=0 N − m − s m
n 2 o
t
exp − 2m − o(1)
,
(2πm)1/2
where we have used the fact that m = o(N).
It follows that
m+m1/2 Z 1
1 − o(1) 2 /2 1
∑ uk ≥ e−x dx ≥
k=m (2π)1/2 x=0 3
and the lemma follows from (1.9).
Lemmas 1.2 and 1.3 are surprisingly applicable. In fact, since the Gn,p
model is computationally easier to handle than Gn,m , we will repeatedly use
both lemmas to show that P(Gn,p ∈ P) → 0 implies that P(Gn,m ∈ P) → 0
when n → ∞. In other situations we can use a stronger and more widely appli-
cable result. The theorem below, which we state without proof, gives precise
conditions for the asymptotic equivalence of random graphs Gn,p and Gn,m . It
is due to Łuczak [547].

p
Theorem 1.4 Let 0 ≤ p0 ≤ 1, s(n) = n p(1 − p) → ∞, and ω(n) → ∞ ar-
bitrarily slowly as n → ∞.
 1.2 Thresholds and Sharp Thresholds 9

(i) Suppose that P is a graph property such that P(Gn,m ∈ P) → p0 for all
    
n n
m∈ p − ω(n)s(n), p + ω(n)s(n) .
2 2
Then P(Gn,p ∈ P) → p0 as n → ∞,
(ii) Let p− = p − ω(n)s(n)/n3 and p+ = p + ω(n)s(n)/n3 Suppose that P is
a monotone graph property such that P(Gn,p− ∈ P) → p0 and P(G

n,p+ ∈
P) → p0 . Then P(Gn,m ∈ P) → p0 , as n → ∞, where m = b n2 pc.

1.2 Thresholds and Sharp Thresholds

One of the most striking observations regarding the asymptotic properties of
random graphs is the “abrupt” nature of the appearance and disappearance of
certain graph properties. To be more precise in the description of this phe-
nomenon, let us introduce threshold functions (or just thresholds) for mono-
tone graph properties. We start by giving the formal definition of a threshold
for a monotone increasing graph property P.

Definition 1.5 A function m∗ = m∗ (n) is a threshold for a monotone increas-

ing property P in the random graph Gn,m if
(
0 if m/m∗ → 0,
lim P(Gn,m ∈ P) =
n→∞ 1 if m/m∗ → ∞,
as n → ∞.

A similar definition applies to the edge probability p = p(n) in a random

graph Gn,p .

Definition 1.6 A function p∗ = p∗ (n) is a threshold for a monotone increas-

ing property P in the random graph Gn,p if
(
0 if p/p∗ → 0,
lim P(Gn,p ∈ P) =
n→∞ 1 if p/p∗ → ∞,
as n → ∞.

It is easy to see how to define thresholds for monotone decreasing graph

properties and therefore we will leave this to the reader.
10 Random Graphs

Notice also that the thresholds defined above are not unique since any func-
tion which differs from m∗ (n) (resp. p∗ (n)) by a constant factor is also a thresh-
old for P.
A large body of the theory of random graphs is concerned with the search for
thresholds for various properties, such as containing a path or cycle of a given
length, or, in general, a copy of a given graph, or being connected or Hamilto-
nian, to name just a few. Therefore the next result is of special importance. It
was proved by Bollobás and Thomason [152].

Theorem 1.7 Every non-trivial monotone graph property has a threshold.

Proof Without loss of generality assume that P is a monotone increasing

graph property. Given 0 < ε < 1 we define p(ε) by

P(Gn,p(ε) ∈ P) = ε.

Note that p(ε) exists because

P(Gn,p ∈ P) = ∑ p|E(G)| (1 − p)N−|E(G|

G∈P

is a polynomial in p that increases from 0 to 1. This is not obvious from the

expression, but it is obvious from the fact that P is monotone increasing and
that increasing p increases the likelihood that Gn,p ∈ P.
We will show that p∗ = p(1/2) is a threshold for P. Let G1 , G2 , . . . , Gk
be independent copies of Gn,p . The graph G1 ∪ G2 ∪ . . . ∪ Gk is distributed as
Gn,1−(1−p)k . Now 1 − (1 − p)k ≤ kp, and therefore by the coupling argument

Gn,1−(1−p)k ⊆ Gn,kp ,

/ P implies G1 , G2 , . . . , Gk ∈
and so Gn,kp ∈ / P. Hence

/ P) ≤ [P(Gn,p ∈
P(Gn,kp ∈ / P)]k .

Let ω be a function of n such that ω → ∞ arbitrarily slowly as n → ∞, ω 

log log n. (We say that f (n)  g(n) or f (n) = o(g(n)) if f (n)/g(n) → 0 as
n → ∞. Of course in this case we can also write g(n)  f (n).) Suppose also
that p = p∗ = p(1/2) and k = ω. Then

/ P) ≤ 2−ω = o(1).
P(Gn,ω p∗ ∈

On the other hand for p = p∗ /ω,

1
/ P) ≤ P(Gn,p∗ /ω ∈
/ P) .
 ω
= P(Gn,p∗ ∈
2
 1.2 Thresholds and Sharp Thresholds 11

So
/ P) ≥ 2−1/ω = 1 − o(1).
P(Gn,p∗ /ω ∈

In order to shorten many statements of theorems in the book we say that a

sequence of events En occurs with high probability (w.h.p.) if

lim P(En ) = 1.
n→∞

Thus the statement that says p∗ is a threshold for a property P in Gn,p is the
same as saying that Gn,p 6∈ P w.h.p. if p  p∗ , while Gn,p ∈ P w.h.p. if
p  p∗ .
In many situations we can observe that for some monotone graph properties
more “subtle” thresholds hold. We call them “sharp thresholds”. More pre-
cisely,

Definition 1.8 A function m∗ = m∗ (n) is a sharp threshold for a monotone

increasing property P in the random graph Gn,m if for every ε > 0,

0 i f m/m∗ ≤ 1 − ε

lim P(Gn,m ∈ P) =
n→∞ 1 i f m/m∗ ≥ 1 + ε.

A similar definition applies to the edge probability p = p(n) in the random

graph Gn,p .

Definition 1.9 A function p∗ = p∗ (n) is a sharp threshold for a monotone

increasing property P in the random graph Gn,p if for every ε > 0

0 i f p/p∗ ≤ 1 − ε

lim P(Gn,p ∈ P) =
n→∞ 1 i f p/p∗ ≥ 1 + ε.

We will illustrate both types of threshold in a series of examples dealing with

very simple graph properties. Our goal at the moment is to demonstrate basic
techniques to determine thresholds rather than to “discover” some “striking”
facts about random graphs.
We will start with the random graph Gn,p and the property

P = {all non-empty (non-edgeless) labeled graphs on n vertices}.

This simple graph property is clearly monotone increasing and we will show
below that p∗ = 1/n2 is a threshold for a random graph Gn,p of having at least
one edge (being non-empty).
12 Random Graphs

Lemma 1.10 Let P be the property defined above, i.e., stating that Gn,p
contains at least one edge. Then
(
0 if p  n−2
lim P(Gn,p ∈ P) =
n→∞ 1 if p  n−2 .

Proof Let X be a random variable
counting edges in
Gn,p . Since X has the
Binomial distribution, then E X = n2 p, and Var X = n2 p(1− p) = (1− p) E X.
A standard way to show the first part of the threshold statement, i.e. that
w.h.p. a random graph Gn,p is empty when p = o(n−2 ), is a very simple conse-
quence of Markov’s inequality, called the First Moment Method, see Lemma
21.2. It states that if X is a non-negative integer valued random variable, then

P(X > 0) ≤ EX.

Hence, in our case

n2
P(X > 0) ≤ p→0
2
as n → ∞, since p  n−2 .
On the other hand, if we want to show that P(X > 0) → 1 as n → ∞ then
we cannot use the First Moment Method and we should use the Second Mo-
ment Method, which is a simple consequence of the Chebyshev inequality,
see Lemma 21.3. We will use the inequality to show concentration around the
mean. By this we mean that w.h.p. X ≈ E X. The Chebyshev inequality states
that if X is a non-negative integer valued random variable then
Var X
P(X > 0) ≥ 1 − .
(E X)2

Hence P(X > 0) → 1 as n → ∞ whenever Var X/(E X)2 → 0 as n → ∞. (For

proofs of both of the above Lemmas see Section 21.1 of Chapter 21.)
Now, if p  n−2 then E X → ∞ and therefore
Var X 1− p
= →0
(E X)2 EX
as n → ∞, which shows that the second statement of Lemma 1.10 holds, and
so p∗ = 1/n2 is a threshold for the property of Gn,p being non-empty.

Let us now look at the degree of a fixed vertex in both models of random
graphs. One immediately notices that if deg(v) denotes the degree of a fixed
vertex in Gn,p , then deg(v) is a binomially distributed random variable, with
 1.2 Thresholds and Sharp Thresholds 13

parameters n − 1 and p, i.e., for d = 0, 1, 2 . . . , n − 1,

 
n−1 d
P(deg(v) = d) = p (1 − p)n−1−d ,
d
while in Gn,m the distribution of deg(v) is Hypergeometric, i.e.,
n−1
n−1
( 2 )

d m−d
P(deg(v) = d) = .
(n2)

m

Consider the monotone decreasing graph property that a graph contains an

isolated vertex, i.e. a vertex of degree zero:

P = {all labeled graphs on n vertices containing isolated vertices}.

We will show that m∗ = 21 n log n is the sharp threshold function for the above
property P in Gn,m .
Lemma 1.11 Let P be the property that a graph on n vertices contains at
least one isolated vertex and let m = 12 n(log n + ω(n)). Then
(
1 if ω(n) → −∞
lim P(Gn,m ∈ P) =
n→∞ 0 if ω(n) → ∞.
Proof To see that the second statement of Lemma 1.11 holds we use the First
Moment Method. Namely, let X0 = Xn,0 be the number of isolated vertices in
the random graph Gn,m . Then X0 can be represented as the sum of indicator
random variables
X0 = ∑ Iv ,
v∈V

where
(
1 if v is an isolated vertex in Gn,m
Iv =
0 otherwise.
So
(n−1
2 )


m
E X0 = ∑ E Iv = n =
v∈V
(n)
2
m
 m m−1  
n−2 4i
n ∏ 1− =
n i=0 n(n − 1)(n − 2) − 2i(n − 2)
n−2 m (log n)2
    
n 1+O , (1.10)
n n
14 Random Graphs

assuming that ω = o(log n).

Hence,
 m
n−2 2m
E X0 ≤ n ≤ ne− n = e−ω ,
n

for m = 12 n(log n + ω(n)).

(1 + x ≤ ex is one of the basic inequalities stated in Lemma 22.1.)
So E X0 → 0 when ω(n) → ∞ as n → ∞ and the First Moment Method im-
plies that X0 = 0 w.h.p.
To show that Lemma 1.11 holds in the case when ω → −∞ we first observe
from (1.10) that in this case
n−2 m
 
E X0 = (1 − o(1))n
n
 
2m
≥ (1 − o(1))n exp −
n−2
≥ (1 − o(1))e−ω → ∞, (1.11)

The second inequality in the above comes from Lemma 22.1(b), and we have
once again assumed that ω = o(log n) to justify the first equation.
We caution the reader that E X0 → ∞ does not prove that X0 > 0 w.h.p. In
Chapter 5 we will see an example of a random variable XH , where E XH → ∞
and yet XH = 0 w.h.p.
We will now use a stronger version of the Second Moment Method (for
its proof see Section 21.1 of Chapter 21). It states that if X is a non-negative
integer valued random variable then

(E X)2 Var X
P(X > 0) ≥ = 1− . (1.12)
EX2 EX2
Notice that

!2
E X02 = E ∑ Iv = ∑ E(Iu Iv )
v∈V u,v∈V

= ∑ P(Iu = 1, Iv = 1)
u,v∈V

= ∑ P(Iu = 1, Iv = 1) + ∑ P(Iu = 1, Iv = 1)
u6=v u=v
n−2
( )
2
= n(n − 1) mn
+ E X0
() 2
m
 1.2 Thresholds and Sharp Thresholds 15

n − 2 2m
 
≤ n2 + E X0
n
= (1 + o(1))(E X0 )2 + E X0 .

The last equation follows from (1.10).

Hence, by (1.12),
(E X0 )2
P(X0 > 0) ≥
E X02
(E X0 )2
=
1 + o(1))((E X0 )2 + E X0 )
1
=
(1 + o(1)) + (E X0 )−1
= 1 − o(1),

on using (1.11). Hence P(X0 > 0) → 1 when ω(n) → −∞ as n → ∞, and so we

can conclude that m = m(n) is the sharp threshold for the property that Gn,m
contains isolated vertices.
For this simple random variable, we worked with Gn,m . We will in general
work with the more congenial independent model Gn,p and translate the results
to Gn,m if so desired.
For another simple example of the use of the second moment method, we
will prove

Theorem 1.12 If m/n → ∞ then w.h.p. Gn,m contains at least one triangle.

Proof Because having a triangle is a monotone increasing property we can

prove the result in Gn,p assuming that np → ∞.
Assume first that np = ω ≤ log n where ω = ω(n) → ∞ and let Z be the
number of triangles in Gn,p . Then

ω3
 
n 3
EZ = p ≥ (1 − o(1)) → ∞.
3 6
We remind the reader that simply having E Z → ∞ is not sufficient to prove that
Z > 0 w.h.p.
Next let T1 , T2 , . . . , TM , M = n3 denote the triangles of Kn . Then

M
E Z2 = ∑ P(Ti , T j ∈ Gn,p )
i, j=1
M M
= ∑ P(Ti ∈ Gn,p ) ∑ P(T j ∈ Gn,p | Ti ∈ Gn,p ) (1.13)
i=1 j=1
16 Random Graphs
M
= M P(T1 ∈ Gn,p ) ∑ P(T j ∈ Gn,p | T1 ∈ Gn,p ) (1.14)
j=1
M
= E Z × ∑ P(T j ∈ Gn,p | T1 ∈ Gn,p ).
j=1

Here (1.14) follows from (1.13) by symmetry.

Now suppose that T j , T1 share σ j edges. Then
M
∑ P(Tj ∈ Gn,p | T1 ∈ Gn,p )
j=1

= 1+ ∑ P(T j ∈ Gn,p | T1 ∈ Gn,p )+

j:σ j =1

∑ P(T j ∈ Gn,p | T1 ∈ Gn,p )

j:σ j =0
  
n
= 1 + 3(n − 3)p2 + − 3n + 8 p3
3
3ω 2
≤ 1+ + E Z.
n
It follows that
3ω 2
 
Var Z ≤ (E Z) 1 + + E Z − (E Z)2 ≤ 2 E Z.
n
Applying the Chebyshev inequality we get
Var Z 2
P(Z = 0) ≤ P(|Z − E Z| ≥ E Z) ≤ ≤ = o(1).
(E Z)2 EZ

This proves the theorem for p ≤ logn n . For larger p we can use (1.7).
We can in fact use the second moment method to show that if m/n → ∞ then
w.h.p. Gn,m contains a copy of a k-cycle Ck for any fixed k ≥ 3. See Theorem
5.3, see also Exercise 1.4.7.

1.3 Pseudo-Graphs
We sometimes use one of the two the following models that are related to Gn,m
and have a little more independence. (We will use Model A in Section 7.3 and
Model B in Section 6.4).
Model A: We let x = (x1 , x2 , . . . , x2m ) be chosen uniformly at random from
[n]2m .
 1.3 Pseudo-Graphs 17

Model B: We let x = (x1 , x2 , . . . , x2m ) be chosen uniformly at random from

[n]
m
2 .
(X)
The (multi-)graph Gn,m , X ∈ {A, B} has vertex set [n] and edge set Em =
{{x2i−1 , x2i } : 1 ≤ i ≤ m}. Basically, we are choosing edges with replacement.
In Model A we allow loops and in Model B we do not. We get simple graphs
(X∗)
from by removing loops and multiple edges to obtain graphs Gn,m with m∗
edges. It is not difficult to see that for X ∈ {A, B} and conditional on the value
(X∗)
of m∗ that Gn,m is distributed as Gn,m∗ , see Exercise (1.4.11).
More importantly, we have that for G1 , G2 ∈ Gn,m ,
(X) (X) (X) (X)
P(Gn,m = G1 | Gn,m is simple) = P(Gn,m = G2 | Gn,m is simple), (1.15)

for X = A, B.
This is because for i = 1, 2,
(A) m!2m (B) m!2m
P(Gn,m = Gi ) = and P(G n,m = G i ) = n
m m .
n2m 2 2

Indeed, we can permute the edges in m! ways and permute the vertices within
(X)
edges in 2m ways without changing the underlying graph. This relies on Gn,m
being simple.
Secondly, if m = cn for a constant c > 0 then with N = n2 , and using Lemma

22.2,

N m!2m
 
(X)
P(Gn,m is simple) ≥ ≥
m n2m
Nm m2 m3 m!2m
 
(1 − o(1)) exp − −
m! 2N 6N 2 n2m
2 +c)
= (1 − o(1))e−(c . (1.16)

It follows that if P is some graph property then

(X) (X)
P(Gn,m ∈ P) = P(Gn,m ∈ P | Gn,m is simple) ≤
2 +c (X)
(1 + o(1))ec P(Gn,m ∈ P). (1.17)

Here we have used the inequality P(A | B) ≤ P(A)/ P(B) for events A, B.
(X)
We will use this model a couple of times and (1.17) shows that if P(Gn,m ∈
P) = o(1) then P(Gn,m ∈ P) = o(1), for m = O(n).
(A)
Model Gn,m was introduced independently by Bollobás and Frieze [142] and
by Chvátal [189].
 18 Random Graphs

1.4 Exercises
We point out here that in the following exercises, we have not asked for best
possible results. These exercises are for practise. You will need to use the in-
equalities from Section 22.1.

1.4.1 Suppose that p = d/n where d = o(n1/3 ). Show that w.h.p. Gn,p has no
copies of K4 .
1.4.2 Suppose that p = d/n where d > 1. Show that w.h.p. Gn,p contains an
induced path of length (log n)1/2 .
1.4.3 Suppose that p = d/n where d = O(1). Prove that w.h.p., in Gn,p , for all
S ⊆ [n], |S| ≤ n/ log n, we have e(S) ≤ 2|S|, where e(S) is the number of
edges contained in S.
1.4.4 Suppose that p = log n/n. Let a vertex of Gn,p be small if its degree is
less than log n/100. Show that w.h.p. there is no edge of Gn,p joining
two small vertices.
1.4.5 Suppose that p = d/n where d is constant. Prove that w.h.p., in Gn,p , no
vertex belongs to more than one triangle.
1.4.6 Suppose that p = d/n where d is constant. Prove that w.h.p. Gn,p con-
tains a vertex of degree exactly (log n)1/2 .
1.4.7 Suppose that k ≥ 3 is constant and that np → ∞. Show that w.h.p. Gn,p
contains a copy of the k-cycle, Ck .
1.4.8 Suppose that 0 < p < 1 is constant. Show that w.h.p. Gn,p has diameter
two.
1.4.9 Let f : [n] → [n] be chosen uniformly at random from all nn functions
from [n] → [n]. Let X = { j :6 ∃i s.t. f (i) = j}. Show that w.h.p. |X| ≈
e−1 n.
1.4.10 Prove Theorem 1.4.
1.4.11 Show that conditional on the value of mX∗ that GX∗ n,m is distributed as
Gn,m∗ , where X = A, B.

1.5 Notes
Friedgut and Kalai [321] and Friedgut [322] and Bourgain [156] and Bourgain
and Kalai [155] provide much greater insight into the notion of sharp thresh-
olds. Friedgut [320] gives a survey of these aspects. For a graph property A
let µ(p, A ) be the probability that the random graph Gn,p has property A . A
threshold is coarse if it is not sharp. We can identify coarse thresholds with
)
p dµ(p,A
dp < C for some absolute constant 0 < C. The main insight into coarse
 1.5 Notes 19

thresholds is that to exist, the occurrence of A can in the main be attributed

to the existence of one of a bounded number of small subgraphs. For exam-
ple, Theorem 2.1 of [320] states that there exists a function K(C, ε) such that
the following holds. Let A be a monotone property of graphs that is invariant
)
under automorphism and assume that p dµ(p,A dp < C for some constant 0 < C.
Then for every ε > 0 there exists a finite list of graphs G1 , G2 , . . . , Gm all of
which have no more than K(ε,C) edges, such that if B is the family of graphs
having one of these graphs as a subgraph then µ(p, A ∆B) ≤ ε.
 2
Evolution

Here begins our story of the typical growth of a random graph. All the results
up to Section 2.3 were first proved in a landmark paper by Erdős and Rényi
[280]. The notion of the evolution of a random graph stems from a dynamic
view of a graph process: viz. a sequence of graphs:

G0 = ([n], 0),
/ G1 , G2 , . . . , Gm , . . . , GN = Kn .

where Gm+1
is obtained from Gm by adding a random edge em . We see that

there are n2 ! such sequences and Gm and Gn,m have the same distribution.
In process of the evolution of a random graph we consider properties possessed
by Gm or Gn,m w.h.p., when m = m(n) grows from 0 to n2 , while in the case

of Gn,p we analyse its typical structure when p = p(n) grows from 0 to 1 as

n → ∞.
In the current chapter we mainly explore how the typical component struc-
ture evolves as the number of edges m increases.

2.1 Sub-Critical Phase

The evolution of Erdős-Rényi type random graphs has clearly distinguishable
phases. The first phase, at the beginning of the evolution, can be described
as a period when a random graph is a collection of small components which
are mostly trees. Indeed the first result in this section shows that a random
graph Gn,m is w.h.p. a collection of tree-components as long as m = o(n), or,
equivalently, as long as p = o(n−1 ) in Gn,p . For clarity, all results presented
in this chapter are stated in terms of Gn,m . Due to the fact that computations
are much easier for Gn,p we will first prove results in this model and then the
results for Gn,m will follow by the equivalence established either in Lemmas
1.2 and 1.3 or in Theorem 1.4. We will also assume, throughout this chapter,
that ω = ω(n) is a function growing slowly with n, e.g. ω = log log n will
suffice.

Theorem 2.1 If m  n, then Gm is a forest w.h.p.

20
 2.1 Sub-Critical Phase 21

Proof Suppose m = n/ω and let N = n2 , so p = m/N ≤ 3/(ωn). Let X be

the number of cycles in Gn,p . Then

n  
n (k − 1)! k
EX = ∑ p
k=3 k 2
n
nk 3k
≤ ∑ k k
k=3 2k ω n
= O(ω −3 ) → 0.
Therefore, by the First Moment Method, (see Lemma 21.2),
P(Gn,p is not a forest) = P(X ≥ 1) ≤ E X = o(1),
which implies that
P(Gn,p is a forest) → 1 as n → ∞.
Notice that the property that a graph is a forest is monotone decreasing, so by
Lemma 1.3
P(Gm is a forest) → 1 as n → ∞.
(Note that we have actually used Lemma 1.3 to show that P(Gn,p is not a
forest)=o(1) implies that P(Gm is not a forest)=o(1).)
We will next examine the time during which the components of Gm are
isolated vertices and single edges only, w.h.p.

Theorem 2.2 If m  n1/2 then Gm is the union of isolated vertices and edges
w.h.p.

Proof Let p = m/N, m = n1/2 /ω and let X be the number of paths of length
two in the random graph Gn,p . By the First Moment Method,
n4
 
n 2
P(X > 0) ≤ E X = 3 p ≤ → 0,
3 2N 2 ω 2
as n → ∞. Hence
P(Gn,p contains a path of length two) = o(1).
Notice that the property that a graph contains a path of a given length two is
monotone increasing, so by Lemma 1.3,
P(Gm contains a path of length two) = o(1),
22 Evolution

and the theorem follows.

Now we are ready to describe the next step in the evolution of Gm .

Theorem 2.3 If m  n1/2 , then Gm contains a path of length two w.h.p.

m
Proof Let p = N ,m = ωn1/2 and X be the number of paths of length two in
Gn,p . Then
 
n 2
EX = 3 p ≈ 2ω 2 → ∞,
3
as n → ∞. This however does not imply that X > 0 w.h.p.! To show that X > 0
w.h.p. we will apply the Second Moment Method
Let P2 be the set of all paths of length two in the complete graph Kn , and let
X̂ be the number of isolated paths of length two in Gn,p i.e. paths that are also
components of Gn,p . We will show that w.h.p. Gn,p contains such an isolated
path. Now,
X̂ = ∑ IP⊆i Gn,p .
P∈P2

We always use IE to denote the indicator for an event E . The notation ⊆i in-
dicates that P is contained in Gn,p as a component (i.e. P is isolated). Having
a path of length two is a monotone increasing property. Therefore we can as-
sume that m = o(n) and so np = o(1) and the result for larger m will follow
from monotonicity and coupling. Then
 
n 2
E X̂ = 3 p (1 − p)3(n−3)+1
3
n3 4ω 2 n
≥ (1 − o(1)) (1 − 3np) → ∞,
2 n4
as n → ∞.
In order to compute the second moment of the random variable X̂ notice
that,
∗
X̂ 2 = ∑ ∑ IP⊆i Gn,p IQ⊆i Gn,p = ∑P,Q∈P IP⊆i Gn,p IQ⊆i Gn,p ,
2
P∈P2 Q∈P2

where the last sum is taken over P, Q ∈ P2 such that either P = Q or P and
Q are vertex disjoint. The simplification that provides the last summation is
precisely the reason that we introduce path-components (isolated paths). Now
( )
E X̂ 2 = ∑ ∑ P(Q ⊆i Gn,p | P ⊆i Gn,p ) P(P ⊆i Gn,p ).
P Q
 2.1 Sub-Critical Phase 23

The expression inside the brackets is the same for all P and so
 

E X̂ 2 = E X̂ 1 + ∑ P(Q ⊆i Gn,p | P(1,2,3) ⊆i Gn,p ) ,

Q∩P(1,2,3) =0/

where P{1,2,3} denotes the path on vertex set [3] = {1, 2, 3} with middle vertex
2. By conditioning on the event P(1,2,3) ⊆i Gn,p , i.e, assuming that P(1,2,3) is a
component of Gn,p , we see that all of the nine edges between Q and P(1,2,3)
must be missing. Therefore
   
n 2
2 3(n−6)+1
≤ E X̂ 1 + (1 − p)−9 E X̂ .


E X̂ ≤ E X̂ 1 + 3 p (1 − p)
3
So, by the Second Moment Method (see Lemma 21.5),

(E X̂)2 (E X̂)2
P(X̂ > 0) ≥ ≥

E X̂ 2 E X̂ 1 + (1 − p)−9 E X̂
1
= →1
(1 − p)−9 + [E X̂]−1

as n → ∞, since p → 0 and E X̂ → ∞. Thus

P(Gn,p contains an isolated path of length two) → 1,

which implies that P(Gn,p contains a path of length two) → 1. As the property
of having a path of length two is monotone increasing it in turn implies that

P(Gm contains a path of length two) → 1

for m  n1/2 and the theorem follows.

From Theorems 2.2 and 2.3 we obtain the following corollary.

Corollary 2.4 The function m∗ (n) = n1/2 is the threshold for the property
that a random graph Gm contains a path of length two, i.e.,
(
o(1) if m  n1/2 .
P(Gm contains a path of length two) =
1 − o(1) if m  n1/2 .

As we keep adding edges, trees on more than three vertices start to appear.
Note that isolated vertices, edges and paths of length two are also trees on one,
two and three vertices, respectively. The next two theorems show how long we
have to “wait” until trees with a given number of vertices appear w.h.p.
24 Evolution
k−2
Theorem 2.5 Fix k ≥ 3. If m  n k−1 , then w.h.p. Gm contains no tree with k
vertices.
k−2
2 3
Proof Let m = n k−1 /ω and then p = Nm ≈ ωnk/(k−1) ≤ ωnk/(k−1) . Let Xk denote
the number of trees with k vertices in Gn,p . Let T1 , T2 , . . . , TM be an enumeration
of the copies of k-vertex trees in Kn . Let
Ai = {Ti occurs as a subgraph in Gn,p }.
The probability that a tree T occurs in Gn,p is pe(T ) , where e(T ) is the number
of edges of T . So,
M
E Xk = ∑ P(At ) = M pk−1 .
t=1
n
k−2
since one can choose a set of k vertices in nk ways and then


But M = k k
by Cayley’s formula choose a tree on these vertices in kk−2 ways. Hence
 
n k−2 k−1
E Xk = k p . (2.1)
k
Noting also that (see Lemma 22.1(c))
   
n ne k
≤ ,
k k
we see that
 ne k  k−1
k−2 3
E Xk ≤ k
k ωnk/(k−1)
3k−1 ek
= → 0,
k2 ω k−1
as n → ∞, seeing as k is fixed.
Thus we see by the first moment method that,
P(Gn,p contains a tree with k vertices) → 0.
This property is monotone increasing and therefore
P(Gm contains a tree with k vertices) → 0.

Let us check what happens if the number of edges in Gm is much larger than
k−2
n .
k−1

k−2
Theorem 2.6 Fix k ≥ 3. If m  n k−1 , then w.h.p. Gm contains a copy of every
fixed tree with k vertices.
 2.1 Sub-Critical Phase 25
k−2
Proof Let p = Nm , m = ωn k−1 where ω = o(log n) and fix some tree T with k
vertices. Denote by X̂k the number of isolated copies of T (T -components) in
Gn,p . Let aut(H) denote the number of automorphisms of a graph H. Note that
there are k!/aut(T ) copies of T in the complete graph Kk . To see this choose
a copy of T with vertex set [k]. There are k! ways of mapping the vertices of
T to the vertices of Kk . Each map f induces a copy of T and two maps f1 , f2
induce the same copy iff f2 f1−1 is an automorphism of T .
So,
 
n k! k
E X̂k = pk−1 (1 − p)k(n−k)+(2)−k+1 (2.2)
k aut(T )
(2ω)k−1
= (1 + o(1)) → ∞.
aut(T )
In (2.2) we have used the fact that ω = o(log n) in order to show that (1 −
k
p)k(n−k)+(2)−k+1 = 1 − o(1).
Next let T be the set of copies of T in Kn and T[k] be a fixed copy of T on
vertices [k] of Kn . Then, arguing as in (2.3),

E(X̂k2 ) = ∑ P(T2 ⊆i Gn,p | T1 ⊆i Gn,p ) P(T1 ⊆i Gn,p )

T1 ,T2 ∈T
 
 
= E X̂k 
1 + ∑ P(T2 ⊆i Gn,p | T[k] ⊆i Gn,p )

T2 ∈T
V (T2 )∩[k]=0/
 2

≤ E X̂k 1 + (1 − p)−k E Xk .

2
Notice that the (1 − p)−k factor comes from conditioning on the event
T[k] ⊆i Gn,p which forces the non-existence of fewer than k2 edges.
Hence, by the Second Moment Method,

(E X̂k )2
P(X̂k > 0) ≥ 2

→ 1.
E X̂k 1 + (1 − p)−k E X̂k

Then, by a similar reasoning to that in the proof of Theorem 2.3,

P(Gm contains a copy of T ) → 1,

as n → ∞.

Combining the two above theorems we arrive at the following conclusion.

26 Evolution
k−2
Corollary 2.7 The function m∗ (n) = n k−1 is the threshold for the property
that a random graph Gm contains a tree with k ≥ 3 vertices, i.e.,
( k−2
o(1) if m  n k−1
P(Gm ⊇ k-vertex-tree) = k−2
1 − o(1) if m  n k−1

In the next theorem we show that “on the threshold” for k vertex trees, i.e.,
k−2
if m = cn k−1 , where c is a constant, c > 0, the number of tree components of a
given order asymptotically follows the Poisson distribution. This time we will
formulate both the result and its proof in terms of Gm .

k−2
Theorem 2.8 If m = cn k−1 , where c > 0, and T is a fixed tree with k ≥ 3
vertices, then

P(Gm contains an isolated copy of tree T ) → 1 − e−λ ,

k−1
as n → ∞, where λ = (2c)
aut(T ) .
More precisely, the number of copies of T is asymptotically distributed as
the Poisson distribution with expectation λ .

Proof Let T1 , T2 , . . . , TM be an enumeration of the copies of some k vertex

tree T in Kn .
Let
Ai = {Ti occurs as a component in Gm }.

Suppose J ⊆ [M] = {1, 2, . . . , M} with |J| = t, where t is fixed. Let AJ =

j∈J A j . We have P(AJ ) = 0 if there are i, j ∈ J such that Ti , T j share a ver-
T

tex. Suppose Ti , i ∈ J are vertex disjoint. Then

(n−kt
2 )


m−(k−1)t
P(AJ ) = N

.
m

Note that in the numerator we count the number of ways of choosing m edges
so that AJ occurs.
If, say, t ≤ log n, then
       
n − kt kt kt kt
= N 1− 1− = N 1−O ,
2 n n−1 n
and so
m2
n−kt

→ 0.
2
 2.1 Sub-Critical Phase 27

Then from Lemma 22.1(f),

m−(k−1)t
n−kt

N 1 − O ktn
 
2 = (1 + o(1))
m − (k − 1)t (m − (k − 1)t)!
m−(k−1)t 1 − O mkt



N n
= (1 + o(1))
(m − (k − 1)t)!
N m−(k−1)t
= (1 + o(1)) .
(m − (k − 1)t)!
Similarly, again by Lemma 22.1,
Nm
 
N
= (1 + o(1)) ,
m m!
and so
m!  m (k−1)t
P(AJ ) = (1 + o(1)) N −(k−1)t = (1 + o(1)) .
(m − (k − 1)t)! N
Thus, if ZT denotes the number of components of Gm that are copies of T ,
then,
    t  
ZT 1 n k! m (k−1)t
E ≈
t t! k, k, k, . . . , k aut(T ) N
!(k−1)t
t
nkt cn(k−2)/(k−1)
 
k!
≈
t!(k!)t aut(T ) N
λt
≈ ,
t!
where
(2c)k−1
λ= .
aut(T )
So by Corollary 21.11 the number of copies of T -components is asymptotically
distributed as the Poisson distribution with expectation λ given above, which
combined with the statements of Theorem 2.1 and Corollary 2.7 proves the the-
orem. Note that Theorem 2.1 implies that w.h.p. there are no non-component
copies of T .

We complete our presentation of the basic features of a random graph in

its sub-critical phase of evolution with a description of the order of its largest
component.
28 Evolution

Theorem 2.9 If m = 12 cn, where 0 < c < 1 is a constant, then w.h.p. the order
of the largest component of a random graph Gm is O(log n).

The above theorem follows from the next three lemmas stated and proved in
terms of Gn,p with p = c/n, 0 < c < 1. In fact the first of those three lemmas
covers a little bit more than the case of p = c/n, 0 < c < 1.

Lemma 2.10 If p ≤ n1 − n4/3 ω

, where ω = ω(n) → ∞, then w.h.p. every com-
ponent in Gn,p contains at most one cycle.
Proof Suppose that there is a pair of cycles that are in the same component.
If such a pair exists then there is minimal pair C1 ,C2 , i.e., either C1 and C2 are
connected by a path (or meet at a vertex) or they form a cycle with a diagonal
path (see Figure 2.1). Then in either case, C1 ∪ C2 consists of a path P plus
another two distinct edges, one from each endpoint of P joining it to another
vertex in P. The number of such graphs on k labeled vertices can be bounded
by k2 k!.

Let X be the number of subgraphs of the above kind (shown in Figure 2.1) in
the random graph Gn,p . By the first moment method (see Lemma 21.2),

n
 
n 2 k+1
P(X > 0) ≤ E X ≤ ∑ k k!p (2.3)
k=4 k
n
nk 1  ω k+1
≤ ∑ k2 k! k+1 1 − 1/3
k=4 k! n n
Z ∞ 2
x  ωx 
≤ exp − 1/3 dx
0 n n
2
= 3
ω
= o(1).

We remark for later use that if p = c/n, 0 < c < 1 then (2.3) implies
n
P(X > 0) ≤ ∑ k2 ck+1 n−1 = O(n−1 ). (2.4)
k=4

Hence, in determining the order of the largest component we may con-

centrate our attention on unicyclic components and tree-components (isolated
 2.1 Sub-Critical Phase 29
111
000 111
000
000
111 000
111
000
111 000
111
000
111 000
111
000
111 000
111 000
111
00
11 111
000
11
00 000
111
00
11 000
111
000
111
00
11

000
111 111
000 00
11 111
000 111
000 111
000
111
000 000
111 11
00 000
111 000
111 000
111
000
111 000
111 00
11 000
111 000
111 000
111
000
111 000
111
000
111 00
11 000
111
000
111 000
111
000
111 000
111
000
111
000
111
111
000
000
111
000
111
111
000 000
111
000
111 000
111
000
111
000
111
000
111 000
111 000
111
111
000 111
000
000
111 000
111
000
111 000
111
000
111 000
111
000
111
000
111 111
000
111
000 000
111
000
111
000
111 000
111
000
111
000
111

111
000
000
111
00
11 000
111
000
111
11
00 000
111
00
11
00
11

111
000
000
111
000
111
000
111
111
000
000
111 000
111
000
111
000
111
000
111 000
111
111
000
000
111
000
111
00
11 000
111 000
111
111
000
11
00 000
111
00
11 000
111
00
11 000
111
11
00
00
11
00
11
00
11
11
00
00
11
111
000
000
111 00
11
000
111 00
11
000
111
000
111
00
11
11
00
00
11
00
11 111
000
000
111
000
111
000
111
000
111
00
11
11
00
00
11
00
11

111
000
111
000 000
111
000
111
000
111
000
111 000
111
000
111 000
111
000
111 11
00
00
11
00
11
00
11

11
00
00
11
00
11
00
11

111
000
000
111
000
111
000
111
000
111 000
111 000
111
111
000
000
111 111
000
000
111 000
111
000
111 000
111
000
111

Figure 2.1 C1 ∪C2

trees). However the number of vertices on unicyclic components tends to be

rather small, as is shown in the next lemma.

Lemma 2.11 If p = c/n, where c 6= 1 is a constant, then in Gn,p w.h.p. the

number of vertices in components with exactly one cycle, is O(ω) for any grow-
ing function ω.

Proof Let Xk be the number of vertices on unicyclic components with k ver-

tices. Then
   
n k−2 k k
E Xk ≤ k kpk (1 − p)k(n−k)+(2)−k . (2.5)
k 2
30 Evolution

The factor kk−2 2k in (2.5) is the number of choices for a tree plus an edge on k

vertices in [k]. This bounds the number C(k, k) of connected graphs on [k] with
k edges. This is off by a factor O(k1/2 ) from the exact formula which is given
below for completeness:
k  
r
k (r − 1)! k−r−1 π k−1/2
C(k, k) = ∑ rk ≈ k . (2.6)
r=3 r 2 8

The remaining factor, other than nk , in (2.5) is the probability that the k edges

of the unicyclic component exist and that there are no other edges on Gn,p
incident with the k chosen vertices.
Noting also that by Lemma 22.1(d),
nk k(k−1)
 
n
≤ e− 2n ,
k k!
and so we get
nk − k(k−1) k+1 ck −ck+ ck(k−1) + ck
E Xk ≤ e 2n k e 2n 2n
k! nk
ek k(k−1) k(k−1) c
≤ k e− 2n kk+1 ck e−ck+ 2n + 2
k

k c
≤ k ce1−c e 2 .
So,
n n
k c
E ∑ Xk ≤ ∑k ce1−c e 2 = O(1), (2.7)
k=3 k=3

since ce1−c < 1 for c 6= 1. By Markov’s inequality, if ω = ω(n) → ∞, (see

Lemma 21.1)
!
n  
1
P ∑ Xk ≥ ω = O → 0 as n → ∞,
k=3 ω

and the Lemma follows.

After proving the first two lemmas one can easily see that the only remaining
candidate for the largest component of our random graph is an isolated tree.
Lemma 2.12 Let p = nc , where c 6= 1 is a constant, α = c − 1 − log c, and
ω = ω(n) → ∞, ω = o(log log n). Then

(i) w.h.p. there exists an isolated tree of order

 
1 5
k− = log n − log log n − ω,
α 2
 2.1 Sub-Critical Phase 31

(ii) w.h.p. there is no isolated tree of order at least

 
1 5
k+ = log n − log log n + ω
α 2
Proof Note that our assumption on c means that α is a positive constant.
Let Xk be the number of isolated trees of order k. Then
 
n k−2 k−1 k
E Xk = k p (1 − p)k(n−k)+(2)−k+1 . (2.8)
k
To prove (i) suppose k = O(log n). Then by using Lemma 22.1(a),(b) and Stir-
lings approximation (1.5) for k! we see that
n kk−2 −c k
E Xk = (1 + o(1)) (ce ) (2.9)
c k!
(1 + o(1)) n
= √ (ce1−c )k
c 2π k5/2
(1 + o(1)) n −αk
= √ e .
c 2π k5/2
Putting k = k− we see that

(1 + o(1)) n eαω (log n)5/2

E Xk = √ ≥ Aeαω , (2.10)
c 2π k5/2 n
for some constant A > 0.
We continue via the Second Moment Method, this time using the Chebyshev
inequality as we will need a little extra precision for the proof of Theorem
2.14. Using essentially the same argument as for a fixed tree T of order k (see
Theorem 2.6), we get
 2

E Xk2 ≤ E Xk 1 + (1 − p)−k E Xk .

So
 2

Var Xk ≤ E Xk + (E Xk )2 (1 − p)−k − 1
≤ E Xk + 2ck2 (E Xk )2 /n.

Thus, by the Chebyshev inequality (see Lemma 21.3), we see that for any fixed
ε > 0,
1 2ck2
P (|Xk − E Xk | ≥ ε E Xk ) ≤ + = o(1). (2.11)
ε 2 E Xk ε 2n
Thus w.h.p. Xk ≥ Aeαω/2 and this completes the proof of (i).
32 Evolution

For (ii) we go back to the formula (2.8) and write, for some new constant
A > 0,
k k−1  c k−1 −ck+ ck2
 
A  ne k k−2
E Xk ≤ √ k 1− e 2n
k k 2n n
2An  1−bck k
≤ cbk e ,
cbk k5/2
k


where cbk = c 1 − 2n .
In the case c < 1 we have cbk e1−bck ≤ ce1−c and cbk ≈ c and so we can write

k
n
3An n ce1−c 3An ∞
∑ E Xk ≤ c ∑ k5/2 ≤ 5/2 ∑ e−αk =
k=k+ k=k+ ck+ k=k+
3Ane−αk+ (3A + o(1))α 5/2 e−αω
= = = o(1). (2.12)
5/2
ck+ (1 − e−α ) c(1 − e−α )

If c > 1 then for k ≤ n

bk e1−bck = e−α−O(1/ log n)
log n we use c and for k > n
log n we
use ck ≥ c/2 and cbk e1−bck ≤ 1 and replace (2.12) by
n n/ log n n
3An 6An 1
∑ E Xk ≤ 5/2 ∑ e−(α+O(1/ log n))k + ∑ = o(1).
k=k+ ck+ k=k+ c k=n/ log n k5/2

Finally, applying Lemmas 2.11 and 2.12 we can prove the following useful
identity: Suppose that x = x(c) is given as
(
c c≤1
x = x(c) = −x −c
.
The solution in (0, 1) to xe = ce c>1

Note that xe−x increases continuously as x increases from 0 to 1 and then de-
creases. This justifies the existence and uniqueness of x.
Lemma 2.13 If c > 0, c 6= 1 is a constant, and x = x(c) is defined above,
then
1 ∞ kk−1
k
∑ ce−c = 1.
x k=1 k!
Proof Let p = nc . Assume first that c < 1 and let X be the total number of
vertices of Gn,p that lie in non-tree components. Let Xk be the number of tree-
components of order k. Then,
n
n= ∑ kXk + X.
k=1
 2.2 Super-Critical Phase 33

So,
n
n= ∑ k E Xk + E X.
k=1

Now,

(i) by (2.4) and (2.7), E X = O(1),

(ii) by (2.9), if k < k+ then
n k−2
k
E Xk = (1 + o(1)) k ce−c .
ck!
So, by Lemma 2.12,

n k+ kk−1
k
n = o(n) + ∑ ce−c
c k=1 k!
n ∞ kk−1
k
= o(n) + ∑ k! ce−c .
c k=1

Now divide through by n and let n → ∞.

This proves the identity for the case c < 1. Suppose now that c > 1. Then,
since x is a solution of equation ce−c = xe−x , 0 < x < 1, we have
∞
kk−1
k ∞ k−1
k
k
∑ ce−c = ∑ xe−x = x,
k=1 k! k=1 k!

by the first part of the proof (for c < 1).

We note that in fact, Lemma 2.13 is also true for c = 1.

2.2 Super-Critical Phase

The structure of a random graph Gm changes dramatically when m = 12 cn
where c > 1 is a constant. We will give a precise characterisation of this phe-
nomenon, presenting results in terms of Gm and proving them for Gn,p with
p = c/n, c > 1.

Theorem 2.14 If m = cn/2, c > 1, then w.h.p. Gm consists of aunique gi-
2
ant component, with 1 − xc + o(1) n vertices and 1 − xc2 + o(1) cn


2 edges.
Here 0 < x < 1 is the solution of the equation xe−x = ce−c . The remaining
components are of order at most O(log n).
34 Evolution

Proof Suppose that Zk is the number of components of order k in Gn,p . Then,

bounding the number of such components by the number of trees with k ver-
tices that span a component, we get
 
n k−2 k−1
E Zk ≤ k p (1 − p)k(n−k) (2.13)
k
 ne k  c k−1 2
≤A kk−2 e−ck+ck /n
k n
An  k
≤ 2 ce1−c+ck/n
k
Now let β1 = β1 (c) be small enough so that

ce1−c+cβ1 < 1,

and let β0 = β0 (c) be large enough so that

 β0 log n 1
ce1−c+o(1) < 2.
n
If we choose β1 and β0 as above then it follows that w.h.p. there is no compo-
nent of order k ∈ [β0 log n, β1 n].
Our next task is to estimate the number of vertices on small components i.e.
those of size at most β0 log n.
We first estimate the total number of vertices on small tree components, i.e.,
on isolated trees of order at most β0 log n.
1
Assume first that 1 ≤ k ≤ k0 , where k0 = 2α log n, where α is from Lemma
2.12. It follows from (2.9) that
!
k0
n k0 kk−1
k
E ∑ kXk ≈ ∑ ce−c
k=1 c k=1 k!
n ∞ kk−1
k
≈ ∑ ce−c ,
c k=1 k!

using kk−1 /k! < ek , and ce−c < e−1 for c 6= 1 to extend the summation from k0
to infinity.
Putting ε = 1/ log n and using (2.11) we see that the probability that any
Xk , 1 ≤ k ≤ k0 , deviates from its mean by more than 1 ± ε is at most
k0 
(log n)2 (log n)4
 
∑ 1/2−o(1) + O = o(1),
k=1 n n

where the n1/2−o(1) term comes from putting ω ≈ k0 /2 in (2.10). (ω = o(log log n)
in Lemma 2.12. We can however use ω ≈ k0 /2 for k at most k+ /2.)
 2.2 Super-Critical Phase 35

Thus, if x = x(c), 0 < x < 1 is the unique solution in (0, 1) of the equation
xe−x = ce−c , then w.h.p.,

k0
n kk−1
∞
k
∑ kXk ≈ c ∑ xe−x
k=1 k=1 k!
nx
= ,
c
by Lemma 2.13.
Now consider k0 < k ≤ β0 log n.
!
β0 log n
n β0 log n  1−c+ck/n k
∑ kXk ≤ ce
c k=k∑+1
E
k=k0 +1 0
 
= O n(ce1−c )k0
 
= O n1/2+o(1) .

So, by the Markov inequality (see Lemma 21.1), w.h.p.,

β0 log n
∑ kXk = o(n).
k=k0 +1

Now consider the number Yk of non-tree components with k vertices, 1 ≤ k ≤

β0 log n.

!
β0 log n β0 log n    
n k−1 k  c k  c k(n−k)
E ∑ kYk ≤ ∑ k 1−
k=1 k=1 k 2 n n
β0 log n  k
≤ ∑ k ce1−c+ck/n
k=1
= O(1).
So, again by the Markov inequality, w.h.p.,
β0 log n
∑ kYk = o(n).
k=1

Summarising, we have proved so far that w.h.p. there are approximately nx c

vertices on components of order k, where 1 ≤ k ≤ β0 log n and all the remaining
giant components are of size at least β1 n.
We complete the proof by showing the uniqueness of the giant component. Let
log n c1
c1 = c − and p1 = .
n n
36 Evolution

Define p2 by
1 − p = (1 − p1 )(1 − p2 )
log n
and note that p2 ≥ n2
. Then, see Section 1.2,
Gn,p = Gn,p1 ∪ Gn,p2 .
If x1 e−x1 = c1 e−c1 , then x1 ≈ x and so, by our previous analysis, w.h.p.,
Gn,p1 has no components with number of vertices in the range [β0 log n, β1 n].
Suppose there are components C1 ,C2 , . . . ,Cl with |Ci | > β1 n. Here l ≤ 1/β1 .
Now we add edges of Gn,p2 to Gn,p1 . Then
 
l 2
P (∃i, j : no Gn,p2 edge joins Ci with C j ) ≤ (1 − p2 )(β1 n)
2
2
≤ l 2 e−β1 log n
= o(1).
So w.h.p. Gn,p

has a unique component with more than β0 log n vertices and it
has ≈ 1 − xc n vertices.
We now consider the number of edges in the giant C0 . Now we switch to
G = Gn,m . Suppose that the edges of G are e1 , e2 , . . . , em in random order. We
estimate the probability that e = em = {x, y} is an edge of the giant. Let G1
be the graph induced by {e1 , e2 , . . . , em−1 }. G1 is distributed as Gn,m−1 and so
we know that w.h.p. G1 has a unique giant C1 and other components are of
size O(log n). So the probability that e is an edge of the giant is o(1) plus the
probability that x or y is a vertex of C1 . Thus,
  x    x 
P e 6∈ C0 | |C1 | ≈ n 1 − = P e ∩C1 = 0/ | |C1 | ≈ n 1 −
c     c
|C1 | |C1 | − 1 x 2
= 1− 1− ≈ . (2.14)
n n c
It follows that the expected number of edges in the giant is as claimed. To
prove concentration, it is simplest to use the Chebyshev inequality, see Lemma
 now fix i, j ≤ m and let C2 denote the unique giant component of
21.3. So,
Gn,m − ei , e j . Then, arguing as for (2.14),

P(ei , e j ⊆ C0 ) =
o(1) + P(e j ∩C2 6= 0/ | ei ∩C2 6= 0)
/ P(ei ∩C2 6= 0)
/
= (1 + o(1)) P(ei ⊆ C0 ) P(e j ⊆ C0 ).
In the o(1) term, we hide the probability of the event

ei ∩C2 6= 0,
/ e j ∩C2 6= 0,
/ ei ∩ e j 6= 0/
 2.2 Super-Critical Phase 37

which has probability o(1). We should double this o(1) probability here to
account for switching the roles of i, j.
The Chebyshev inequality can now be used to show that the number of edges
is concentrated as claimed.
We will see later, see Lemma 2.17, that w.h.p. each of the small components
have at most one cycle.
From the above theorem and the results of previous sections we see that,
when m = cn/2 and c passes the critical value equal to 1, the typical structure of
a random graph changes from a scattered collection of small trees and unicyclic
components to a coagulated lump of components (the giant component) that
dominates the graph. This short period when the giant component emerges
is called the phase transition. We will look at this fascinating period of the
evolution more closely in Section 2.3.
We know that w.h.p.  component of Gn,m , m = cn/2, c > 1 has ≈
 the giant
2
1 − xc vertices and ≈ 1 − xc2 cn 2 edges. So, if we look at the graph H induced
by the vertices outside the giant, then w.h.p. H has ≈ n1 = nx c vertices and
≈ m1 = xn1 /2 edges. Thus we should expect H to resemble Gn1 .m1 , which is
sub-critical since x < 1. This can be made precise, but the intuition is clear.
Now increase m further and look on the outside of the giant component.
The giant component subsequently consumes the small components not yet
attached to it. When m is such that m/n → ∞ then unicyclic components dis-
appear and a random graph Gm achieves the structure described in the next
theorem.

Theorem 2.15 Let ω = ω(n) → ∞ as n → ∞ be some slowly growing func-

tion. If m ≥ ωn but m ≤ n(log n − ω)/2, then Gm is disconnected and all com-
ponents, with the exception of the giant, are trees w.h.p.

Tree-components of order k die out in the reverse order they were born, i.e.,
larger trees are ”swallowed” by the giant earlier than smaller ones.

Cores
Given a positive integer k, the k-core of a graph G = (V, E) is the largest set
S ⊆ V such that the minimum degree δS in the vertex induced subgraph G[S]
is at least k. This is unique because if δS ≥ k and δT ≥ k then δS∪T ≥ k. Cores
were first discussed by Bollobás [131]. It was shown by Łuczak [551] that for
k ≥ 3 either there is no k-core in Gn,p or one of linear size, w.h.p. The precise
size and first occurrence of k-cores for k ≥ 3 was established in Pittel, Spencer
38 Evolution

and Wormald [645]. The 2-core, C2 which is the set of vertices that lie on at
least one cycle behaves differently to the other cores, k ≥ 3. It grows gradually.
We will need the following result in Section 17.2.
Lemma 2.16 Suppose that c > 1 and that x < 1 is the solution to xe
−x = ce−c .
Then w.h.p. the 2-core C2 of Gn,p , p = c/n has (1 − x) 1 − xc + o(1) n vertices

2
and 1 − xc + o(1) cn 2 edges.

Proof Fix v ∈ [n]. We estimate P(v ∈ C2 ). Let C1 denote the unique giant
component of G1 = Gn,p − v. Now G1 is distributed as Gn−1,p and so C1 ex-
ists w.h.p. To be in C2 , either (i) v has two neighbors in C1 or (ii) v has two
neighbors in some other component. Now because all components other than
C1 have size O(log n) w.h.p., we see that
 
O(log n)  c 2
P((ii)) = o(1) + n = o(1).
2 n
Now w.h.p. |C1 | ≈ 1 − xc n and it is independent of the edges incident with v

and so
P((i)) = o(1) + 1 − P(0 or 1 neighbors in C1 ) =
  
c |C1 |  c |C1 |−1 c
= o(1) + (1 + o(1)) E 1 − 1 − + |C1 | 1 − (2.15)
n n n
= o(1) + 1 − (e−c+x + (c − x)e−c+x )
 x
= o(1) + (1 − x) 1 − ,
c
where the last line follows from the fact that e−c+x = xc . Also, one has to be

|C |
careful when estimating something like E 1 − nc 1 . For this we note that
Jensen’s inequality implies that
 c |C1 |  c E |C1 |
E 1− ≥ 1− = e−c+x+o(1) .
n n
On the other hand, if ng = 1 − xc n,

 c |C1 |
E 1− ≤
n
  
 c |C1 | 
E 1− |C1 | ≥ (1 − o(1))ng P (|C1 | ≥ (1 − o(1))ng )
n
+ P (|C1 | ≤ (1 − o(1))ng ) = e−c+x+o(1) .
It follows from (2.15) that E(|C2 |) ≈ (1 − x) 1 − xc n. To prove concentra-

tion of |C2 |, we can use the Chebyshev inequality as we did in the proof of
Theorem 2.14 to prove concentration for the number of edges in the giant.
 2.3 Phase Transition 39

To estimate the expected number of edges in C2 , we proceed as in Theorem

2.14 and turn to G = Gn,m and estimate the probability that e1 ⊆ C2 . If G0 =
G \ e and C10 is the giant of G0 then e1 is an edge of C2 iff e1 ⊆ C10 or e1 is
contained in a small component. This latter condition is unlikely. Thus,
 0 2
|C1 |  x 2
P(e1 ⊆ C2 ) = o(1) + E = o(1) + 1 − .
n c
The estimate for the expectation of the number of edges in the 2-core follows
immediately and one can prove concentration using the Chebyshev inequality.

2.3 Phase Transition

In the previous two sections we studied the asymptotic behavior of Gm (and
Gn,p ) in the “sub-critical phase” when m = cn, c < 1/2 (p = c/n, c < 1), as well
as in the “super-critical phase” when m > n/2 (p = c/n, c > 1) of its evolution.
We have learned that when m = cn, c > 1/2 our random graph consists w.h.p.
of tree components and components with exactly one cycle (see Theorem 2.1
and Lemma 2.11). We call such components simple while components which
are not simple, i.e. components with at least two cycles, will be called complex.
All components during the sub-critical phase are rather small, of order log n,
tree-components dominate the typical structure of Gm , and there is no sig-
nificant gap in the order of the first and the second largest component. This
follows from Lemma 2.12. The proof of this lemma shows that w.h.p. there
are many trees of height k− . The situation changes when m > n/2, i.e., when
we enter the super-critical phase and then w.h.p. Gm consists of a single giant
complex component (of the order comparable to n), and some number of sim-
ple components, i.e., tree components and components with exactly one cycle
(see Theorem 2.14). One can also observe a clear gap between the order of the
largest component (the giant) and the second largest component which is of the
order O(log n). This phenomenon of dramatic change of the typical structure
of a random graph is called its phase transition.
A natural question arises as to what happens when m/n → 1/2, either from
below or above, as n → ∞. It appears that one can establish, a so called, scaling
window or critical window for the phase transition in which Gm is undergoing
a rapid change in its typical structure. A characteristic feature of this period is
that a random graph can w.h.p. consist of more than one complex component
(recall: there are no complex components in the sub-critical phase and there is
a unique complex component in the super-critical phase).
40 Evolution

Erdős and Rényi [280] studied the size of the largest tree in the random
graph Gn,m when m = n/2 and showed that it was likely to be around n2/3 .
They called the transition from O(log n) through Θ(n2/3 ) to Ω(n) the “dou-
ble jump”. They did not study the regime m = n/2 + o(n). Bollobás [130]
opened the detailed study of this and Łuczak [549] continued this analysis.
He established the precise size of the “scaling window” by removing a log-
arithmic factor from Bollobás’s estimates. The component structure of Gn,m
for m = n/2 + o(n) is rather complicated and the proofs are technically chal-
lenging. We will begin by stating several results that give a an idea of the
component structure in this range, referring the reader elsewhere for proofs:
Chapter 5 of Janson, Łuczak and Ruciński [441]; Aldous [15]; Bollobás [130];
Janson [428]; Janson, Knuth, Łuczak and Pittel [445]; Łuczak [549], [550],
[554]; Łuczak, Pittel and Wierman [557]. We will finish with a proof by Nach-
mias and Peres that when p = 1/n the largest component is likely to have size
of order n2/3 .
The first theorem is a refinement of Lemma 2.10.
Theorem 2.17 Let m = n2 − s, where s = s(n) ≥ 0.

(a) The probability that Gn,m contains a complex component is at most n2 /4s3 .
(b) If s  n2/3 then w.h.p. the largest component is a tree of size asymptotic to
n 3
2s2
log sn .

The next theorem indicates when the phase in which we may have more than
one complex component “ends”, i.e., when a single giant component emerges.

Theorem 2.18 Let m = 2n + s, where s = s(n) ≥ 0. Then the probability that

Gn,m contains more than one complex component is at most 6n2/9 /s1/3 .

For larger s, the next theorem gives a precise estimate of the size of the
largest component for s  n2/3 . For s > 0 we let s̄ > 0 be defined by
       
2s̄ 2s̄ 2s 2s
1− exp = 1+ exp − .
n n n n

Theorem 2.19 Let m = n2 + s where s  n2/3 . Then with probability at least

1 − 7n2/9 /s1/3 ,
2/3
 
L1 − 2(s + s̄)n  ≤ n
 
 n + 2s  5
 2.3 Phase Transition 41

where L1 is the size of the largest component in Gn,m . In addition, the largest
component is complex and all other components are either trees or unicyclic
components.
To get a feel for this estimate of L1 we remark that
4s2
 3
s
s̄ = s − +O 2 .
3n n
The next theorem gives some information about `-components inside the
scaling window m = n/2 + O(n2/3 ). An `-component is one that has ` more
edges than vertices. So trees are (-1)-components.
Theorem 2.20 Let m = 2n + O(n2/3 ) and let r` denote the number of `-
components in Gn,m . For every 0 < δ < 1 there exists Cδ such that if n is
sufficiently large, then with probability at least 1 − δ , ∑`≥3 `r` ≤ Cδ and the
number of vertices on complex components is at most Cδ n2/3 .
One of the difficulties in analysing the phase transition stems from the need
to estimate C(k, `), which is the number of connected graphs with vertex set [k]
and ` edges. We need good estimates for use in first moment calculations. We
have seen the values for C(k, k − 1) (Cayley’s formula) and C(k, k), see (2.6).
For ` > 0, things become more tricky. Wright [748], [749], [750] showed that
Ck,k+` ≈ γ` kk+(3`−1)/2 for ` = o(k1/3 ) where the Wright coefficients γ` satisfy
an explicit recurrence and have been related to Brownian motion, see Aldous
[15] and Spencer [710]. In a breakthrough paper, Bender, Canfield and McKay
[72] gave an asymptotic formula valid for all k. Łuczak [548] in a beautiful
argument simplified a large part of their argument, see Exercise (4.3.6). Bol-
lobás [132] proved the useful simple estimate Ck,k+` ≤ c`−`/2 kk+(3`−1)/2 for
some absolute constant c > 0. It is difficult to prove tight statements about
Gn,m in the phase transition window without these estimates. Nevertheless, it
is possible to see that the largest component should be of size order n2/3 , us-
ing a nice argument from Nachmias and Peres. They have published a stronger
version of this argument in [615].

Theorem 2.21 Let p = n1 and A be a large constant. Let Z be the size of the
largest component in Gn,p . Then
 
1
(i) P Z ≤ n2/3 = O(A−1 ),
A
 
(ii) P Z ≥ An2/3 = O(A−1 ).
42 Evolution

Proof We will prove part (i) of the theorem first. This is a standard application
of the first moment method, see for example Bollobás [132]. Let
 Xk be the
number of tree components of order k and let k ∈ A1 n2/3 , An2/3 . Then, see


also (2.8),
 
n k−2 k−1 k
E Xk = k p (1 − p)k(n−k)+(2)−k+1 .
k
But
k 2 /2
(1 − p)k(n−k)+(2)−k+1 ≈ (1 − p)kn−k
= exp{(kn − k2 /2) log(1 − p)}
kn − k2 /2
 
≈ exp − .
n
Hence, by the above and Lemma 22.2,
k3
 
n
E Xk ≈ √ exp − 2 . (2.16)
2π k5/2 6n
So if
An2/3
X= ∑ Xk ,
1 2/3
An

then
3
1 A e−x /6
Z
EX ≈ √ dx
2π x= A1 x5/2
4
= √ A3/2 + O(A1/2 ).
3 π
Arguing as in Lemma 2.12 we see that

E Xk2 ≤ E Xk + (1 + o(1))(E Xk )2 ,

E(Xk Xl ) ≤ (1 + o(1))(E Xk )(E Xl ), k 6= l.

It follows that
E X 2 ≤ E X + (1 + o(1))(E X)2 .

Applying the second moment method, Lemma 21.6, we see that

1
P(X > 0) ≥
(E X)−1 + 1 + o(1)
= 1 − O(A−1 ),
 2.3 Phase Transition 43

which completes the proof of part (i).

To prove (ii) we first consider a breadth first search (BFS) starting from, say,
vertex x. We construct a sequence of sets S1 = {x}, S2 , . . ., where
Si+1 = {v 6∈ Si : ∃w ∈ Si such that (v, w) ∈ E(Gn,p )}.
We have
 
E(|Si+1 | |Si ) ≤ (n − |Si |) 1 − (1 − p)|Si |
≤ (n − |Si |)|Si |p
≤ |Si |.
So
E |Si+1 | ≤ E |Si | ≤ · · · ≤ E |S1 | = 1. (2.17)
We prove next that
4
πk = P(Sk 6= 0)
/ ≤ . (2.18)
k
This is clearly true for k ≤ 4 and we obtain (2.18) by induction from
n−1  
n−1 i
πk+1 ≤ ∑ p (1 − p)n−1−i (1 − (1 − πk )i ). (2.19)
i=1 i
To explain the above inequality note that we can couple the construction of
S1 , S2 , . . . , Sk with a (branching) process where T1 = {1} and Tk+1 is obtained
from Tk as follows: each Tk independently spawns Bin(n − 1, p) individuals.
Note that |Tk | stochastically dominates |Sk |. This is because in the BFS process,
each w ∈ Sk gives rise to at most Bin(n − 1, p) new vertices. Inequality (2.19)
follows, because Tk+1 6= 0/ implies that at least one of 1’s children give rise to
descendants at level k. Going back to (2.19) we get
πk+1 ≤ 1 − (1 − p)n−1 − (1 − p + p(1 − πk ))n−1 + (1 − p)n−1
= 1 − (1 − pπk )n−1
   
n−1 2 2 n−1 3 3
≤ 1 − 1 + (n − 1)pπk − p πk + p πk
2 3
   
1 1
≤ πk − + o(1) πk2 + + o(1) πk3
2 6
     
1 1
= πk 1 − πk + o(1) − + o(1) πk
2 6
 
1
≤ πk 1 − πk .
4
44 Evolution

This expression increases for 0 ≤ πk ≤ 1 and immediately gives π5 ≤ 3/4 ≤

4/5. In general we have by induction that
 
4 1 4
πk+1 ≤ 1− ≤ ,
k k k+1
completing the inductive proof of (2.18).

Let Cx be the component containing x and let ρx = max{k : Sk 6= 0}

/ in the
BFS from x. Let
n o
X =  x : |Cx | ≥ n2/3  ≤ X1 + X2 ,
 

where
n o
X1 =  x : |Cx | ≥ n2/3 and ρx ≤ n1/3  ,
 

n o
X2 =  x : ρx > n1/3  .
 

It follows from (2.18) that

4
P(ρx > n1/3 ) ≤
n1/3
and so
E X2 ≤ 4n2/3 .

Furthermore,
n o
P |Cx | ≥ n2/3 and ρx ≤ n1/3
 
≤ P |S1 | + . . . + |Sn1/3 | ≥ n2/3
E(|S1 | + . . . + |Sn1/3 |)
≤
n2/3
1
≤ 1/3 ,
n
after using (2.17). So E X1 ≤ n2/3 and E X ≤ 5n2/3 .
Now let Cmax denote the size of the largest component. Now

Cmax ≤ |X| + n2/3

where the addition of n2/3 accounts for the case where X = 0.

So we have
ECmax ≤ 6n2/3
 2.4 Exercises 45

and part (ii) of the theorem follows from the Markov inequality (see Lemma
21.1).

2.4 Exercises
2.4.1 Prove Theorem 2.15.
2.4.2 Show that if p = ω/n where ω = ω(n) → ∞ then w.h.p. Gn,p contains no
unicyclic components. (A component is unicyclic if it contains exactly
one cycle i.e. is a tree plus one extra edge).
2.4.3 Prove Theorem 2.17.
2.4.4 Suppose that m = cn/2 where c > 1 is a constant. Let C1 denote the giant
component of Gn,m , assuming that it exists. Suppose that C1 has n0 ≤ n
vertices and m0 ≤ m edges. Let G1 , G2 be two connected graphs with n0
vertices from [n] and m0 edges. Show that
P(C1 = G1 ) = P(C1 = G2 ).
(I.e. C1 is a uniformly random connected graph with n0 vertices and m0
edges).
2.4.5 Suppose that Z is the length of the cycle in a randomly chosen connected
n

unicyclic graph on vertex set [n]. Show that, where N = 2 ,
nn−2 (N − n + 1)
EZ = .
C(n, n)
2.4.6 Suppose that c < 1. Show that w.h.p. the length of the longest path in
log n
Gn,p , p = nc is ≈ log 1/c .
2.4.7 Suppose that c 6= 1 is constant. Show that w.h.p. the number of edges in
log n
the largest component that is a path in Gn,p , p = nc is ≈ c−log c.
2.4.8 Let Gn,n,p denote the random bipartite graph derived from the complete
bipartite graph Kn,n where each edge is included independently with
probability p. Show that if p = c/n where c > 1 is a constant then w.h.p.
Gn,n,p has a unique giant component of size ≈ 2G(c)n where G(c) is as
in Theorem 2.14.
2.4.9 Consider the bipartite random graph Gn,n,p=c/n , with constant c > 1. De-
fine 0 < x < 1 to be the solution to xe−x
= ce−c . Prove that w.h.p.
the
2
2-core of Gn,n,p=c/n has ≈ 2(1 − x) 1 − xc n vertices and ≈ c 1 − xc n
edges.
 46 Evolution

2.4.10 Let p = 1+εn . Show that if ε is a small positive constant then w.h.p. Gn,p
contains a giant component of size (2ε + O(ε 2 ))n.
2.4.11 Let m = 2n + s, where s = s(n) ≥ 0. Show that if s  n2/3 then w.h.p.
the random graph Gn,m contains exactly one complex component. (A
component C is complex if it contains at least two distinct cycles. In
terms of edges, C is complex iff it contains at last |C| + 1 edges).
2.4.12 Let mk (n) = n(log n + (k − 1) log log n + ω)/(2k), where |ω| → ∞, |ω| =
o(log n). Show that
(
o(1) if ω → −∞
P(Gmk 6⊇ k-vertex-tree-component) = .
1 − o(1) if ω → ∞

2.4.13 Let k ≥ 3 be fixed and let p = nc . Show that if c is sufficiently large, then
w.h.p. the k-core of Gn,p is non-empty.
2.4.14 Let k ≥ 3 be fixed and let p = nc . Show that there exists θ = θ (c, k) > 0
such that w.h.p. all vertex sets S with |S| ≤ θ n contain fewer than k|S|/2
edges. Deduce that w.h.p. either the k-core of Gn,p is empty or it has size
at least θ n.
2.4.15 Suppose that p = nc where c > 1 is a constant. Show that w.h.p. the giant
component of Gn,p is non-planar. (Hint: Assume that c = 1 + ε where ε
is small. Remove a few vertices from the giant so that the girth is large.
Now use Euler’s formula).
2.4.16 Show that if ω = ω(n) → ∞ then w.h.p. Gn,p has at most ω complex
components.
2.4.17 Suppose that np → ∞ and 3 ≤ k = O(1). Show that Gn,p contains a k-
cycle w.h.p.
2.4.18 Suppose that p = c/n where c > 1 is constant and let β = β (c) be the
smallest root of the equation
1  
cβ + (1 − β )ce−cβ = log c(1 − β )(β −1)/β .
2
1 Show that if ω → ∞ and ω ≤ k ≤ β n then w.h.p. Gn,p contains no
maximal induced tree of size k.
2 Show that w.h.p. Gn,p contains an induced tree of size (log n)2 .
3 Deduce that w.h.p. Gn,p contains an induced tree of size at least β n.
2.4.19 Show that if c 6= 1 and xe−x = ce−c where 0 < x < 1 then
(
1 ∞ kk−2 −c k 1− c c < 1.
∑ (ce ) = x 2 x

c k=1 k! 1− c > 1.
c 2
 2.5 Notes 47
≥k
2.4.20 Let GδN,M denote a graph chosen uniformly at random from the set of
graphs with vertex set [N], M edges and minimum degree at least k.
Let Ck denote the k core of Gn,m (if it exists). Show that conditional on
|Ck | = N and |E(Ck )| = M that the graph induced by Ck is distributed as
≥k
GδN,M .

2.5 Notes
Phase transition
The paper by Łuczak, Pittel and Wierman [557] contains a great deal of in-
formation about the phase transition. In particular, [557] shows that if m =
n/2 + λ n2/3 then the probability that Gn,m is planar tends to a limit p(λ ),
where p(λ ) → 0 as λ → ∞. The landmark paper by Janson, Knuth, Łuczak
and Pittel [445] gives the most detailed analysis to date of the events in the
scaling window.
Outside of the critical window n2 ± O(n2/3 ) the size of the largest component
is asymptotically determined. Theorem 2.17 describes Gn,m before reaching
the window and on the other hand a unique “giant” component of size ≈ 4s
begins to emerge at around m = n2 + s, for s  n2/3 . Ding, Kim, Lubetzky and
Peres [249] give a useful model for the structure of this giant.

Achlioptas processes
Dimitris Achlipotas proposed the following variation on the basic graph pro-
cess. Suppose that instead of adding a random edge ei to add to Gi−1 to create
Gi , one is given a choice of two random edges ei , fi and one chooses one of
them to add. He asked whether it was possible to come up with a choice rule
that would delay the occurrence of some graph property P. As an initial chal-
lenge he asked whether it was possible to delay the production of a giant com-
ponent beyond n/2. Bohman and Frieze [114] showed that this was possible
by the use of a simple rule. Since that time this has grown into a large area of
research. Kang, Perkins and Spencer [473] have given a more detailed analy-
sis of the “Bohman-Frieze” process. Bohman and Kravitz [121] and in greater
generality Spencer and Wormald [712] analyse “bounded size algorithms” in
respect of avoiding giant components. Flaxman, Gamarnik and Sorkin [311]
consider how to speed up the occurrence of a giant component. Riordan and
Warnke [666] discuss the speed of transition at a critical point in an Achlioptas
process.
48 Evolution

The above papers concern component structure. Krivelevich, Loh and Su-
dakov [516] considered rules for avoiding specific subgraphs. Krivelevich, Lu-
betzky and Sudakov [517] discuss rules for speeding up Hamiltonicity.

Graph Minors
Fountoulakis, Kühn and Osthus [317] show that for every ε > 0 there exists Cε
such that if np >  p = o(1) then w.h.p. Gn,p contains a complete minor
 Cε and
2
n p
of size (1 ± ε) log np . This improves earlier results of Bollobás, Catlin and
Erdős [136] and Krivelevich and Sudakov [522]. Ajtai, Komlós and Szemerédi
[9] showed that if np ≥ 1 + ε and np = o(n1/2 ) then w.h.p. Gn,p contains a top-
logical clique of size almost as large as the maximum degree. If we know that
Gn,p is non-planar w.h.p. then it makes sense to determine its thickness. This is
the minimum number of planar graphs whose union is the whole graph. Cooper
[202] showed that the thickness of Gn,p is strongly related to its arboricity and
is asymptotic to np/2 for a large range of p.
 3
Vertex Degrees

In this chapter we study some typical properties of the degree sequence of a

random graph. We begin by discussing the typical degrees in a sparse random
graph i.e. one with O(n) edges and prove some results on the asymptotic dis-
tribution of degrees. Next we look at the typical values of the minimum and
maximum degrees in dense random graphs. We then describe a simple canoni-
cal labelling algorithm for the graph isomorphism problem on a dense random
graph.

3.1 Degrees of Sparse Random Graphs

Recall that the degree of an individual vertex of Gn,p is a Binomial random
variable with parameters n − 1 and p. One should also notice that the degrees
of different vertices are only mildly correlated.
We will first prove some simple but often useful properties of vertex degrees
when p = o(1). Let X0 = Xn,0 be the number of isolated vertices in Gn,p . In
Lemma 1.11, we established the sharp threshold for “disappearance” of such
vertices. Now we will be more precise and determine the asymptotic distribu-
tion of X0 “below”, “on” and “above” the threshold. Obviously,
E X0 = n(1 − p)n−1 ,
and an easy computation shows that, as n → ∞,

∞

 if np − log n → −∞
E X0 → e−c if np − log n → c, c < ∞, (3.1)

if np − log n → ∞

0

We denote by Po(λ ) a random variable with the Poisson distribution with

parameter λ , while N(0, 1) denotes the random variable with the Standard Nor-
D
mal distribution. We write Xn → X to say that a random variable Xn converges
in distribution to a random variable X, as n → ∞.
The following theorem shows that the asymptotic distribution of X0 passes
through three phases: it starts in the Normal phase; next when isolated vertices

49
50 Vertex Degrees

are close to “dying out”, it moves through a Poisson phase; it finally ends up
at the distribution concentrated at 0.

Theorem 3.1 Let X0 be the random variable counting isolated vertices in a

random graph Gn,p . Then, as n → ∞,
D
(i) X̃0 = (X0 − E X0 )/(Var X0 )1/2 → N(0, 1),
if n2 p → ∞ and np − log n → −∞,
D
(ii) X0 → Po(e−c ), if np − log n → c, c < ∞,
D
(iii) X0 → 0, if np − log n → ∞.
Proof For the proof of (i) we refer the reader to Chapter 6 of Janson, Łuczak
and Ruciński [441] (or to [61] and [510]).
To prove (ii) one has to show that if p = p(n) is such that np − log n → c ,
then

e−ck −e−c
lim P(X0 = k) = e , (3.2)
n→∞ k!
for k = 0, 1, ... . Now,
X0 = ∑ Iv ,
v∈V

where
(
1 if v is an isolated vertex in Gn,p
Iv =
0 otherwise.
So
E X0 = ∑ E Iv = n(1 − p)n−1
v∈V
= n exp{(n − 1) log(1 − p)}
( )
∞
pk
= n exp −(n − 1) ∑
k=1 k

= n exp −(n − 1)p + O(np2 )



(log n)2
  
= n exp −(log n + c) + O
n
≈ e−c . (3.3)
The easiest way to show that (3.2) holds is to apply the Method of Moments
(see Chapter 21). Briefly, since the distribution of the random variable X0 is
 3.1 Degrees of Sparse Random Graphs 51

uniquely determined by its moments, it is enough to show, that either the kth

moment E X0 (X0 − 1) · · · (X0 − k + 1) of X0 , or its binomial moment
factorial
E Xk0 , tend to the respective moments of the Poisson distribution, i.e., to either
e−ck or e−ck /k!. We choose the binomial moments, and so let
 
(n) X0
Bk = E ,
k
then, for every non-negative integer k,
(n)
Bk = ∑ P(Ivi1 = 1, Ivi2 = 1, . . . , Ivik = 1),
1≤i1 <i2 <···<ik ≤n
 
n k
= (1 − p)k(n−k)+(2) .
k
Hence
(n) e−ck
lim Bk = ,
n→∞ k!
and part (ii) of the theorem follows by Theorem 21.11, with λ = e−c .
For part (iii), suppose that np = log n + ω where ω → ∞. We repeat the
calculation estimating E X0 and replace ≈ e−c in (3.3) by ≤ (1 + o(1))e−ω → 0
and apply the first moment method.
From the above theorem we immediately see that if np − log n → c then
−c
lim P(X0 = 0) = e−e . (3.4)
n→∞

We next give a more general result describing the asymptotic distribution

of the number Xd = Xn,d , d ≥ 1 of vertices of any fixed degree d in a random
graph.
Recall, that the degree of a vertex in Gn,p has the binomial distribution
Bin(n − 1, p). Hence,
 
n−1 d
E Xd = n p (1 − p)n−1−d . (3.5)
d
Therefore, as n → ∞,


 0 if p  n−(d+1)/d ,

if p ≈ cn−(d+1)/d , c < ∞,



 λ1

if p  n−(d+1)/d) but

∞
E Xd → (3.6)


 pn − log n − d log log n → −∞,

if pn − log n − d log log n → c, c < ∞,



 λ2

0 if pn − log n − d log log n → ∞,

52 Vertex Degrees

where
cd e−c
λ1 = and λ2 = . (3.7)
d! d!
The asymptotic behavior of the expectation of the random variable Xd sug-
gests possible asymptotic distributions for Xd , for a given edge probability p.

Theorem 3.2 Let Xd = Xn,d be the number of vertices of degree d,

d ≥ 1, in Gn,p and let λ1 , λ2 be given by (3.7). Then, as n → ∞,
D
(i) Xd → 0 if p  n−(d+1)/d ,
D
(ii) Xd → Po(λ1 ) if p ≈ cn−(d+1)/d , c < ∞,
D
(iii) X̃d := (Xd − E Xd )/(Var Xd )1/2 → N(0, 1) if p  n−(d+1)/d , but
pn − log n − d log log n → −∞
D
(iv) Xd → Po(λ2 ) if pn − log n − d log log n → c, −∞ < c < ∞,
D
(v) Xd → 0 if pn − log n − d log log n → ∞
Proof The proofs of statements (i) and (v) are straightforward applications of
the first moment method, while the proofs of (ii) and (iv) can be found in Chap-
ter 3 of Bollobás [124] (see also Karoński and Ruciński [481] for estimates of
the rate of convergence). The proof of (iii) can be found in [61].

The next theorem shows the concentration of Xd around its expectation when
in Gn,p the edge probability p = c/n, i.e., when the average vertex degree is c.

Theorem 3.3 Let p = c/n where c is a constant. Let Xd denote the number
of vertices of degree d in Gn,p . Then, for d = O(1), w.h.p.
cd e−c
Xd ≈ n.
d!
Proof Assume that vertices of Gn,p are labeled 1, 2, . . . , n. We first compute
E Xd . Thus,
E Xd = n P(deg(1) = d) =
 
n − 1  c d  c n−1−d
=n 1−
d n n
d
  2       
n d c d c 1
=n 1+O exp −(n − 1 − d) +O 2
d! n n n n
cd e−c
  
1
=n 1+O .
d! n
 3.1 Degrees of Sparse Random Graphs 53

We now compute the second moment. For this we need to estimate

P(deg(1) = deg(2) = d)
c n−1−d 2
  
c n − 2  c d−1 
= 1−
n d −1 n n
c n−2−d 2
    
 c  n−2 c d
+ 1− 1−
n d n n
  
1
= P(deg(1) = d) P(deg(2) = d) 1 + O .
n
The first line here accounts for the case where {1, 2} is an edge and the second
line deals with the case where it is not.
Thus

Var Xd =
n n
= ∑ ∑ [P(deg(i) = d, deg( j) = d) − P(deg(1) = d) P(deg(2) = d)]
i=1 j=1
n  
1
≤ ∑ O + E Xd ≤ An,
i6= j=1 n

for some constant A = A(c).

Applying the Chebyshev inequality ( Lemma 21.3), we obtain
A
P(|Xd − E Xd | ≥ tn1/2 ) ≤ ,
t2
which completes the proof.
We conclude this section with a look at the asymptotic behavior of the max-
imum vertex degree, when a random graph is sparse.

Theorem 3.4 Let ∆(Gn,p ) (δ (Gn,p )) denotes the maximum (minimum) degree
of vertices of Gn,p .

(i) If p = c/n for some constant c > 0 then w.h.p.

log n
∆(Gn,p ) ≈ .
log log n
(ii) If np = ω log n where ω → ∞, then w.h.p. δ (Gn,p ) ≈ ∆(Gn,p ) ≈ np.
l m
log n
Proof (i) Let d± = log log n±2 log log log n . Then, if d = d− ,
 
n − 1  c d
P(∃v : deg(v) ≥ d) ≤ n
d n
54 Vertex Degrees
 ce d
≤n
d
= exp {log n − d log d + O(d)} (3.8)
log log log n
Let λ = log log n . Then
log n 1
d log d ≥ · · (log log n − log log log n + o(1))
log log n 1 − 2λ
log n
= (1 + 2λ + O(λ 2 ))(log log n − log log log n + o(1))
log log n
log n
= (log log n + log log log n + o(1)). (3.9)
log log n
Plugging this into (3.8) shows that ∆(Gn,p ) ≤ d− w.h.p.
Now let d = d+ and let Xd be the number of vertices of degree d in Gn,p .
Then
 
n − 1  c d  c n−d−1
E(Xd ) = n 1−
d n n
= exp {log n − d log d + O(d)}
 
log n
= exp log n − (log log n − log log log n + o(1)) + O(d) (3.10)
log log n
→ ∞.
Here (3.10) is obtained by using −λ in place of λ in the argument for (3.9).
Now, for vertices v, w, by the same argument as in the proof of Theorem 3.3,
we have
P(deg(v) = deg(w) = d) = (1 + o(1)) P(deg(v) = d) P(deg(w) = d),
and the Chebyshev inequality implies that Xd > 0 w.h.p. This completes the
proof of (i).
Statement (ii) is an easy consequence of the Chernoff bounds, Corollary
22.7. Let ε = ω −1/3 . Then
2 np/3 1/3 /3
P(∃v : |deg(v) − np| ≥ εnp) ≤ 2ne−ε = 2n−ω = o(n−1 ).

3.2 Degrees of Dense Random Graphs

In this section we will concentrate on the case where edge probability p is con-
stant and see how the degree sequence can be used to solve the graph isomor-
 3.2 Degrees of Dense Random Graphs 55

phism problem w.h.p. The main result deals with the maximum vertex degree
in dense random graph and is instrumental in the solution of this problem.
p
Theorem 3.5 Let d± = (n − 1)p + (1 ± ε) 2(n − 1)pq log n, where q = 1 −
p. If p is constant and ε > 0 is a small constant, then w.h.p.

(i) d− ≤ ∆(Gn,p ) ≤ d+ .
(ii) There is a unique vertex of maximum degree.

Proof We break the proof of Theorem 3.5 into two lemmas.

p
Lemma 3.6 Let d = (n−1)p+x (n − 1)pq, p be constant, x ≤ n1/3 (log n)2 ,
where q = 1 − p. Then
  s
n−1 d 1 2
Bd = p (1 − p)n−1−d
= (1 + o(1)) e−x /2 .
d 2πnpq

Proof Stirling’s formula gives

s  d   d !n−1
(n − 1)p n−1 (n − 1)q 1− n−1

1
Bd = (1 + o(1)) .
2πnpq d n−1−d
(3.11)
Now
  d  r  d
d n−1 q n−1
= 1+x =
(n − 1)p (n − 1)p
x2 q x3
 r    r 
q pq
= exp x − + O 3/2 p+x
(n − 1)p 2(n − 1)p n n−1
2
 r  3 
pq x q x
= exp x + + O 3/2 ,
n − 1 2(n − 1) n
whereas
 d 1− d
n − 1 − d 1− n−1
  r
p n−1
= 1−x =
(n − 1)q (n − 1)q
x2 p
  r  3   r 
p x pq
= exp − x + + O 3/2 q−x
(n − 1)q 2(n − 1)q n n−1
2
 r  3 
pq x p x
= exp −x + + O 3/2 ,
n − 1 2(n − 1) n
So
d 1− d
x2
     3 
d n−1 n−1−d n−1 x
= exp + O 3/2 ,
(n − 1)p (n − 1)q 2(n − 1) n
56 Vertex Degrees

and lemma follows from (3.11).

The next lemma proves a strengthing of Theorem 3.5.
Lemma 3.7 Let ε = 1/10, and p be constant and q = 1 − p. If
p
d± = (n − 1)p + (1 ± ε) 2(n − 1)pq log n.
then w.h.p.
(i) ∆(Gn,p ) ≤ d+ ,
(ii) There are Ω(n2ε(1−ε) ) vertices of degree at least d− ,
(iii) 6 ∃ u 6= v such that deg(u), deg(v) ≥ d− and |deg(u) − deg(v)| ≤ 10.
Proof We first prove that as x → ∞,
  Z∞
1 −x2 /2 1 2 1 2
e 1− 2 ≤ e−y /2 dy ≤ e−x /2 . (3.12)
x x x x
To see this notice
1 −y2 /2 0
Z ∞ Z ∞ 
2 /2
e−y dy = − e dy
x y x
1 −y2 /2 ∞
 
1 −y2 /2
Z ∞
=− e − 2
e dy
y x x y
 ∞
1 2 1 1 −y2 /2
Z ∞
2
= e−x /2 + 3 e−y /2 + 3 4
e dy
x y x x y
   
1 2 1 1 2
= e−x /2 1 − 2 + O 4 e−x /2 .
x x x
We can now prove statement (i).
Let Xd be the number of vertices of degree d. Then E Xd = nBd and so Lemma
3.6 implies that
 ! 
r
n  1 d − (n − 1)p 2 
E Xd = (1 + o(1)) exp − p
2π pq  2 (n − 1)pq 

assuming that
p
d ≤ dL = (n − 1)p + (log n)2 (n − 1)pq.
Also, if d > (n − 1)p then
Bd+1 (n − d − 1)p
= <1
Bd (d + 1)q
and so if d ≥ dL ,
E Xd ≤ E XdL ≤ n exp{−Ω(log n)4 }.
 3.2 Degrees of Dense Random Graphs 57

It follows that
∆(Gn,p ) ≤ dL w.h.p. (3.13)

Now if Yd = Xd + Xd+1 + · · · + XdL for d = d± then

 !2 
dL r
n  1 l − (n − 1)p 
EYd ≈ ∑ exp − p
l=d 2π pq  2 (n − 1)pq 
 !2 
∞ r
n  1 l − (n − 1)p 
≈∑ exp − p (3.14)
l=d 2π pq  2 (n − 1)pq 
 !2 
1 λ − (n − 1)p 
r
n
Z ∞ 
≈ exp − p dλ .
2π pq λ =d  2 (n − 1)pq 

The justification for (3.14) comes from

 ! 
∞ r
n  1 l − (n − 1)p 2 
∑ 2π pq exp − 2 p(n − 1)pq  =
l=dL
∞
2 /2 2 /3
= O(n) ∑ e−x = O(e−(log n) ),
x=(log n)2

and
 !2 
d+ − (n − 1)p
r  1
n 
exp − p = n−O(1) .
2π pq  2 (n − 1)pq 
p
If d = (n − 1)p + x (n − 1)pq then, from (3.12) we have
 !2 
1 λ − (n − 1)p 
r
n
Z ∞ 
EYd ≈ exp − p dλ
2π pq λ =d  2 (n − 1)pq 
r
n p
Z ∞
2
= (n − 1)pq e−y /2 dy
2π pq y=x
n 1 −x2 /2
≈√ e
2π x
(
≤ n−2ε(1+ε) d = d+
. (3.15)
≥ n2ε(1−ε) d = d−

Part (i) follows from (3.15).

When d = d− we see from (3.15) that EYd → ∞. We use the second moment
58 Vertex Degrees

method to show that Yd− 6= 0 w.h.p.

dL
EYd (Yd − 1) = n(n − 1) ∑ P(deg(1) = d1 , deg(2) = d2 )
d≤d1 ,d2
dL
= n(n − 1) ∑ ˆ = d1 − 1, d(2)
(p P(d(1) ˆ = d2 − 1)
d≤d1 ,d2
ˆ = d1 , d(2)
+ (1 − p) P(d(1) ˆ = d2 )),
ˆ is the number of neighbors of x in {3, 4, . . . , n}. Note that d(1)
where d(x) ˆ and
ˆ
d(2) are independent, and
n−2

d1 −1 (1 − p)
ˆ = d1 − 1)
P(d(1) d1 (1 − p)
= =
ˆ
P(d(1) = d1 ) n−2

p (n − 1 − d1 )p
d 1

= 1 + Õ(n−1/2 ).
In Õ we ignore polylog factors.
Hence
E(Yd (Yd − 1))
dL h i
= n(n − 1) ∑ P(d(1) ˆ = d2 )(1 + Õ(n−1/2 ))
ˆ = d1 ) P(d(2)
d≤d1 ,d2
dL h i
= n(n − 1) ∑ P(deg(1) = d1 ) P(deg(2) = d2 )(1 + Õ(n−1/2 ))
d≤d1 ,d2

= EYd (EYd − 1)(1 + Õ(n−1/2 )),

since
n−2

ˆ = d1 )
P(d(1) d1
= n−1

(1 − p)−1
P(deg(1) = d1 ) d1
−1/2
= 1 + Õ(n ).
So, with d = d−
 
1
P Yd ≤ EYd
2
E(Yd (Yd − 1)) + EYd − (EYd )2
≤
(EYd )2 /4
 
1
= Õ ε
n
= o(1).
 3.2 Degrees of Dense Random Graphs 59

This completes the proof of statement (ii). Finally,

  dL
n
P(¬(iii)) ≤ o(1) + ∑ P(deg(1) = d1 , deg(2) = d2 )
2 d∑
1 =d− |d2 −d1 |≤10
  dL
n h
ˆ = d1 − 1) P(d(2)
ˆ = d2 − 1)
= o(1) + ∑ p P(d(1)
2 d∑
1 =d− |d2 −d1 |≤10
i
ˆ = d1 ) P(d(2)
+ (1 − p) P(d(1) ˆ = d2 ) ,

Now
dL
∑ ∑ ˆ = d1 − 1) P(d(2)
P(d(1) ˆ = d2 − 1)
d1 =d− |d2 −d1 |≤10
dL
≤ 21(1 + Õ(n−1/2 )) ˆ = d1 − 1) 2 ,
 
∑ P(d(1)
d1 =d−

and by Lemma 3.6 and by (3.12) we have with

d− − (n − 1)p p
x= p ≈ (1 − ε) 2 log n,
(n − 1)pq

dL
1
Z ∞
ˆ = d1 − 1) 2 ≈ 2
e−y dy
 
∑ P(d(1)
d1 =d− 2π pqn y=x

1 ∞ Z
−z2 /2
=√ √ e dz
8π pqn z=x 2
1 1 2
≈√ √ n−2(1−ε) ,
8π pqn x 2
ˆ = d1 2 . Thus
We get a similar bound for ∑ddL1 =d− ∑|d2 −d1 |≤10 P(d(1)
 

 2

P(¬(iii)) = o n2−1−2(1−ε)
= o(1).

Application to graph isomorphism

In this section we describe a procedure for canonically labelling a graph G. It
is taken from Babai, Erdős and Selkow [43]. If the procedure succeeds then
it is possible to quickly tell whether G ∼
= H for any other graph H. (Here ∼ =
stands for graph isomorphism).
Algorithm LABEL
60 Vertex Degrees

Step 0: Input graph G and parameter L.

Step 1: Re-label the vertices of G so that they satisfy

dG (v1 ) ≥ dG (v2 ) ≥ · · · ≥ dG (vn ).

If there exists i < L such that dG (vi ) = dG (vi+1 ), then FAIL.

Step 2: For i > L let

Xi = { j ∈ {1, 2, . . . , L} : vi , v j ∈ E(G)}.

Re-label vertices vL+1 , vL+2 , . . . , vn so that these sets satisfy

XL+1  XL+2  · · ·  Xn

where  denotes lexicographic order.

If there exists i < n such that Xi = Xi+1 then FAIL.

Suppose now that the above ordering/labelling procedure LABEL succeeds

for G. Given an n vertex graph H, we run LABEL on H.

(i) If LABEL fails on H then G ∼6 H.

=
(ii) Suppose that the ordering generated on V (H) is w1 , w2 , . . . , wn . Then

G∼
= H ⇔ vi → wi is an isomorphism.

It is straightforward to verify (i) and (ii).

Theorem 3.8 Let p be a fixed constant, q = 1 − p, and let ρ = p2 + q2 and

let L = 3 log1/ρ n. Then w.h.p. LABEL succeeds on Gn,p .

Proof Lemma 3.7 implies that Step 1 succeeds w.h.p. We must now show that
w.h.p. Xi 6= X j for all i 6= j > L. There is a slight problem because the edges
from vi , i > L to v j , j ≤ L are conditioned by the fact that the latter vertices are
those of highest degree.
Now fix i, j and let Ĝ = Gn,p \ {vi , v j }. It follows from Lemma 3.7 that
if i, j > L then w.h.p. the L largest degree vertices of Ĝ and Gn,p coincide.
So, w.h.p., we can compute Xi , X j with respect to Ĝ to create X̂i , X̂ j , which
are independent of the edges incident with vi , v j . It follows that if i, j > L then
X̂i = Xi and X̂ j = X j and this avoids our conditioning problem. Denote by NĜ (v)
the set of the neighbors of vertex v in graph Ĝ. Then

P(Step 2 fails)
 3.3 Exercises 61

≤ o(1) + P(∃vi , v j : NĜ (vi ) ∩ {v1 , . . . , vL } = NĜ (v j ) ∩ {v1 , . . . , vL })

 
n
≤ o(1) + (p2 + q2 )L
2
= o(1).

Corollary 3.9 If 0 < p < 1 is constant then w.h.p. Gn,p has a unique auto-
morphism, i.e. the identity automorphism.

See Exercise 3.3.7.

Application to edge coloring

The chromatic index χ 0 (G) of a graph G is the minimum number of colors that
can be used to color the edges of G so that if two edges share a vertex, then
they have a different color. Vizing’s theorem states that

∆(G) ≤ χ 0 (G) ≤ ∆(G) + 1.

Also, if there is a unique vertex of maximum degree, then χ 0 (G) = ∆(G). So, it
follows from Theorem 3.5 (ii) that, for constant p, w.h.p. we have χ 0 (Gn,p ) =
∆(Gn,p ).

3.3 Exercises
3.3.1 Suppose that m = dn/2 where d is constant. Prove that the number of
k e−d
vertices of degree k in Gn,m is asymptotically equal to d k! n for any
fixed positive integer k.
3.3.2 Suppose that c > 1 and that x < 1 is the solution to xe−x = ce−c . Show
c
that if c = O(1)
 is fixed  then w.h.p. the giant component of Gn,p , p = n
k −c k
has ≈ c k! e
1 − xc


n vertices of degree k ≥ 1.
3.3.3 Suppose that p ≤ 1+ε n
n where n
1/4 ε → 0. Show that if Γ is the sub-graph
n
of Gn,p induced by the 2-core C2 , then Γ has maximum degree at most
three.
3.3.4 Let p = log n+d log
n
log n+c
, d ≥ 1. Using the method of moments, prove
that the number of vertices of degree d in Gn,p is asymptotically Poisson
−c
with mean ed! .
3.3.5 Prove parts (i) and (v) of Theorem 3.2.
62 Vertex Degrees

3.3.6 Show that if 0 < p < 1 is constant then w.h.p. the minimum degree δ in
Gn,p satisfies
p p
|δ − (n − 1)q − 2(n − 1)pq log n| ≤ ε 2(n − 1)pq log n,

where q = 1 − p and ε = 1/10.

3.3.7 Use the canonical labelling of Theorem 3.8 to show that w.h.p. Gn,1/2
has exactly one automprphism, the identity automorphism. (An auto-
morphism of a graph G = (V, E) is a map ϕ : V → V such that {x, y} ∈ E
if and only if {ϕ(x), ϕ(y)} ∈ E.)

3.4 Notes
For the more detailed account of the properties of the degree sequence of Gn,p
the reader is referred to Chapter 3 of Bollobás [132].
Erdős and Rényi [279] and [281] were first to study the asymptotic distribu-
tion of the number Xd of vertices of degree d in relation with connectivity of
a random graph. Bollobás [128] continued those investigations and provided
detailed study of the distribution of Xd in Gn,p when 0 < lim inf np(n)/ log n ≤
lim sup np(n)/ log n < ∞. Palka [628] determined certain range of the edge
probability p for which the number of vertices of a given degree of a ran-
dom graph Gn,p has a Normal distribution. Barbour [58] and Karoński and
Ruciński [481] studied the distribution of Xd using the Stein–Chen approach.
A complete answer to the asymptotic Normality of Xd was given by Barbour,
Karoński and Ruciński [61] (see also Kordecki [510]). Janson [434] extended
those results and showed that random variables counting vertices of given de-
gree are jointly normal, when p ≈ c/n in Gn,p and m ≈ cn in Gn,m , where c is
a constant.
Ivchenko [423] was the first to analyze the asymptotic behavior the kth-
largest and kth smallest element of the degree sequence of Gn,p . In particular
he analysed the span between the minimum and the maximum degree of sparse
Gn,p . Similar results were obtained independently by Bollobás [126] (see also
Palka [629]). Bollobás [128] answered the question for what values of p(n),
Gn,p w.h.p. has a unique vertex of maximum degree (see Theorem 3.5).
Bollobás [123], for constant p, 0 < p < 1, i.e., when Gn,p is dense, gave
an estimate of the probability that maximum degree does not exceed pn +
√
O( n log n). A more precise result was proved by Riordan and Selby [663]
who showed that for constant p,pthe probability that the maximum degree
of Gn,p does not exceed pn + b np(1 − p), where b is fixed, is equal to
 3.4 Notes 63

(c + o(1))n , for c = c(b) the root of a certain equation. Surprisingly, c(0) =

0.6102... is greater than 1/2 and c(b) is independent of p.
McKay and Wormald [583] proved that for a wide range of functions p =
p(n), the distribution of the degree sequence of Gn,p can be approximated
by {(X1 , . . . , Xn )| ∑ Xi is even}, where X1 , . . . , Xn are independent random vari-
ables each having the Binomial distribution Bin(n − 1, p0 ), where p0 is itself a
random variable with a particular truncated normal distribution
 4
Connectivity

We first establish, rather precisely, the threshold for connectivity. We then view
this property in terms of the graph process and show that w.h.p. the random
graph becomes connected at precisely the time when the last isolated vertex
joins the giant component. This “hitting time” result is the pre-cursor to several
similar results. After this we deal with k-connectivity.

4.1 Connectivity
The first result of this chapter is from Erdős and Rényi [279].

Theorem 4.1 Let m = 12 n (log n + cn ). Then


0

 if cn → −∞,
−c
lim P(Gm is connected) = e−e if cn → c (constant)
n→∞ 
if cn → ∞.

1

Proof To prove the theorem we consider, as before, a random graph Gn,p . It

suffices to prove that, when p = lognn+c ,
−c
P(Gn,p is connected ) → e−e .

and use Theorem 1.4 to translate to Gm and then use (1.7) and monotonicity
for cn → ±∞.
Let Xk = Xk,n be the number of components with k vertices in Gn,p and
consider the complement of the event that Gn,p is connected. Then

P(Gn,p is not connected )

 
n/2
[
= P (Gn,p has a component of order k) =
k=1
 
n/2
[
P {Xk > 0} .
k=1

64
 4.1 Connectivity 65

Note that X1 counts here isolated vertices and therefore

n/2
P(X1 > 0) ≤ P(Gn,p is not connected ) ≤ P(X1 > 0) + ∑ P(Xk > 0).
k=2

Now
n/2 n/2 n/2   n/2
n k−2 k−1
∑ P(Xk > 0) ≤ ∑ E Xk ≤ ∑ k p (1 − p)k(n−k) = ∑ uk .
k=2 k=2 k=2 k k=2

Now, for 2 ≤ k ≤ 10,

 k−1
log n + c log n+c
uk ≤ e n k k
e−k(n−10) n
n
 k−1
log n
≤ (1 + o(1))ek(1−c) ,
n
and for k > 10
log n + c k−1 −k(log n+c)/2
 ne k 
uk ≤ kk−2 e
k n
!k
e1−c/2+o(1) log n
≤n .
n1/2
So
n/2
e−c log n n/2 1+o(1)−k/2
∑ uk ≤ (1 + o(1)) n
+ ∑ n
k=2 k=10
 
= O no(1)−1 .

It follows that
P(Gn,p is connected ) = P(X1 = 0) + o(1).
But we already know (see Theorem 3.1) that for p = (log n + c)/n the num-
ber of isolated vertices in Gn,p has an asymptotically Poisson distribution and
therefore, as in (3.4)
−c
lim P(X1 = 0) = e−e ,
n→∞

and so the theorem follows.

It is possible to tweak the proof of Theorem 4.1 to give a more precise result
stating that a random graph becomes connected exactly at the moment when
the last isolated vertex disappears.
 66 Connectivity

Theorem 4.2 Consider the random graph process {Gm }. Let

m∗1 = min{m : δ (Gm ) ≥ 1},

m∗c = min{m : Gm is connected}.

Then, w.h.p.,
m∗1 = m∗c .
Proof Let
1 1
m± = n log n ± n log log n,
2 2
and
m± log n ± log log n
p± = ≈ .
N n
We first show that w.h.p.
(i) Gm− consists of a giant connected component plus a set V1 of at most 2 log n
isolated vertices,
(ii) Gm+ is connected.
Assume (i) and (ii). It follows that w.h.p.
m− ≤ m∗1 ≤ m∗c ≤ m+ .
Since Gm− consists of a connected component and a set of isolated vertices V1 ,
to create Gm+ we add m+ − m− random edges. Note that m∗1 = m∗c if none of
these edges is contained in V1 . Thus
1 2
2 |V1 |
P(m∗1 < m∗c ) ≤ o(1) + (m+ − m− )
N − m+
2n((log n)2 ) log log n
≤ o(1) + 1 2
2 n − O(n log n)
= o(1).
Thus to prove the theorem, it is sufficient to verify (i) and (ii).
Let
m− log n − log log n
p− = ≈ ,
N n
and let X1 be the number of isolated vertices in Gn,p− . Then
E X1 = n(1 − p− )n−1
≈ ne−np−
 4.1 Connectivity 67

≈ log n.

Moreover

E X12 = E X1 + n(n − 1)(1 − p− )2n−3

≤ E X1 + (E X1 )2 (1 − p− )−1 .

So,
Var X1 ≤ E X1 + 2(E X1 )2 p− ,

and

P(X1 ≥ 2 log n) = P(|X1 − E X1 | ≥ (1 + o(1)) E X1 )

 
1
≤ (1 + o(1)) + 2p−
E X1
= o(1).

Having at least 2 log n isolated vertices is a monotone property and so w.h.p.

Gm− has less then 2 log n isolated vertices.
To show that the rest of Gm is a single connected component we let Xk , 2 ≤
k ≤ n/2 be the number of components with k vertices in G p− . Repeating the
calculations for p− from the proof of Theorem 4.1, we have
!
n/2  
E ∑ Xk = O no(1)−1 .
k=2

Let
E = {∃ component of order 2 ≤ k ≤ n/2}.

Then
√
P(Gm− ∈ E ) ≤ O( n) P(Gn,p− ∈ E )
= o(1),

and this completes the proof of (i).

To prove (ii) (that Gm+ is connected w.h.p.) we note that (ii) follows from
the fact that Gn,p is connected w.h.p. for np − log n → ∞ ( see Theorem 4.1).
By implication Gm is connected w.h.p. if n Nm − log n → ∞. But,

nm+ n( 1 n log n + 21 n log log n)

= 2
N N
≈ log n + log log n.
 68 Connectivity

4.2 k-connectivity

In this section we show that the threshold for the existence of vertices of degree
k is also the threshold for the k-connectivity of a random graph. Recall that a
graph G is k-connected if the removal of at most k − 1 vertices of G does not
disconnect it. In the light of the previous result it should be expected that a
random graph becomes k-connected as soon as the last vertex of degree k − 1
disappears. This is true and follows from the results of Erdős and Rényi [281].
Here is a weaker statement.

Theorem 4.3 Let m = 12 n (log n + (k − 1) log log n + cn ) , k = 1, 2, . . .. Then



 0 if cn → −∞
e−c

−
lim P(Gm is k-connected) = e (k−1)! if cn → c
n→∞ 

1 if cn → ∞.


Proof Let
log n+(k − 1) log log n + c
p= .
n
We will prove that, in Gn,p , with edge probability p above,
(i) the expected number of vertices of degree at most k − 2 is o(1),
e−c
(ii) the expected number of vertices of degree k − 1 is, approximately (k−1)! .

We have

E(number of vertices of degree t ≤ k − 1)

nt (log n)t e−c
 
n−1 t
=n p (1 − p)n−1−t ≈ n t
t t! n n(log n)k−1
and (i) and (ii) follow immediately.
The distribution of the number of vertices of degree k − 1 is asymptotically
Poisson, as may be verified by the method of moments. (See Exercise 3.3.4).
We now show that, if
A (S, T ) = T is a component of Gn,p \ S


then
 
1
P ∃S, T, |S| < k, 2 ≤ |T | ≤ (n − |S|) : A (S, T ) = o(1).
2
This implies that if δ (Gn,p ) ≥ k then Gn,p is k-connected and Theorem 4.3
follows. |T | ≥ 2 because if T = {v} then v has degree less than k.
 4.3 Exercises 69

We can assume that S is minimal and then N(T ) = S and denote s = |S|, t =
|T |. T is connected, and so it contains a tree with t − 1 edges. Also each vertex
of S is incident with an edge from S to T and so there are at least s edges
between S and T . Thus, if p = (1 + o(1)) logn n then

P(∃S, T ) ≤ o(1)+
k−1 (n−s)/2     
n n t−2 t−1 st s
∑ ∑ s t t p p (1 − p)t(n−s−t)
s=1 t=2 s
k−1 (n−s)/2  s  t
ne
≤ p−1 ∑ ∑ · (te) · p · et p ne · p · e−(n−t)p
s=1 t=2 s
k−1 (n−s)/2
≤ p−1 ∑ ∑ At Bs (4.1)
s=1 t=2

where

A = nepe−(n−t)p = e1+o(1) n−1+(t+o(t))/n log n

B = ne2t pet p = e2+o(1)tn(t+o(t))/n log n.

Now if 2 ≤ t ≤ log n then A = n−1+o(1) and B = O((log n)2 ). On the other hand,
if t > log n then we can use A ≤ n−1/3 and B ≤ n2 to see that the sum in (4.1)
is o(1).

4.3 Exercises
4.3.1 Let m = m∗1 be as in Theorem 4.2 and let em = (u, v) where u has degree
one. Let 0 < c < 1 be a positive constant. Show that w.h.p. there is no
triangle containing vertex v.
4.3.2 Let m = m∗1 as in Theorem 4.2 and let em = (u, v) where u has degree
one. Let 0 < c < 1 be a positive constant. Show that w.h.p. the degree of
v in Gm is at least c log n.
4.3.3 Suppose that n log n  m ≤ n3/2 and let d = 2m/n. Let Si (v) be the set
of vertices at distance i from vertex v. Show that w.h.p. |Si (v)| ≥ (d/2)i
for all v ∈ [n] and 1 ≤ i ≤ 32 log
log n
d.
4.3.4 Suppose that m  n log n and let d = m/n. Using the previous question,
show that w.h.p. there are at least d/2 internally vertex disjoint paths of
length at most 34 log
log n
d between any pair of vertices in Gn,m .
4.3.5 Suppose that m  n log n and let d = m/n. Suppose that we randomly
70 Connectivity
(log n) 2
color the edges of Gn,m with q colors where q  (log d)2
. Show that w.h.p.
there is a rainbow path between every pair of vertices. (A path is rainbow
if each of its edges has a different color).
4.3.6 Let Ck,k+` denote the number of connected graphs with vertex set [k] and
k + ` edges where ` → ∞ with k and ` = o(k). Use the inequality
 
n k n
Ck,k+` pk+` (1 − p)(2)−k−`+k(n−k) ≤
k k
and a careful choice of p, n to prove (see Łuczak [548]) that
r p !`/2
k3 e + O( `/k)
Ck,k+` ≤ kk+(3`−1)/2 .
` 12`

4.3.7 Let Gn,n,p be the random bipartite graph with vertex bi-partition V =
(A, B), A = [1, n], B = [n + 1, 2n] in which each of the n2 possible edges
appears independently with probability p. Let p = log n+ωn , where ω →
∞. Show that w.h.p. Gn,n,p is connected.

4.4 Notes
Disjoint paths
Being k-connected means that we can find disjoint paths between any two sets
of vertices A = {a1 , a2 , . . . , ak } and B = {b1 , b2 , . . . , bk }. In this statement there
is no control over the endpoints of the paths i.e. we cannot specify a path
from ai to bi for i = 1, 2, . . . , k. Specifying the endpoints leads to the notion
of linkedness. Broder, Frieze, Suen and Upfal [165] proved that when we are
above the connectivity threshold, we can w.h.p. link any two k-sets by edge
disjoint paths, provided some natural restrictions apply. The result is optimal
up to constants. Broder, Frieze, Suen and Upfal [164] considered the case of
vertex disjoint paths. Frieze and Zhao [359] considered the edge disjoint path
version in random regular graphs.

Rainbow Connection
The rainbow connection rc(G) of a connected graph G is the minimum number
of colors needed to color the edges of G so that there is a rainbow path be-
tween every
p pair of vertices. Caro, Lev, Roditty, Tuza and Yuster [173] proved
that p = log n/n is the sharp threshold for the property rc(G) ≤ 2. This
was sharpened to a hitting time result by Heckel and Riordan [410]. He and
 4.4 Notes 71

Liang [409] further studied the rainbow connection of random graphs. Specif-
ically, they obtain a threshold for the property rc(G) ≤ d where d is constant.
Frieze and Tsourakakis [358] studied the rainbow connection of G = G(n, p) at
the connectivity threshold p = log n+ω
n where ω → ∞ and ω = o(log n). They
showed that w.h.p. rc(G) is asymptotically equal to max {diam(G), Z1 (G)},
where Z1 is the number of vertices of degree one.
 5
Small Subgraphs

Graph theory is replete with theorems stating conditions for the existence of a
subgraph H in a larger graph G. For example
Turán’s theorem [729] states that
2
a graph with n vertices and more than 1 − 1r n2 edges must contain a copy of
Kr+1 . In this chapter we see instead how many random edges are required to
have a particular fixed size subgraph w.h.p. In addition, we will consider the
distribution of the number of copies.

5.1 Thresholds
In this section we will look for a threshold for the appearance of any fixed
graph H, with vH = |V (H)| vertices and eH = |E(H)| edges. The property that
a random graph contains H as a subgraph is clearly monotone increasing. It
is also transparent that ”denser” graphs appear in a random graph ”later” than
”sparser” ones. More precisely, denote by
eH
d(H) = , (5.1)
vH
the density of a graph H. Notice that 2d(H) is the average vertex degree in H.
We begin with the analysis of the asymptotic behavior of the expected number
of copies of H in the random graph Gn,p .

Lemma 5.1 Let XH denote the number of copies of H in Gn,p .

 
n vH ! eH
E XH = p ,
vH aut(H)

where aut(H) is the number of automorphisms of H.

Proof The complete graph on n vertices Kn contains vnH aH distinct copies

of H, where aH is the number of copies of H in KvH . Thus

 
n
E XH = aH peH ,
vH

72
 5.1 Thresholds 73

and all we need to show is that

aH × aut(H) = vH !.
Each permutation σ of [vH ] = {1, 2, . . . , vH } defines a unique copy of H as
follows: A copy of H corresponds to a set of eH edges of KvH . The copy Hσ
corresponding to σ has edges {(xσ (i) , yσ (i) ) : 1 ≤ i ≤ eH }, where {(x j , y j ) : 1 ≤
j ≤ eH } is some fixed copy of H in KvH . But Hσ = Hτσ if and only if for each
i there is j such that (xτσ (i) , yτσ (i) ) = (xσ ( j) , yσ ( j) ) i.e., if τ is an automorphism
of H.

 
Theorem 5.2 Let H be a fixed graph with eH > 0. Suppose p = o n−1/d(H) .
Then w.h.p. Gn,p contains no copies of H.
Proof Suppose that p = ω −1 n−1/d(H) where ω = ω(n) → ∞ as n → ∞. Then
 
n vH ! eH
E XH = p ≤ nvH ω −eH n−eH /d(H) = ω −eH .
vH aut(H)
Thus
P(XH > 0) ≤ E XH → 0 as n → ∞.

From our previous experience one would expect that when E XH → ∞ as

n → ∞ the random graph Gn,p would contain H as a subgraph w.h.p. Let us
check whether such a phenomenon holds also in this case. So consider the
case when pn1/d(H) → ∞, i.e. where p = ωn−1/d(H) and ω = ω(n) → ∞ as
n → ∞. Then for some constant cH > 0
E XH ≥ cH nvH ω eH n−eH /d(H) = cH ω eH → ∞.
However, as we will see, this is not always enough for Gn,p to contain a copy
of a given graph H w.h.p. To see this, consider the graph H given in Figure 5.1
below.

Here vH = 6 and eH = 8. Let p = n−5/7 . Now 1/d(H) = 6/8 > 5/7 and so
E XH ≈ cH n6−8×5/7 → ∞.
On the other hand, if Ĥ = K4 then
E XĤ ≤ n4−6×5/7 → 0,
74 Small Subgraphs
000
111
000
111
000
111
000
111
000
111
111111
000000
111111 0111111
000
111
1000000
000000
000000 1
111111 0111111
000000
111111
000000
000000
111111 0
1
0000000
111111
111111
000000 1
0111111
1000000
000000
111111
000000 1
111111 0
1000000
111111
000000
111111
000000
111111 0
0000000
111111
000000
111111 1
0111111
1000000
000000
111111
111111 0
1000000
111111
000000 1
111111
000000 0111111
000000
000
111
111111
000000 0111111
1
0
000000
111111
000000000
111
000
111
111111
000000 11
0111111
000000000
111
000
111
000
11111111111111111
00000000000000
0
1 000
111
000
111
1111111
00000001
000
111
1111111 1111111
0
0000000000
111
0000000
000
111
1111111
0000000
1111111
0
1
0000000
1111111
0
1
0000000000
111
00000000
1111111
0000000
1111111 0
0000000
1111111
1
0000000
1111111 0000000
1111111
1
0
0000000
1111111
1
0000000
1111111
00000001
1111111 0
0000000
1111111
0
1111111
0000000
0000000
1111111
1
1111111
0
0000000
1
1111111
0000000
000000010
1111111
0000000
1111111
0000000
1111111 0
0000000
1111111
1
0
1111111
0000000
0000000
1111111
1
0
1111111
0000000
1111111
0000000
1
1111111
0
0000000
1
000
111
000
111
000
111
000
111
000
111 00000000
11111111
00011111111
111 00000000
00000000
11111111 000
111
00000000
11111111 000
111
000
111
00000000 1111111111
11111111
00000000
11111111 000
111
0000000000
00000000
11111111
000
111 000
111
0000000000
1111111111
000
111
00000000
11111111 0000000000
1111111111
000
111
0000000000
1111111111
000
111
000
111
000
111
000
111

Figure 5.1 A Kite

and so w.h.p. there are no copies of Ĥ and hence no copies of H.

The reason for such ”strange” behavior is quite simple. Our graph H is in
fact not balanced, since its overall density is smaller than the density of one
of its subgraphs, i.e., of Ĥ = K4 . So we need to introduce another density
characteristic of graphs, namely the maximum subgraph density defined as
follows:
m(H) = max{d(K) : K ⊆ H}. (5.2)

A graph H is balanced if m(H) = d(H). It is strictly balanced if d(H) > d(K)

for all proper subgraphs K ⊂ H.
Now we are ready to determine the threshold for the existence of a copy
of H in Gn,p . Erdős and Rényi [280] proved this result for balanced graphs.
The threshold for any graph H was first found by Bollobás in [124] and an
alternative, deterministic argument to derive the threshold was presented in
[480]. A simple proof, given here, is due to Ruciński and Vince [677].

Theorem 5.3 Let H be a fixed graph with eH > 0. Then

(
0 if pn1/m(H) → 0
lim P(H ⊆ Gn,p ) =
n→∞ 1 if pn1/m(H) → ∞.

Proof Let ω = ω(n) → ∞ as n → ∞. The first statement follows from Theo-

rem 5.2. Notice, that if we choose Ĥ to be a subgraph of H with d(Ĥ) = m(H)
(such a subgraph always exists since we do not exclude Ĥ = H), then p =
ω −1 n−1/d(Ĥ) implies that E XĤ → 0. Therefore, w.h.p. Gn,p contains no copies
of Ĥ, and so it does not contain H as well.
 5.1 Thresholds 75

To prove the second statement we use the Second Moment Method. Sup-
pose now that p = ωn−1/m(H) . Denote by H1 , H2 , . . . , Ht all copies of H in the
complete graph on {1, 2, . . . , n}. Note that
 
n vH !
t= , (5.3)
vH aut(H)
where aut(H) is the number of automorphisms of H. For i = 1, 2, . . . ,t let
(
1 if Hi ⊆ Gn,p ,
Ii =
0 otherwise.

Let XH = ∑ti=1 Ii . Then

t t t t
Var XH = ∑ ∑ Cov(Ii , I j ) = ∑ ∑ (E(Ii I j ) − (E Ii )(E I j ))
i=1 j=1 i=1 j=1
t t
= ∑ ∑ (P(Ii = 1, I j = 1) − P(Ii = 1) P(I j = 1))
i=1 j=1
t t
= ∑ ∑ P(Ii = 1, I j = 1) − p2eH .


i=1 j=1

Observe that random variables Ii and I j are independent iff Hi and H j are edge
disjoint. In this case Cov(Ii , I j ) = 0 and such terms vanish from the above
summation. Therefore we consider only pairs (Hi , H j ) with Hi ∩ H j = K , for
some graph K with eK > 0. So,


Var XH = O ∑ n2vH −vK p2eH −eK − p2eH
K⊆H,eK >0
 
= O n2vH p2eH ∑ n−vK p−eK .
K⊆H,eK >0

On the other hand,

 
n vH ! eH
E XH = p = Ω (nvH peH ) ,
vH aut(H)
thus, by Lemma 21.4,
!
Var XH
P(XH = 0) ≤ =O ∑ n−vK p−eK
(E XH )2 K⊆H,eK >0
 eK !
1
=O ∑
K⊆H,eK >0 ωn1/d(K)−1/m(H)
= o(1).
76 Small Subgraphs

Hence w.h.p., the random graph Gn,p contains a copy of the subgraph H when
pn1/m(H) → ∞.

5.2 Asymptotic Distributions

We will now study the asymptotic distribution of the number XH of copies
of a fixed graph H in Gn,p . We start at the threshold, so we assume that
npm(H) → c, c > 0, where m(H) denotes as before, the maximum subgraph
density of H. Now, if H is not balanced, i.e., its maximum subgraph density
exceeds the density of H, then E XH → ∞ as n → ∞, and one can show that
there is a sequence of numbers an , increasing with n, such that the asymp-
totic distribution of XH /an coincides with the distribution of a random variable
counting the number of copies of a subgraph K of H for which m(H) = d(K).
Note that K is itself a balanced graph. However the asymptotic distribution of
balanced graphs on the threshold, although computable, cannot be given in a
closed form. The situation changes dramatically if we assume that the graph
H whose copies in Gn,p we want to count is strictly balanced, i.e., when for
every proper subgraph K of H, d(K) < d(H) = m(H).
The following result is due to Bollobás [124], and Karoński and
Ruciński [479].

Theorem 5.4 If H is a strictly balanced graph and npm(H) → c,

D
c > 0, then XH → Po(λ ), as n → ∞, where λ = cvH /aut(H).
Proof Denote, as before, by H1 , H2 , . . . , Ht all copies of H in the complete
graph on {1, 2, . . . , n}. For i = 1, 2, . . . ,t, let
(
1 if Hi ⊆ Gn,p
IHi =
0 otherwise

Then XH = ∑ti=1 IHi and the kth factorial moment of XH , k = 1, 2 . . .,

E(XH )k = E[XH (XH − 1) · · · (XH − k + 1)],

can be written as

E(XH )k = ∑ P(IHi1 = 1, IHi2 = 1, . . . , IHik = 1)

i1 ,i2 ,...,ik

= Dk + Dk ,

where the summation is taken over all k-element sequences of distinct indices
 5.2 Asymptotic Distributions 77

i j from {1, 2, . . . ,t}, while Dk and Dk denote the partial sums taken over all (or-
dered) k tuples of copies of H which are, respectively, pairwise vertex disjoint
(Dk ) and not all pairwise vertex disjoint (Dk ). Now, observe that
Dk = ∑ P(IHi1 = 1) P(IHi2 = 1) · · · P(IHik = 1)
i1 ,i2 ,...,ik
 
n
= (aH peH )k
vH , vH , . . . , vH
≈ (E XH )k .

So assuming that npd(H) = npm(H) → c as n → ∞,

 v k
cH
Dk ≈ . (5.4)
aut(H)
On the other hand we will show that
Dk → 0 as n → ∞. (5.5)
Consider the family Fk of all (mutually non-isomorphic) graphs obtained
by taking unions of k not all pairwise vertex disjoint copies of the graph H.
Suppose F ∈ Fk has vF vertices (vH ≤ vF ≤ kvH − 1) and eF edges, and let
d(F) = eF /vF be its density. To prove that (5.5) holds we need the following
Lemma.
Lemma 5.5 If F ∈ Fk then d(F) > m(H).
Proof Define
fF = m(H)vF − eF . (5.6)
We will show (by induction on k ≥ 2) that fF < 0 for all F ∈ Fk . First note that
fH = 0 and that fK > 0 for every proper subgraph K of H, since H is strictly
balanced. Notice also that the function f is modular, i.e., for any two graphs
F1 and F2 ,
fF1 ∪F2 = fF1 + fF2 − fF1 ∩F2 . (5.7)
Assume that the copies of H composing F are numbered in such a way that
Hi1 ∩ Hi2 6= 0.
/ If F = Hi1 ∪ Hi2 then (5.6) and fH1 = fH2 = 0 implies
fHi1 ∪Hi2 = − fHi1 ∩Hi2 < 0.

For arbitrary k ≥ 3, let F 0 = k−1 0

j=1 Hi j and K = F ∩ Hik . Then by the inductive
S

assumption we have fF 0 < 0 while fK ≥ 0 since K is a subgraph of H (in

extreme cases K can be H itself or an empty graph). Therefore
fF = fF 0 + fHik − fK = fF 0 − fK < 0,
78 Small Subgraphs

which completes the induction and implies that d(F) > m(H).
Let CF be the number of sequences Hi1 , Hi2 , . . . , Hik of k distinct copies of
H, such that
k k
Hi j ∼
[
[
V Hi j = {1, 2, . . . , vF } and = F.
j=1 j=1

Then, by Lemma 5.5,



n
Dk = ∑ vF CF peF = O(nvF peF )
F∈Fk
 v(F) 
= O npd(F) = o(1),

and so (5.5) holds.

Summarizing,
k
cvH

E(XH )k ≈ ,
aut(H)
and the theorem follows by the Method of Moments (see Theorem 21.11).
The following theorem describes the asymptotic behavior of the number of
copies of a graph H in Gn,p past the threshold for the existence of a copy of
H. It holds regardless of whether or not H is balanced or strictly balanced. We
state the theorem but we do not supply a proof (see Ruciński [676]).

Theorem 5.6 Let H be a fixed (not-empty) graph. If npm(H) → ∞ and n2 (1 −

D
p) → ∞, then (XH − E XH )/(Var XH )1/2 → N(0, 1), as n → ∞

5.3 Exercises
5.3.1 Draw a graph which is : (a) balanced but not strictly balanced, (b) un-
balanced.
5.3.2 Are the small graphs listed below, balanced or unbalanced: (a) a tree, (b)
a cycle, (c) a complete graph, (d) a regular graph, (d) the Petersen graph,
(e) a graph composed of a complete graph on 4 vertices and a triangle,
sharing exactly one vertex.
5.3.3 Determine (directly, not from the statement of Theorem 5.3) thresholds
p̂ for Gn,p ⊇ G, for graphs listed in exercise (ii). Do the same for the
thresholds of G in Gn,m .
 5.4 Notes 79

5.3.4 For a graph G a balanced extension of G is a graph F, such that G ⊆

F and m(F) = d(F) = m(G). Applying the result of Győri, Rothschild
and Ruciński [397] that every graph has a balanced extension, deduce
Bollobás’s result (Theorem 5.3) from that of Erdős and Rényi (threshold
for balanced graphs).
5.3.5 Let F be a graph obtained by taking a union of triangles such that not
every pair of them is vertex-disjoint, Show (by induction) that eF > vF .
5.3.6 Let fF be a graph function defined as

fF = a vF + b eF ,

where a, b are constants, while vF and eF denote, respectively, the num-

ber of vertices and edges of a graph F. Show that the function fF is
modular.
5.3.7 Determine (directly, using exercise (v)) when the random variable count-
ing the number of copies of a triangle in Gn,p has asymptotically the
Poisson distribution.
5.3.8 Let Xe be the number of isolated edges (edge-components) in Gn,p and
let
ω(n) = 2pn − log n − log log n.

Prove that
(
0 if p  n−2 or ω(n) → ∞
P(Xe > 0) →
1 if p  n−2 and ω(n) → ∞.

5.3.9 Determine when the random variable Xe defined in exercise (vii) has
asymptotically the Poisson distribution.
5.3.10 Use Janson’s inequality, Theorem 22.13, to prove (5.8) below.

5.4 Notes
Distributional Questions
In 1982 Barbour [58] adapted the Stein–Chen technique for obtaining esti-
mates of the rate of convergence to the Poisson and the normal distribution
(see Section 21.3 or [59]) to random graphs. The method was next applied by
Karoński and Ruciński [481] to prove the convergence results for semi-induced
graph properties of random graphs.
Barbour, Karoński and Ruciński [61] used the original Stein’s method for
normal approximation to prove a general central limit theorem for the wide
80 Small Subgraphs

class of decomposable random variables. Their result is illustrated by a variety

of applications to random graphs. For example, one can deduce from it the
asymptotic distribution of the number of k-vertex tree-components in Gn,p , as
well as of the number of vertices of fixed degree d in Gn,p (in fact, Theorem
3.2 is a direct consequence of the last result).
Barbour, Janson, Karoński and Ruciński [60] studied the number Xk of max-
imal complete subgraphs (cliques) of a given fixed size k ≥ 2 in the random
graph Gn,p . They show that if the edge probability p = p(n) is such that the
E Xk tends to a finite constant λ as n → ∞, then Xk tends in distribution to the
Poisson random variable with the expectation λ . When its expectation tends
to infinity, Xk converges in distribution to a random variable which is normally
distributed. Poisson convergence was proved using Stein–Chen method, while
for the proof of the normal part, different methods for different ranges of p
were used such as the first projection method or martingale limit theorem (for
details of these methods see Chapter 6 of Janson, Łuczak and Ruciński [441]).
Svante Janson in an a sequence of papers [424],[425], [426], [429] (see also
[442]) developed or accommodated various methods to establish asymptotic
normality of various numerical random graph characteristics. In particular, in
[425] he established the normal convergence by higher semi-invariants of sums
of dependent random variables with direct applications to random graphs. In
[426] he proved a functional limit theorem for subgraph count statistics in
random graphs (see also [442]).
In 1997 Janson [424] answered the question posed by Paul Erdős: What is
the length Yn of the first cycle appearing in the random graph process Gm ? He
proved that
1
Z 1
2 /4 √
lim P(Yn = j) = t j−1 et/2+t 1 − t dt, for every j ≥ 3.
n→∞ 2 0

Tails of Subgraph Counts in Gn,p .

Often one needs exponentially small bounds for the probability that XH devi-
ates from its expectation. In 1990 Janson [427] showed that for fixed ε ∈ (0, 1],
P (XH ≤ (1 − ε)EXH ) = exp {−Θ(ΦH )} , (5.8)
where ΦH = minK⊆H:eK >0 nvK peK .
The upper tail P (XH ≥ (1 + ε)EXH ) proved to be much more elusive. To
simplify the results, let us assume that ε is fixed, and p is above the exis-
tence threshold, that is, p  n−1/m(H) , but small enough to make sure that
(1 + ε)EXH is at most the number of copies of H in Kn .
Given a graph G, let ∆G be the maximum degree of G and αG∗ the fractional
 5.4 Notes 81

independence number of G, defined as the maximum of ∑v∈V (G) w(v) over all
functions w : V (G) → [0, 1] satisfying w(u) + w(v) ≤ 1 for every uv ∈ E(G).
In 2004, Janson, Oleszkiewicz and Ruciński [439] proved that
exp {−O(MH log(1/p))} ≤ P(XH ≥ (1 + ε)EXH ) ≤ exp {−Ω(MH )} , (5.9)
where the implicit constants in (5.9) may depend on ε, and
∗
(
minK⊆H (nvK peK )1/αK , if n−1/m(H) ≤ p ≤ n−1/∆H ,
MH =
n2 p∆H , if p ≥ n−1/∆H .

For example, if H is k-regular, then MH = n2 pk for every p.

The logarithms of the upper and lower bounds in (5.9) differ by a multiplica-
tive factor log(1/p). In 2011, DeMarco and Kahn formulated the following
plausible conjecture (stated in [236] for ε = 1).
Conjecture: For any H and ε > 0,
P (XH ≥ (1 + ε)EXH ) = exp (−Θ (min {ΦH , MH log(1/p)})) . (5.10)
A careful look reveals that, when ∆H ≥ 2, the minimum in (5.10) is only at-
tained by ΦH in a tiny range above the existence threshold (when p ≤ n−1/m(H) (log n)aH
for some aH > 0). In 2018, Šileikis and Warnke [706] found counterexample-
graphs (all balanced but not strictly balanced) which violate (5.10) close to the
threshold, and conjectured that (5.10) should hold under the stronger assump-
tion p ≥ n−1/mH +δ .
DeMarco and Kahn [236] proved (5.10) for cliques H = Kk , k = 3, 4, . . . .
Adamczak and Wolff [6] proved a polynomial concentration inequality which
− k−2
confirms (5.10) for any cycle H = Ck , k = 3, 4, . . . and p ≥ n 2(k−1) . Moreover,
Lubetzky and Zhao [546], via a large deviations framework of Chatterjee and
Dembo [176], showed that (5.10) holds for any H and p ≥ n−α for a sufficiently
small constant α > 0. For more recent developments see [196], where it is
shown that one can take α > 1/6∆H .
 6
Spanning Subgraphs

The previous chapter dealt with the existence of small subgraphs of a fixed size.
In this chapter we concern ourselves with the existence of large subgraphs,
most notably perfect matchings and Hamilton Cycles. The celebrated theo-
rems of Hall and Tutte give necessary and sufficient conditions for a bipartite
and arbitrary graph respectively to contain a perfect matching. Hall’s theorem
in particular can be used to establish that the threshold for having a perfect
matching in a random bipartite graph can be identified with that of having no
isolated vertices.
For general graphs we view a perfect matching as half a Hamilton cycle and
prove thresholds for the existence of perfect matchings and Hamilton cycles in
a similar way.
Having dealt with perfect matchings and Hamilton cycles, we turn our at-
tention to long paths in sparse random graphs, i.e. in those where we expect a
linear number of edges. We then analyse a simple greedy matching algorithm
using differential equations.
We then consider random subgraphs of some fixed graph G, as opposed to
random subgraphs of Kn . We give sufficient conditions for the existence of long
paths and cycles.
We finally consider the existence of arbitrary spanning subgraphs H where
we bound the maximum degree ∆(H).

6.1 Perfect Matchings

Before we move to the problem of the existence of a perfect matching, i.e., a
collection of independent edges covering all of the vertices of a graph, in our
main object of study, the random graph Gn,p , we will analyse the same problem
in a random bipartite graph. This problem is much simpler than the respective
one for Gn,p , but provides a general approach to finding a perfect matching in
a random graph.

82
 6.1 Perfect Matchings 83

Bipartite Graphs
Let Gn,n,p be the random bipartite graph with vertex bi-partition V = (A, B),
A = [1, n], B = [n + 1, 2n] in which each of the n2 possible edges appears inde-
pendently with probability p. The following theorem was first proved by Erdős
and Rényi [282].

Theorem 6.1 Let ω = ω(n), c > 0 be a constant, and p = log n+ω

n . Then

0

 if ω → −∞
−2e−c
lim P(Gn,n,p has a perfect matching) = e if ω → c
n→∞ 
if ω → ∞.

1

Moreover,
lim P(Gn,n,p has a perfect matching) = lim P(δ (Gn,n,p ) ≥ 1).
n→∞ n→∞

Proof We will use Hall’s condition for the existence of a perfect matching in
a bipartite graph. It states that a bipartite graph contains a perfect matching if
and only if the following condition is satisfied:
∀S ⊆ A, |N(S)| ≥ |S|, (6.1)
where for a set of vertices S, N(S) denotes the set of neighbors of S.
It is convenient to replace (6.1) by
n
∀S ⊆ A, |S| ≤ , |N(S)| ≥ |S|, (6.2)
2
n
∀T ⊆ B, |T | ≤ , |N(T )| ≥ |T |. (6.3)
2
This is because if |S| > n/2 and |N(S)| < |S| then T = B \ N(S) will violate
(6.3).
Now we can restrict our attention to S, T satisfying (a) |S| = |T | + 1 and (b)
each vertex in T has at least 2 neighbors in S. Take a pair S, T with |S| + |T | as
small as possible. If the minimum degree δ ≥ 1 then |S| ≥ 2.
(i) If |S| > |T |+1, we can remove |S|−|T |−1 vertices from |S| – contradiction.
(ii) Suppose ∃w ∈ T such that w has less than 2 neighbors in S. Remove w and
its (unique) neighbor in |S| – contradiction..
It follows that
P(∃v : v is isolated) ≤ P(6 ∃ a perfect matching)
≤ P(∃v : v is isolated) + 2 P(∃S ⊆ A, T ⊆ B, 2 ≤ k = |S| ≤ n/2,
84 Spanning Subgraphs

|T | = k − 1, N(S) ⊆ T and e(S : T ) ≥ 2k − 2).

Here e(S : T ) denotes the number of edges between S and T , and e(S : T ) cab
be assumed to be at least 2k − 2, because of (b) above.
Suppose now that p = lognn+c for some constant c. Then let Y denote the
number of sets S and T not satisfying the conditions (6.2), (6.3). Then
n/2    
n n k(k − 1) 2k−2
EY ≤ 2 ∑ p (1 − p)k(n−k)
k=2 k k − 1 2k − 2
n/2 
ne k−1 ke(log n + c) 2k−2 −npk(1−k/n)
   
ne k
≤2∑ e
k=2 k k−1 2n
!k
n/2
eO(1) nk/n (log n)2
≤ ∑n
k=2 n
n/2
= ∑ uk .
k=2

Case 1: 2 ≤ k ≤ n3/4 .

uk = n((eO(1) n−1 log n)2 )k .

So
n3/4  
1
∑ uk = O n1−o(1)
.
k=2

Case 2: n3/4 < k ≤ n/2.

uk ≤ n1−k(1/2−o(1))
So
n/2  3/4 
u
∑ k = O n−n /3 .
n3/4

So
P(6 ∃ a perfect matching) = P(∃ isolated vertex) + o(1).
Let X0 denote the number of isolated vertices in Gn,n,p . Then
E X0 = 2n(1 − p)n ≈ 2e−c .
By previously used techniques we have
−c
P(X0 = 0) ≈ e−2e
 6.1 Perfect Matchings 85

To prove the case for |ω| → ∞ we can use monotonicity and (1.7) and the fact
−2c −2c
that e−e → 0 if c → −∞ and e−e → 1 if c → ∞.

Non-Bipartite Graphs
We now consider Gn,p . We could try to replace Hall’s theorem by Tutte’s the-
orem. A proof along these lines was given by Erdős and Rényi [283]. We can
however get away with a simpler approach based on simple expansion proper-
ties of Gn,p . The proof here can be traced back to Bollobás and Frieze [142].
Theorem 6.2 Let ω = ω(n), c > 0 be a constant, and let p = log n+c
n
n
. Then

0

 if cn → −∞
−e−c
lim
n→∞
P(G n,p has a perfect matching) = e if cn → c
n even

if c → ∞.

1
n

Moreover,

lim P(Gn,p has a perfect matching) = lim P(δ (Gn,p ) ≥ 1).

n→∞ n→∞

Proof We will for convenience only consider the case where cn = ω → ∞ and
ω = o(log n). If cn → −∞ then there are isolated vertices, w.h.p. and our proof
can easily be modified to handle the case cn → c.
Our combinatorial tool that replaces Tutte’s theorem is the following: We
say that a matching M isolates a vertex v if no edge of M contains v.
For a graph G we let

µ(G) = max {|M| : M is a matching in G} . (6.4)

Let G = (V, E) be a graph without a perfect matching i.e. µ(G) < b|V |/2c.
Fix v ∈ V and suppose that M is a maximum matching that isolates v. Let
S0 (v, M) = {u 6= v : M isolates u}. If u ∈ S0 (v, M) and e = {x, y} ∈ M and f =
{u, x} ∈ E then flipping e, f replaces M by M 0 = M + f − e. Here e is flipped-
out. Note that y ∈ S0 (v, M 0 ).
Now fix a maximum matching M that isolates v and let

S0 (v, M 0 )
[
A(v, M) =
M0

where we take the union over M 0 obtained from M by a sequence of flips.

Lemma 6.3 Let G be a graph without a perfect matching and let M be a
maximum matching and v be a vertex isolated by M. Then |NG (A(v, M))| <
|A(v, M)|.
86 Spanning Subgraphs

Proof Suppose that x ∈ NG (A(v, M)) and that f = {u, x} ∈ E where u ∈

A(v, M). Now there exists y such that e = {x, y} ∈ M, else x ∈ S0 (M) ⊆ A(v, M).
We claim that y ∈ A(v, M) and this will prove the lemma. Since then, every
neighbor of A(v, M) is the neighbor via an edge of M.
Suppose that y ∈/ A(v, M). Let M 0 be a maximum matching that (i) isolates
u and (ii) is obtainable from M by a sequence of flips. Now e ∈ M 0 because
if e has been flipped out then either x or y is placed in A(v, M). But then we
can do another flip with M 0 , e and the edge f = {u, x}, placing y ∈ A(v, M),
contradiction.
We now change notation and write A(v) in place of A(v, M), understanding
that there is some maximum matching that isolates v. Note that if u ∈ A(v)
then there is some maximum matching that isolates u and so A(u) is well-
defined. Furthermore, it always that case that if v is isolated by some maximum
matching and u ∈ A(v) then µ(G + {u, v}) = µ(G) + 1.
Now let
log n + θ log log n + ω
p=
n
where θ ≥ 0 is a fixed integer and ω → ∞ and ω = o(log log n).
We have introduced θ so that we can use some of the following results for
the Hamilton cycle problem.
We write
Gn,p = Gn,p1 ∪ Gn,p2 ,
where
log n + θ log log n + ω/2
p1 =
n
and
ω
1 − p = (1 − p1 )(1 − p2 ) so that p2 ≈ .
2n
Note that Theorem 4.3 implies:
The minimum degree in Gn,p1 is at least θ + 1 w.h.p. (6.5)
We consider a process where we add the edges of Gn,p2 one at a time to
Gn,p1 . We want to argue that if the current graph does not have a perfect match-
ing then there is a good chance that adding such an edge {x, y} will increase the
size of a largest matching. This will happen if y ∈ A(x). If we know that w.h.p.
every set S for which |NGn,p1 (S)| < |S| satisfies |S| ≥ αn for some constant
α > 0, then
αn

− i α2
P(y ∈ A(x)) ≥ 2 n
≥ , (6.6)
2
2
 6.1 Perfect Matchings 87

provided i = O(n).
This is because

the edges we add will be uniformly random and there will
be at least αn2 edges {x, y} where y ∈ A(x). Here given an initial x we can
include edges {x0 , y0 } where x0 ∈ A(x) and y0 ∈ A(x0 ). We have subtracted i to
account for not re-using edges in f1 , f2 , . . . , fi−1 .
In the light of this we now argue that sets S, with |NGn,p1 (S)| < (1 + θ )|S|
are w.h.p. of size Ω(n).
n
Lemma 6.4 Let M = 100(θ +7). W.h.p. S ⊆ [n], |S| ≤ 2e(θ +5)M implies |N(S)| ≥
(θ + 1)|S|, where N(S) = NGn,p1 (S).

Proof Let a vertex of graph G1 = Gn,p1 be large if its degree is at least λ =

log n
100 , and small otherwise. Denote by LARGE and SMALL, the set of large and
small vertices in G1 , respectively.

Claim 1 W.h.p. if v, w ∈ SMALL then dist(v, w) ≥ 5.

Proof If v, w are small and connected by a short path P, then v, w will have
few neighbors outside P and conditional on P existing, v having few neighbors
outside P is independent of w having few neighbors outside P. Hence,
P(∃v, w ∈ SMALL in Gn,p1 such that dist(v, w) < 5)
  3 ! !2
λ  
n l l+1 n k n−k−5
≤ ∑ n p1 ∑ p1 (1 − p1 )
2 l=0 k=0 k
!2
4
λ
(log n)k (log n)(θ +1)/100 · e−ω/2
≤ n(log n) ∑ ·
k=0 k! n log n
!2
λ (log n)(θ +1)/100 · e−ω/2
4 (log n)
≤ 2n(log n) · (6.7)
λ! n log n
!
(log n)O(1) 2 log n
=O (100e) 100
n
= O(n−3/4 )
= o(1).
l l uk+1


The bound in (6.7) holds since l! ≥ e and uk > 100 for k ≤ l, where

(log n)k (log n)(θ +1)/100 · e−ω/2

uk = · .
k! n log n
88 Spanning Subgraphs

Claim 2 W.h.p. Gn,p1 does not have a 4-cycle containing a small vertex.
Proof
P(∃ a 4-cycle containing a small vertex )
(log n)/100  
4 4 n−4 k
≤ 4n p1 ∑ p1 (1 − p1 )n−4−k
k=0 k
≤ n−3/4 (log n)4
= o(1).

n |S| log n
Claim 3 W.h.p. in Gn,p1 for every S ⊆ [n], |S| ≤ 2eM , e(S) < M .

Proof
 
n |S| log n
P ∃|S| ≤ and e(S) ≥
2eM M
n/2eM   s

n 2 s log n/M
≤ ∑ p1
s=log n/M
s s log n/M
 !log n/M s
n/2eM 1+o(1)
ne Me s
≤ ∑  
s=log n/M
s 2n
n/2eM   s
s −1+log n/M 1+o(1) log n/M
≤ ∑ · (Me )
s=log n/M
n
= o(1).

Claim 4 Let M be as in Claim 3. Then, w.h.p. in Gn,p1 , if S ⊆ LARGE, |S| ≤

n
2e(θ +5)M then |N(S)| ≥ (θ + 4)|S|.
Proof Let T = N(S), s = |S|,t = |T |. Then we have
|S| log n |S| log n 2|S| log n
e(S ∪ T ) ≥ e(S, T ) ≥ − 2e(S) ≥ − .
100 100 M
n
Then if |T | ≤ (θ + 4)|S| we have |S ∪ T | ≤ (θ + 5)|S| ≤ 2eM and
 
|S ∪ T | 1 2 |S ∪ T | log n
e(S ∪ T ) ≥ − log n = .
θ + 5 100 M M
 6.2 Hamilton Cycles 89

This contradicts Claim 3.

n
We can now complete the proof of Lemma 6.4. Let |S| ≤ 2e(θ +5)M and as-
sume that Gn,p1 has minimum degree at least θ + 1.
Let S1 = S ∩ SMALL and S2 = S \ S1 . Then

|N(S)|
≥ |N(S1 )| + |N(S2 )| − |N(S1 ) ∩ S2 | − |N(S2 ) ∩ S1 | − |N(S1 ) ∩ N(S2 )|
≥ |N(S1 )| + |N(S2 )| − |S2 | − |N(S2 ) ∩ S1 | − |N(S1 ) ∩ N(S2 )|.
But Claim 1 and Claim 2 and minimum degree at least θ + 1 imply that

|N(S1 )| ≥ (θ + 1)|S1 |, |N(S2 ) ∩ S1 | ≤ min{|S1 |, |S2 |},

|N(S1 ) ∩ N(S2 )| ≤ |S2 |.
So, from this and Claim 4 we obtain
|N(S)| ≥ (θ + 1)|S1 | + (θ + 4)|S2 | − 3|S2 | = (θ + 1)|S|.

We now go back to the proof of Theorem 6.2 for the case c = ω → ∞. Let
the edges of Gn,p2 be { f1 , f2 , . . . , fs } in random order, where s ≈ ωn/4. Let
G0 = Gn,p1 and Gi = Gn,p1 + { f1 , f2 , . . . , fi } for i ≥ 1. It follows from Lemmas
6.3 and 6.4 that with µ(G) as in (6.4), and if µ(Gi ) < n/2 then, assuming Gn,p1
1
has the expansion claimed in Lemma 6.4, with θ = 0 and α = 10eM ,
α2
P(µ(Gi+1 ) ≥ µ(Gi ) + 1 | f1 , f2 , . . . , fi ) ≥ , (6.8)
2
see (6.6).
It follows that

P(Gn,p does not have a perfect matching) ≤

o(1) + P(Bin(s, α 2 /2) < n/2) = o(1).
We have used the notion of dominance, see Section 22.9 in order to use the
binomial distribution in the above inequality.

6.2 Hamilton Cycles

This was a difficult question left open in [280]. A breakthrough came with
the result of Pósa [649]. The precise theorem given below can be credited
to Komlós and Szemerédi [504], Bollobás [130] and Ajtai, Komlós and Sze-
merédi [11].
90 Spanning Subgraphs

log n+log log n+cn

Theorem 6.5 Let p = n . Then

0

 if cn → −∞
−c
lim P(Gn,p has a Hamilton cycle) = e−e if cn → c
n→∞ 
if cn → ∞.

1

Moreover,
lim P(Gn,p has a Hamilton cycle ) = lim P(δ (Gn,p ) ≥ 2).
n→∞ n→∞
Proof We will first give a proof of the first statement under the assumption
that cn = ω → ∞ where ω = o(log log n). The proof of the second statement is
postponed to Section 6.3. Under this assumption, we have δ (Gn,p ) ≥ 2 w.h.p.,
see Theorem 4.3. The result for larger p follows by monotonicity.
We now set up the main tool, viz. Pósa’s Lemma. Let P be a path with end
points a, b, as in Figure 6.1. Suppose that b does not have a neighbor outside
of P.

P
000
111
111
000
000
111
000
111
000
111
000
111
000
111
000
111
000
111
000
111
000
111
000
111
00
11
00
11
00
11
00
11
00
11
00
11
y
000
111
000
111 000
111
000
111 000
111
000
111 000
111
000
111 00
11
00
11 00
11
00
11
0000
1111
0000
1111 000
1111111
0000 000 1111
111
000
111 0000
000
111 0000
11110001111
111
000
111 0000
1111
00000001111
111
000
1110000
1111
0000000
111
0000
1111 111
0001111
1110000 111
1111 000 1111
111
0000 1111
0001111
0000111
000 0000
1111000
1110000
1111111
000
000
0000 111
1111 0000000
0000
1111
1111
111
0000
000
000 1111
111 0000
000
111 0000
0000
11110001111
111 0000111
1111 0000111
0001111
1111111000
000
111
0000
1111
0000
1111 000
1110000
1111 000
111
000
111 0000
1111
000
111 0000
1111000
111
000
111 0000
0000
1111000
000
1110000
0000
1111000
111
0000
1111 0001111
111
000
1110000 111
1111
0000 000 0000
1111
0001111
111
0000
1111
000
111 0000111
1111
0000000 0000
1111000
1110000
1111000
111
0000 111
1111
1111 0000000
1111 000 1111
111 0000
000
111 0000
11110001111
111 0000111
1111 0000111
0001111
1111111000
000
111
0000
1111
0000 0001111
111 111
000
0000 111
1111 000 0000
1111
0001111
111 111
000
0000111
1111000 0000
1111
0000000
111
0000000
1111
0000000
111
1111 111
000
0000 0000000
1111111 1111
111
0000
000
000 0000
111 00000001111
111 0000111
1111 0000111
0001111
1111111000
0000
1111 000
1110000
0000
1111 000
111 1111
1110000
000
0000
1111
000
111
1111111
0000
1111000 00000000000111
000
000
111
000
111
0000
1111
000
111 000
111
000
111 000
111
000
111
0000 111
1111
000
111 000
1110000
1111 000
111
000
111 000
111
000
111
0000111
1111
000
111 0000
1111
000
111 00
11
000
1110000
1111
00
11 00
11
000
111
00
11
0000 111
1111
000
111 000
000
111 000
000
1110000
1111
000
111 0000
1111
000
111 000 0000
1111
000
111 000
1110000
1111
00
11
0000 11
1111 000
111
00
11
000
111 000
111 000
111 000
111 000
111 00 00
11
000
111
a 000
111 000
111 000
111 000
111
x 00
11 00
11 b

Figure 6.1 The path P

Notice that the P0 below in Figure 6.2 is a path of the same length as P, ob-
tained by a rotation with vertex a as the fixed endpoint. To be precise, suppose
that P = (a, . . . , x, y, y0 , . . . , b0 , b) and {b, x} is an edge where x is an interior
vertex of P. The path P0 = (a, . . . , x, b, b0 , . . . , y0 , y) is said to be obtained from
P by a rotation.

Now let END = END(P) denote the set of vertices v such that there exists a
path Pv from a to v such that Pv is obtained from P by a sequence of rotations
with vertex a fixed as in Figure 6.3.

Here the set END consists of all the white vertices on the path drawn below
in Figure 6.4.
 6.2 Hamilton Cycles 91
P’
111
000
000
111
000
111
000
111
000
111
000
111
000
111
000
111
000
111
111
000
000
111
000
111
y 00
11
00
11
00
11
00
11
00
11
00
11
000
111
000111
111 000
111
000
111 000
111
000
111 000
111
000
111 00
11
00
11 00
11
00
11
1111
0000 0001111
0000 111
000 1111
111
0000
000 111
000 1111
0000111
0001111
0000
1111
0000 111
0000
1111 0000000
1111 111
000 1111
000
111 0000
000
111 111
000
111
1111
0001111
0000
111
0000111
0000000111
1111
0001111
0000
000
000
111
0000
1111 000
111
1110000
1111 000
111 0000
1111
000
111 000
111 0000
1111000
1110000
1111000
111
0000
1111 111
000
1111
0001111
0000 111
0000 000 1111
111
0000
000
1111
111
0000
000 000
111 0000
1111000
1110000
1111111
000
0000 111
1111
1111 0000000
1111 000 1111
111 0000
000
111 0001111
111 0000111
11111110000111
0001111
1111000
000
111
0000
1111
0000 0001111
111 111
000
0000 111
1111 000 0000
1111
000
111 111
000
111
000 0000
1111
0000000
111
0000000
1111
0000000
111
1111 000
111
0000 1110000 0000
1111
000
111
000 1111
111 0001111
111 0000111
1111 0000111
0001111
1111111000
0000
1111 000
111
000
1111
0000
1111
0000 000
111 111
0000
000
1111
111
0000
000 000
111 00000000000111
000
111
000
0000
1111
1111 111
0001111 000
111
0000 111
1111 1111
111
0000
000 000
111 1111
0000000
1110000
1111111
000
0000
000
111
1111 000
1110000
000
111
0000 111 000
000
111
111
000 0000
1111
000
111
000
111 111
000 1111
0000
000
111
111
000 1111111
0001111
0000
0000111
00
11
111
0000111
0001111000
00
11
000
111
0000
1111 000
1110000
1111
000
111 000
111
000
111 0000
1111
000
111
000
111 000
111
000
111 0000
1111 00
11
0000000
1111000
111
00
11
000
111 000
000
111 000
1111111
111
0000
000
000
111 000
111
0000
1111 00
11 111
000
00
11
000
111 000
111 000
111 000
111 000
111 00
11 00
11
000
111
a 000
111 000
111 000
111
x 000
111 00
11 00
11 b

Figure 6.2 The path P0 obtained after a single rotation

000
111
000
111 000
111
000
111 000
111
000
111 000
111
000
111 00
11
00
11 00
11
00
11
000
111
000
111 000
111
000
111 000
111
000
111 000
111
000
111 00
11
00
11 00
11
00
11
000111
111
1111
0000 000
111
0001111 000
111
0000 111
1111 000
111
1111
111
0000 000
111
1111
0001111 000
111
0000111 00
11
0000
1111000
111 00
11
0000
1111111
000
0000
1111 000
111
0000 111
1111 0000 000
111 0000
1111
000
111
000 1111 0000000 0000
1111
0001111
111 000
111
0000111
1111 0000
1111
0000111
0001111
1111111000
0000
1111 000
111
0000000
1111
0000
1111 000
111 111
0000
000
1111
111
0000
000
1111
0000
1111
0000000
111 00000000000111
000
111
000
0000
1111
0000
1111 0001111
111 000
111
0000 111
1111 000 0000
1111
0001111
111 000
111
0000111
1111000 0000
1111
0000
1111000
111
000
1110000
1111
0000
1111000
111
1111 000
111
0000 111
0000000
1111 0000
1111
000
111
000 1111
111 00000001111
111 0000111
1111 0000111
0001111
1111111000
1111
0000 0000
1111111
000 111
000
0000 111
0000
000
111
1111 0000
1111
1111111
0000
000 111
000
1111111
0000 00000000000000
111
111
000
1111
0000 111
0000000
1111 000 1111
111
0000
000 0000000 1111
0000111
0001111
0000
0000 111
1111
0000
1111 0000000
1111 000 1111
111
000
111 0000
000
111 0000
11110001111
111
000
111 0000111
1111
00000001111
111
0001111
0000
111
0000000
000
111
0000
1111 000
111
000
1110000
1111 000
111
0000 111
1111 0000
1111
000
111
0000
1111
000
111 0000
1111000
111
0000111
1111 0000
1111000
1110000
1111000
111
000
111
000
111
0000
1111 000
111
000
1110000
1111 000
111
000 0000
1111 000
111
000
111 0000
1111 000
111
000 0000
1111 00
11
000
1110000
1111
000
111
000
111
000
111
0000 111
1111 000
000
111
000
111
000
111
000
1110000
1111 000
111
000
111 1111
000
111
000
111
0000111
000
000
111
0000
1111 0000111
00
11
0000111
11110001111
00
11
000
000
111
000
111 000
111 000
111 000
111 000
111 00
11
000
111
a 000
111 000
111 000
111 000
111 00
11 b
000
111
000
111 000
111
000
111 000
111
000
111 00
11
00
11 00
11
00
11
000
111
000
111 000
111
000
111 000
111
000
111 00
11
00
11 00
11
00
11
000111
111
0000
1111 000 000
111
1111
0000 000
111
000 1111
111 111
0000
000 111 00
11
0001111
0000
1111 00
11
0001111
1110000
1111111
000
1111
0000
1111 0001111
111 000
111
0000 111 0000
1111
000
111 111
0000000000
1110000000
111
0000
1111
0000 0001111
111 000
0000 111
1111 000 0000
1111
000
111 111
000
111
000
0000
1111
0000
1111111
000
111
000
1111
0000
1111
0000000
111
000
111
0000 111
1111 0000000
0000
1111 0000
1111
000
111
000 1111
111 0000
000
111 0000000
1111111111
0000 0000111
0001111
1111000
000
111
0000
1111
0000
1111 0001111
111 000
111
0000 111
1111 000 0000
1111
000
111 000
111
000
111
1111
0000
1111000
111
000
1110000
0000
1111000
111
1111 000
111
0000 111
0000000
0000
1111 111 0000
1111
000
111
000 1111
0000
000
111 1111111
0001111111
00001110000111
1111
0001111000
000
111
1111
0000 000
1110000
1111 111
000 0000
1111
000
111 111
00000000000000000
111
0000
1111
0000
1111 0001111
111 000
111
0000 111
1111 000 0000
1111
000
111 000
111
000
1110000
1111
1111
0000000
111
000
1110000
1111
0000
1111000
111
000
111
0000 111
1111 0000000
0000
1111 000
111 0000
1111
000
111
0000
1111
000
111 000
1110000111
1111000
1110000111
1111000
000
111
000
111
0000
1111 000
111
111
000 000
111
000
111
0000 111
1111 1111 000
111
111
0000
000 000
111
000
1110000
1111 00
11
0000000
1111 00
11
111
000
000
111
0000
1111
000
111 000
111
111
000
000
111 000
111
000
000
1111111 000
111
111
0000
000
000
111 000
111
000
1110000
1111
000
111 00
11
000
1110000
1111
00
11 00
11
111
000
00
11
000
111 000
111 000
111 000
111 0000 11
1111
000
111 00 00
11
000
111
a 000
111 000
111 000
111 000
111 00
11 00
11 b
000
111
000
111
000
111
111
000
000
111
000
111
000
111
000
111
000
111
111
000
000
111
000
111
y 00
11
00
11
00
11
00
11
00
11
00
11
000
111
000
111 000
111
000
111 000
111
000
111 000
111
000
111 00
11
00
11 00
11
00
11
1111
0000
0000
1111 0000
1111 000
111
000
111 1111
111
0000
000 000
111
000
111 0000
1111
0000
1111000
111
000
1110000
1111
0000
1111111
000
0000
1111 0000 111
1111
0000
1111 000 0000
1111
000
111
0000
1111
000
111 000
111 0000
1111000
1110000
1111000
111
0000
1111
0000
1111 0000
1111 000 1111
111
000
111 0000
000
111 0001111
111
000
111 0000111
1111
00000000000111
0001111
1111111
0000
000
000
111
0000
1111 0000 111
1111
0000
1111 000 0000
1111
000
111
0000
1111
000
111 000
111 0000
1111000
1110000
1111000
111
1111
0000
1111
0000 1111
0000 111
000 1111
111
000 0000
000
111 111
111
000
1111
0001111
0000
111
0000111
0000000111
1111
0001111
0000
000
000
111
0000
1111 0000
1111 000
111
0000 111
1111
1111
111
0000
000
1111
111
0000
000 000
111 0000
1111000
1110000
1111111
000
111
000
0000
1111
0000
1111 0000
1111 000
111 0000
1111
000
111
000 1111 000
111 0000
1111
0001111
111 000
111
0000111
1111 0000
1111
0000111
0001111
1111111000
0000
1111 0000
1111
0000
1111 000
111 0000
000
111
0000
1111
000
111 000
111 00000000000000
111
000
111
000
111
0000
1111
000
111 000
111
000
111
0000 111
1111 000
1110000
1111 000
111
000
111
000
111 000
111
000
111 0000
1111
000
111 00
11
000
1110000
1111
00
11 00
11
000
111
00
11
0000
1111
000
111 000
000
1110000
1111
000
111
000
111 000
111 0000
1111
000
111 000
1110000
1111
00
11
0000 11
1111 000
111
00
11
000
111 000
111 000
111 000
111 00 00
11
000
111
a 000
111 000
111
x 000
111 00
11 00
11 b

Figure 6.3 A sequence of rotations

Lemma 6.6 If v ∈ P \ END and v is adjacent to w ∈ END then there exists

x ∈ END such that the edge {v, x} ∈ P.
Proof Suppose to the contrary that x, y are the neighbors of v on P and that
v, x, y 6∈ END and that v is adjacent to w ∈ END. Consider the path Pw . Let {r,t}
be the neighbors of v on Pw . Now {r,t} = {x, y} because if a rotation deleted
{v, y} say then v or y becomes an endpoint. But then after a further rotation
from Pw we see that x ∈ END or y ∈ END.
92 Spanning Subgraphs
111
000 111
000 000
111 00
11 00
11
000
111
000
111 000
111
000
111 000
111
000
111 00
11
00
11 00
11
00
11
000
111
000
111 000
111
000
111 000
111
000
111 00
11
00
11 00
11
00
11
0000
1111
0000
1111 1111
0000 000
111
000
111 1111
111
0000
000 1111
0000000
111
000
1110000
1111
0000
1111000
111
000
1110000
1111
0000
1111111
000
1111
0000 0000 111
1111
0000
1111 000 1111
111
0000
0000
1111
000
111
1111
0001111
0000111
00000001111
0000111
0001111
0000111
000
1111
0000
0000
1111 0000
1111 111
000 1111
000
111 0000
000
111 0000
1111111
000
111
1111
0001111
0000
111
0000111
0000000111
1111
0001111
0000
000
000
111
0000
1111 1111
0000 111
000 1111
111
0000 1111
0001111
0000111
0000000
1111000
1110000
1111111
000
0000
1111 0000
1111
0000
1111
1111
111
0000
000
000 1111
111 0000
000
111 0000
0000
11110001111
1110000111
1111 0000111
0001111
1111111000
000
111
1111
0000
1111 0000
1111 111
000 0000
1111
000
111 0000
1111111
00000000000000000
111
0000
0000
1111
111
000
0000 111
1111 000 0000
1111
0001111
111 111
000
0000111
1111000
1111
0000
0000
1111
111
000
000
111
1111
0000
0000
1111000
111
0000
1111 1111
0000 1111
111
0000
000
000 1111
111 00000001111
1110000111
1111 0000111
0001111
1111111000
0000
1111 0000
1111 000
111
0000 111
1111
111
0000
000
0000
1111
000
111
1111
0000000
111
0000111
1111 00000000000111
000
000
111
000
111
1111
0000 0000
1111 000
111
000 0000
1111 000
111
000
111 0000
1111 000
111
0001111
0000 00
11
111
0001111
0000
000
111
1111
0000
000
111 000
111
111
000
000
1110000
1111 000
111
000
111 1111
000
111 000
111
0000111
0001111
000
111
0000
11110000000111
00
11
00001111111
00
11
000
000
111
000
111 000
111 000
111 000
111 00
11
000
111
a 000
111 000
111 000
111 00
11 b

Figure 6.4 The set END

000
111 000
111 000
111
r
000
111 00
11 t00
11
111
000
000
111 000
111
000
111 000
111
000
111 000
111
000
111 00
11
00
11 00
11
00
11
000
111
000111
111 000
111
000
111 000
111
000
111 000
111
000
111 00
11
00
11 00
11
00
11
1111
0000 0001111
0000 111
000 1111
111
0000
000 1111
0000111 1111
0001111
00001111111
0001111
0000111
000
0000
1111
0000
1111 111
0000000
1111 000
111
000
111 1111
111
0000
000 1111
0000000
111
000
111 0000
0000
1111000
111
000
1110000
0000
1111111
000
0000
1111 0001111
111
000
1110000 111
1111
0000 000 0000
1111
0001111
111
0000
1111
000
111 0000111
1111
0000000 0000
1111000
1110000
1111000
111
0000 111
1111
0000
1111 0000000
1111 000 1111
111
000
111 0000
000
111 0000
11110001111
111
000
111 0000111
1111
00000000000111
0001111
1111111
0000
000
000
111
0000
1111 0001111
111
000
1110000 111
1111
0000 000 0000
1111
0001111
111
0000
1111
000
111 0000111
1111
0000000 0000
1111000
1110000
1111000
111
0000 111
1111
1111 0000000
1111 000 1111
111 0000
000
111 0000
11110001111
111 0000111
1111 0000111
0001111
1111111000
000
111
0000
1111
0000 000
1110000
1111 111
000
111
000 0000
1111
000
111 0000
1111111
000
111
000 0000
1111
0000000
111
0000000
1111
0000000
111
1111
0000 1110000
000
1111111 111
000 1111
1111111
0000
000 1111111
0000000 0000
1111111
0001111
0000111
000
000
0000 000
1111 1110000
1111
0000
1111 000
111 1111
111
0000
000
1111
111
0000
000 0000000
111
1111111
0000 0000111
1111000
1110000111
1111000
111
000
000
111
0000
1111
000
111 000
111
000
111 000
111
000
111
0000 111
1111
000
111 000
1110000
1111 000
111
000
111 000
111
000
0000111
1111
000
111 0000
1111
000
111 00
11
0000000
1111
00
11 00
11
000
111
00
11
0000
1111
000
111 000
111
000
111 000
000
1110000
1111
000
111 0000
1111
000
111 000 0000
1111
000
111 000
1110000
1111
00
11
0000 11
1111 000
111
00
11
000
111
000
111 000
111
000
111 000
111
000
111 000
111
000
111 000
111
000
111 00
00
11 00
11
00
11
a v w

Figure 6.5 One of r,t will become an endpoint after a rotation

Corollary 6.7
|N(END)| < 2|END|.

It follows from Lemma 6.4 with θ = 1 that w.h.p. we have

1
|END| ≥ αn where α = . (6.9)
12eM

We now consider the following algorithm that searches for a Hamilton cycle
in a connected graph G. The probability p1 is above the connectivity threshold
and so Gn,p1 is connected w.h.p. Our algorithm will proceed in stages. At the
beginning of Stage k we will have a path of length k in G and we will try to
grow it by one vertex in order to reach Stage k + 1. In Stage n − 1, our aim is
simply to create a Hamilton cycle, given a Hamilton path. We start the whole
procedure with an arbitrary path of G.
Algorithm Pósa:
(a) Let P be our path at the beginning of Stage k. Let its endpoints be x0 , y0 . If
x0 or y0 have neighbors outside P then we can simply extend P to include
one of these neighbors and move to stage k + 1.
 6.2 Hamilton Cycles 93

(b) Failing this, we do a sequence of rotations with x0 as the fixed vertex until
one of two things happens: (i) We produce a path Q with an endpoint y
that has a neighbor outside of Q. In this case we extend Q and proceed to
stage k + 1. (ii) No sequence of rotations leads to Case (i). In this case let
END denote the set of endpoints of the paths produced. If y ∈ END then Py
denotes a path with endpoints x0 , y that is obtained from P by a sequence
of rotations.
(c) If we are in Case (bii) then for each y ∈ END we let END(y) denote the
set of vertices z such that there exists a longest path Qz from y to z such
that Qz is obtained from Py by a sequence of rotations with vertex y fixed.
Repeating the argument above in (b) for each y ∈ END, we either extend
a path and begin Stage k + 1 or we go to (d).
(d) Suppose now that we do not reach Stage k + 1 by an extension and that
we have constructed the sets END and END(y) for all y ∈ END. Suppose
that G contains an edge (y, z) where z ∈ END(y). Such an edge would
imply the existence of a cycle C = (z, Qy , z). If this is not a Hamilton cycle
then connectivity implies that there exist u ∈ C and v ∈ / C such that u, v are
joined by an edge. Let w be a neighbor of u on C and let P0 be the path
obtained from C by deleting the edge (u, w). This creates a path of length
k + 1 viz. the path w, P0 , v, and we can move to Stage k + 1.

A pair z, y where z ∈ END(y) is called a booster in the sense that if we added

this edge to Gn,p1 then it would either (i) make the graph Hamiltonian or (ii)
make the current path longer. We argue now that Gn,p2 can be used to “boost”
P to a Hamilton cycle, if necessary.
We observe now that when G = Gn,p1 , |END| ≥ αn w.h.p., see (6.9). Also,
|END(y)| ≥ αn for all y ∈ END. So we will have Ω(n2 ) boosters.
For a graph G let λ (G) denote the length of a longest path in G, when G
is not Hamiltonian and let λ (G) = n when G is Hamiltonian. Let the edges of
Gn,p2 be { f1 , f2 , . . . , fs } in random order, where s ≈ ωn/4. Let G0 = Gn,p1 and
Gi = Gn,p1 + { f1 , f2 , . . . , fi } for i ≥ 1. It follows from Lemmas 6.3 and 6.4 that
if λ (Gi ) < n then, assuming Gn,p1 has the expansion claimed in Lemma 6.4,

α2
P(λ (Gi+1 ) ≥ λ (Gi ) + 1 | f1 , f2 , . . . , fi ) ≥ , (6.10)
2

see (6.6), replacing A(v) by END(v).

It follows that

P(Gn,p is not Hamiltonian) ≤ o(1) + P(Bin(s, α 2 /2) < n) = o(1). (6.11)

94 Spanning Subgraphs

6.3 Long Paths and Cycles in Sparse Random Graphs

In this section we study the length of the longest path and cycle in Gn,p when
p = c/n where c = O(log n), most importantly for c is a large constant. We
have seen in Chapter 1 that under these conditions, Gn,p will w.h.p. have iso-
lated vertices and so it will not be Hamiltonian. We can however show that it
contains a cycle of length Ω(n) w.h.p.
The question of the existence of a long path/cycle was posed by Erdős and
Rényi in [280]. The first positive answer to this question was given by Ajtai,
Komlós and Szemerédi [10] and by de la Vega [735]. The proof we give here
is due to Krivelevich, Lee and Sudakov. It is subsumed by the more general
results of [515].

Theorem 6.8 Let p = c/n where c is sufficiently large but c = O(log n). Then
w.h.p.
 
(a) Gn,p has a path of length at least 1 − 6 log
c
c
n.
 
(b) Gn,p has a cycle of length at least 1 − 12 log
c
c
n.

Proof We prove this theorem by analysing simple properties of Depth First

Search (DFS). This is a well known algorithm for exploring the vertices of a
component of a graph. We can describe the progress of this algorithm using
three sets: U is the set of unexplored vertices that have not yet been reached by
the search. D is the set of dead vertices. These have been fully explored and no
longer take part in the process. A = {a1 , a2 , . . . , ar } is the set of active vertices
and they form a path from a1 to ar . We start the algorithm by choosing a vertex
v from which to start the process. Then we let

A = {v} and D = 0/ and U = [n] \ {v} and r = 1.

We now describe how these sets change during one step of the algorithm.

Step (a) If there is an edge {ar , w} for some w ∈ U then we choose one such
w and extend the path defined by A to include w.

ar+1 ← w; A ← A ∪ {w};U ← U \ {w}; r ← r + 1.

We now repeat Step (a).

If there is no such w then we do Step (b):
Step (b) We have now completely explored ar .

D ← D ∪ {ar }; A ← A \ {ar }; r ← r − 1.

If r ≥ 1 we go to Step (a). Otherwise, if U = 0/ at this point then

 6.3 Long Paths and Cycles in Sparse Random Graphs 95

we terminate the algorithm. If U 6= 0/ then we choose some v ∈ U to

re-start the process with r = 1. We then go to Step (a).
We make the following simple observations:
• A step of the algorithm increases |D| by one or decreases |U| by one and so
at some stage we must have |D| = |U| = s for some positive integer s.
• There are no edges between D and U because we only add ar to D when
there are no ar : U edges and U does not increase from this point on.
Thus at some stage we have two disjoint sets D,U of size s with no edges
between them and a path of length |A| − 1 = n − 2s −  1. This plus the following
claim implies that Gn,p has a path P of length at least 1 − 6 log
c
c
n w.h.p. Note
that if c is large then
3 log c 2 e
α> implies c > log .
c α α

Claim 5 Let 0 < α < 1 be a positive constant. If p = c/n and c > α2 log αe

then w.h.p. in Gn,p , every pair of disjoint sets S1 , S2 of size at least αn − 1 are
joined by at least one edge.
Proof The probability that there exist sets S1 , S2 of size (at least) αn − 1 with
no joining edge is at most
2 !αn−1
e2+o(1) −cα

n (αn−1)2
(1 − p) ≤ e = o(1).
αn − 1 α2

To complete the proof of the theorem, we apply the above lemma to the
vertices S1 , S2 on the two sub-paths P1 , P2 of length 3 log c
c n at each end of P.
There will w.h.p. be an edge joining S1 , S2 , creating the cycle of the claimed
length.
Krivelevich and Sudakov [523] used DFS to give simple proofs of good
bounds on the size of the largest component in Gn,p for p = 1+ε n where ε is a
small constant. Exercises 6.7.19, 6.7.20 and 6.7.21 elaborate on their results.

Completing the proof of Theorem 6.5

We need to prove part (b). So we let 1 − p = (1 − p1 )(1 − p2 ) where p2 =
1
n log log n . Then

we apply Theorem 6.8(a) to argue that w.h.p. Gn,p1 has a path
 
of length n 1 − O logloglogn n .
96 Spanning Subgraphs

Now, conditional on Gn,p1 having minimum degree at least two, the proof
of the statement of Lemma 6.4 goes through without change for θ = 1 i.e.
n
S ⊆ [n], |S| ≤ 10000 implies |N(S)| ≥ 2|S|. We can then use use the extension-
rotation argument that we  used to prove Theorem
 6.5(c). This time we only
n log log n n
need to close O log n cycles and we have Ω log log n edges. Thus (6.11)
is replaced by

P(Gn,p is not Hamiltonian | δ (Gn,p1 ) ≥ 2) ≤

   
c1 n −8 c2 n log log n
o(1) + P Bin , 10 < = o(1),
log log n log n
for some hidden constants c1 , c2 .

6.4 Greedy Matching Algorithm

In this section we see how we can use differential equations to analyse the per-
formance of a greedy algorithm for finding a large matching in a random graph.
Finding a large matching is a standard problem in Combinatorial Optimisa-
tion. The first polynomial time algorithm to solve this problem was devised
by Edmonds in 1965 and runs in time O(|V |4 ) [275]. Over the years, many
improvements have been made. Currently the fastest such algorithm is p that of
Micali and Vazirani which dates back to 1980. Its running time is O(|E| |V |)
[590]. These algorithms are rather complicated and there is a natural interest
in the performance of simpler heuristic algorithms which should find large, but
not necessarily maximum matchings. One well studied class of heuristics goes
under the general title of the GREEDY heuristic.
The following simple greedy algorithm proceeds as follows: Beginning with
graph G = (V, E) we choose a random edge e = {u, v} ∈ E and place it in a
set M. We then delete u, v and their incident edges from G and repeat. In the
following, we analyse the size of the matching M produced by this algorithm.
Algorithm GREEDY
begin
M ← 0;/
while E(G) 6= 0/ do
begin
A: Randomly choose e = {x, y} ∈ E
G ← G \ {x, y};
M ← M ∪ {e}
end;
 6.4 Greedy Matching Algorithm 97

Output M
end

(G \ {x, y} is the graph obtained from G by deleting the vertices x, y and all
incident edges.)
(B)
We will study this algorithm in the context of the pseudo-graph model Gn,m
of Section 1.3 and apply (1.17) to bring the results back to Gn,m . We will argue
next that if at some stage G has ν vertices and µ edges then G is equally likely
to be any pseudo-graph with these parameters.
We will use the method of deferred decisions, a term coined in Knuth, Mot-
wani and Pittel [499]. In this scenario, we do not expose the edges of the
pseudo-graph until we actually need to. So, as a thought experiment, think that
initially there are m boxes, each containing a uniformly random pair of distinct
integers from [n]. Until the box is opened, the contents are unknown except for
their distribution. Observe that opening box A and observing its contents tells
us nothing more about the contents of box B. This would not be the case if as
in Gn,m we insisted that no two boxes had the same contents.
Remark 6.9 A step of GREEDY involves choosing the first unopened box at
random to expose its contents x, y.
After this, the contents
of the remaining boxes will of course remain uni-
formly random over V (G) 2 . The algorithm will then ask for each box with x
or y to be opened. Other boxes will remain unopened and all we will learn is
that their contents do not contain x or y and so they are still uniform over the
remaining possible edges.

We need the following

Lemma 6.10 Suppose that m = cn for some constant c > 0. Then w.h.p. the
(B)
maximum degree in Gn,m is at most log n.
Proof The degree of a vertex is distributed as Bin(m, 2/n). So, if ∆ denotes
(B)
the maximum degree in Gn,m , then with ` = log n,
   `
2ce `
 
m 2
P(∆ ≥ `) ≤ n ≤n = o(1).
` n `

Now let X(t) = (ν(t), µ(t)),t = 1, 2, . . . , denote the number of vertices and
edges in the graph at the start of the tth iterations of GREEDY. Also, let Gt =
(Vt , Et ) = G at this point and let Gt0 = (Vt , Et \ e) where e is a uniform random
98 Spanning Subgraphs
(B)
edge of Et . Thus ν(1) = n, µ(1) = m and G1 = Gn,m . Now ν(t + 1) = ν(t) − 2
and so ν(t) = n − 2t. Let dt (·) denote degree in Gt0 and let θt (x, y) denote the
number of copies of the edge {x, y} in Gt , excluding e. Then we have
E(µ(t + 1) | Gt ) = µ(t) − (dt (x) + dt (y) − 1 + θt (x, y)).
Taking expectations over Gt we have
E(µ(t + 1)) = E(µ(t)) − E(dt (x)) − E(dt (y)) + 1 + E(θt (x, y)).
Now
ν(t)
dt (i)
E(dt (x) | Gt ) = ∑ 2µ(t) dt (i)
i=1
 
ν(t)
dt (i)
E(dt (y) | Gt ) = Ex  ∑ dt (i)
 
i=1 2µ(t) − 1
i6=x
ν(t)
dt (x)2
 
dt (i)
=∑ dt (i) − E (6.12)
i=1 2µ(t) − 1 2µ(t) − 1
ν(t)
dt (i)2
 
1
=∑ +O .
i=1 2µ(t) n − 2t

We will see momentarity that E(dt (x)2 ) = O(1). In the model GBn,m ,
!
ν(t) µ(t)
2 µ(t)−k
   
2 2 µ(t) 2
E ∑ dt (i) = ν(t) ∑ k 1−
i=1 k=0 k ν(t)k ν(t)
 
2 2µ(t)
= 2µ(t) 1 − + .
ν(t) ν(t)
So,
 
4E(µ(t)) 1
E(µ(t + 1)) = E(µ(t)) − −1+O . (6.13)
n − 2t n − 2t
Here we use Remark 6.9 to argue that E θt (x, y)) = O(1/(n − 2t)).
This suggests that w.h.p. µ(t) ≈ nz(t/n) where z(0) = c and z(τ) is the so-
lution to the differential equation
dz 4z(τ)
=− − 1.
dτ 1 − 2τ
This is easy to solve and gives
 
1 1 − 2τ
z(τ) = c + (1 − 2τ)2 − .
2 2
 6.4 Greedy Matching Algorithm 99
c
The smallest root of z(τ) = 0 is τ = 2c+1 . This suggests the following theorem.

Theorem 6.11 W.h.p., running GREEDY on Gn,cn finds a matching of size

c+o(1)
2c+1 n.
(B)
Proof We will replace Gn,m by Gn,m and consider the random sequence µ(t),
t = 1, 2, . . .. The number of edges in the matching found by GREEDY equals
one less than the first value of t for which µ(t) = 0. We show that w.h.p. µ(t) >
0 if and only if t ≤ c+o(1)
2c+1 n. We will use Theorem 23.1 of Chapter 23.
In our set up for the theorem we let
4x
f (τ, x) = − − 1.
1 − 2τ
 
1 c 1
D = (τ, x) : − < t < TD = ,0 < x < .
n 2c + 1 2
We let X(t) = µ(t) for the statement of the theorem. Then we have to check
the conditions:
(P1) |µ(t)| ≤ cn, ∀t < TD = TD n.
(P2) |µ(t + 1) − µ(t)| ≤ 2 log n, ∀t < TD .
(P3) | E(µ(t + 1) − µ(t)|Ht , E ) − f (t/n, X(t)/n)| ≤ An , ∀t < TD .
Here E = {∆ ≤ log n} and this is needed for (P2).
(P4) f (t, x) is continuous and satisfies a Lipschitz condition
| f (t, x) − f (t 0 , x0 )| ≤ Lk(t, x) − (t 0 , x0 )k∞ where L = 10(2c + 1)2 ,
for (t, x), (t 0 , x0 ) ∈ D ∩ {(t, x) : t ≥ 0}
4x
Here f (t, x) = −1 − 1−2t and we can justify L of P4 as follows:

| f (t, x) − f (t 0 , x0 )|
4x0 
 
 4x
=  −
1 − 2t 1 − 2t 0 
4(x − x0 )   8x0 (t − t 0 )   80t(x − x0 ) 
     

≤   +   +  
(1 − 2t)(1 − 2t 0 )   (1 − 2t)(1 − 2t 0 )   (1 − 2t)(1 − 2t 0 ) 
≤ 10(2c + 1)2 .

Now let β = n1/5 and λ = n−1/20 and σ = T D −10λ and apply the theorem.
This shows that w.h.p. µ(t) = nz(t/n) + O(n19/20 ) for t ≤ σ n.
The result in Theorem 6.11 is taken from Dyer, Frieze and Pittel [273],
where a central limit theorem is proven for the size of the matching produced
by GREEDY.
The use of differential equations to approximate the trajectory of a stochastic
process is quite natural and is often very useful. It is however not always best
100 Spanning Subgraphs

practise to try and use an “off the shelf” theorem like Theorem 23.1 in order to
get a best result. It is hard to design a general theorem that can deal optimally
with terms that are o(n).

6.5 Random Subgraphs of Graphs with Large Minimum

Degree
Here we prove an extension of Theorem 6.8. The setting is this. We have a
sequence of graphs Gk with minimum degree at least k, where k → ∞. We
construct a random subgraph G p of G = Gk by including each edge of G, in-
dependently with probability p. Thus if G = Kn , G p is Gn,p . The theorem we
prove was first proved by Krivelevich, Lee and Sudakov [515]. The argument
we present here is due to Riordan [665].
In the following we abbreviate (Gk ) p to G p where the parameter k is to be
understood.

Theorem 6.12 Let Gk be a sequence of graphs with minimum degree at least

k where k → ∞. Let p be such that pk → ∞ as k → ∞. Then w.h.p. G p contains
a cycle of length at least (1 − o(1))k.
Proof We will assume that G has n vertices. We let T denote the forest pro-
duced by depth first search. We also let D,U, A be as in the proof of Theorem
6.8. Let v be a vertex of the rooted forest T . There is a unique vertical path
from v to the root of its component. We write A (v) for the set of ancestors of
v, i.e., vertices (excluding v) on this path. We write D(v) for the set of descen-
dants of v, again excluding v. Thus w ∈ D(v) if and only if v ∈ A (w). The
distance d(u, v) between two vertices u and v on a common vertical path is just
their graph distance along this path. We write Ai (v) and Di (v) for the set of
ancestors/descendants of v at distance exactly i, and A≤i (v), D≤i (v) for those
at distance at most i. By the depth of a vertex we mean its distance from the
root. The height of a vertex v is max {i : Di (v) 6= 0}.
/ Let R denote the set of
edges of G that are not tested for inclusion in G p during the exploration.
Lemma 6.13 Every edge e of R joins two vertices on some vertical path in
T.
Proof Let e = {u, v} and suppose that u is placed in D before v. When u
is placed in D, v cannot be in U, else {u, v} would have been tested. Also, v
cannot be in D by our choice of u. Therefore at this time v ∈ A and there is a
vertical path from v to u.
 6.5 Random Subgraphs of Graphs with Large Minimum Degree 101

Lemma 6.14 With high probability, at most 2n/p = o(kn) edges are tested
during the depth first search exploration.

Proof Each time an edge is tested, the test succeeds (the edge is found to be
present) with probability p. The Chernoff bound implies that the probability
that more than 2n/p tests are made but fewer than n succeed is o(1). But every
successful test contributes an edge to the forest T , so w.h.p. at most n tests are
successful.
From now on let us fix an arbitrary (small) constant 0 < ε < 1/10. We call
a vertex v full if it is incident with at least (1 − ε)k edges in R.

Lemma 6.15 With high probability, all but o(n) vertices of Tk are full.

Proof Since G has minimum degree at least k, each v ∈ V (G) = V (T ) that

is not full is incident with at least εk tested edges. If for some constant c > 0
there are at least cn such vertices, then there are at least cεkn/2 tested edges.
But the probability of this is o(1) by Lemma 6.14.
Let us call a vertex v rich if |D(v)| ≥ εk, and poor otherwise. In the next
two lemmas, (Tk ) is a sequence of rooted forests with n vertices. We suppress
the dependence on k in notation.

Lemma 6.16 Suppose that T = Tk contains o(n) poor vertices. Then, for any
constant C, all but o(n) vertices of T are at height at least Ck.

Proof For each rich vertex v, let P(v) be a set of dεke descendants of v, ob-
tained by choosing vertices of D(v) one-by-one starting with those furthest
from v. For every w ∈ P(v) we have D(w) ⊆ P(v), so |D(w)| < εk, i.e., w is
poor. Consider the set S1 of ordered pairs (v, w) with v rich and w ∈ P(v). Each
of the n−o(n) rich vertices appears in at least εk pairs, so |S1 | ≥ (1−o(1))εkn.
For any vertex w we have |A≤i (w)| ≤ i, since there is only one ancestor at
each distance, until we hit the root. Since (v, w) ∈ S1 implies that w is poor
and v ∈ A (w), and there are only o(n) poor vertices, at most o(Ckn) = o(kn)
pairs (v, w) ∈ S1 satisfy d(v, w) ≤ Ck. Thus S10 = {(v, w) ∈ S1 : d(v, w) > Ck}
satisfies |S10 | ≥ (1 − o(1))εkn. Since each vertex v is the first vertex of at most
dεke ≈ εk pairs in S1 ⊇ S10 , it follows that n − o(n) vertices v appear in pairs
(v, w) ∈ S10 . Since any such v has height at least Ck, the proof is complete.
Let us call a vertex v light if |D ≤(1−5ε)k (v)| ≤ (1 − 4ε)k, and heavy other-
wise. Let H denote the set of heavy vertices in T .

Lemma 6.17 Suppose that T = Tk contains o(n) poor vertices, and let X ⊆
V (T ) with |X| = o(n). Then, for k large enough, T contains a vertical path P
of length at least ε −2 k containing at most ε 2 k vertices in X ∪ H.
102 Spanning Subgraphs

Proof Let S2 be the set of pairs (u, v) where u is an ancestor of v and 0 <
d(u, v) ≤ (1 − 5ε)k. Since a vertex has at most one ancestor at any given dis-
tance, we have |S2 | ≤ (1 − 5ε)kn. On the other hand, by Lemma 6.16 all but
o(n) vertices u are at height at least k and so appear in at least (1 − 5ε)k pairs
(u, v) ∈ S2 . It follows that only o(n) vertices u are in more than (1 − 4ε)k such
pairs, i.e., |H| = o(n).
Let S3 denote the set of pairs (u, v) where v ∈ X ∪ H, u is an ancestor of v,
and d(u, v) ≤ ε −2 k. Since a given v can only appear in ε −2 k pairs (u, v) ∈ S3 ,
we see that |S3 | ≤ ε −2 k|X ∪ H| = o(kn). Hence only o(n) vertices u appear in
more than ε 2 k pairs (u, v) ∈ S3 .
By Lemma 6.16, all but o(n) vertices are at height at least ε −2 k. Let u be
such a vertex appearing in at most ε 2 k pairs (u, v) ∈ S3 , and let P be the vertical
path from u to some v ∈ D ε −2 k (u). Then P has the required properties.

Proof of Theorem 6.12

Fix ε > 0. It suffices to show that w.h.p. G p contains a cycle of length at least
(1 − 5ε)k, say. Explore G p by depth-first search as described above. We condi-
tion on the result of the exploration, noting that the edges of R are still present
independently with probability p. By Lemma 6.13, {u, v} ∈ R implies that u is
either an ancestor or a descendant of v. By Lemma 6.15, we may assume that
all but o(n) vertices are full.
Suppose that

| {u : {u, v} ∈ R, d(u, v) ≥ (1 − 5ε)k} | ≥ εk. (6.14)

for some vertex v. Then, since εkp → ∞, testing the relevant edges {u, v} one-
by-one, w.h.p we find one present in G p , forming, together with T , the required
long cycle. On the other hand, suppose that (6.14) fails for every v. Suppose
that some vertex v is full but poor. Since v has at most εk descendants, there
are at least (1 − 2ε)k pairs {u, v} ∈ R with u ∈ A (v). Since v has only one
ancestor at each distance, it follows that (6.14) holds for v, a contradiction.
We have shown that we can assume that no poor vertex is full. Hence there
are o(n) poor vertices, and we may apply Lemma 6.17, with X the set of ver-
tices that are not full. Let P be the path whose existence is guaranteed by
the lemma, and let Z be the set of vertices on P that are full and light, so
|V (P) \ Z| ≤ ε 2 k. For any v ∈ Z, since v is full, there are at least (1 − ε)k ver-
tices u ∈ A (v) ∪ D(v) with {u, v} ∈ R. Since (6.14) does not hold, at least
(1 − 2ε)k of these vertices satisfy d(u, v) ≤ (1 − 5ε)k. Since v is light, in turn
at least 2εk of these u must be in A (v). Recalling that a vertex has at most one
ancestor at each distance, we find a set R(v) of at least εk vertices u ∈ A (v)
with {u, v} ∈ R and εk ≤ d(u, v) ≤ (1 − 5ε)k ≤ k.
 6.6 Spanning Subgraphs 103

It is now easy to find a (very) long cycle w.h.p. Recall that Z ⊆ V (P) with
|V (P) \ Z| ≤ ε 2 k. Thinking of P as oriented upwards towards the root, let v0
be the lowest vertex in Z. Since |R(v0 )| ≥ εk and kp → ∞, w.h.p. there is an
edge {u0 , v0 } in G p with u0 ∈ R(v0 ). Let v1 be the first vertex below u0 along
P with v1 ∈ Z. Note that we go up at least εk steps from v0 to u0 and down
at most 1 + |V (P) \ Z| ≤ 2ε 2 k from u0 to v1 , so v1 is above v0 . Again w.h.p.
there is an edge {u1 , v1 } in G p with u1 ∈ R(v1 ), and so at least εk steps above
v1 . Continue downwards from u1 to the first v2 ∈ Z, and so on. Since ε −1 =
O(1), w.h.p. we may  continue in this way to find overlapping chords {ui , vi }
for 0 ≤ i ≤ 2ε −1 , say. (Note that we remain within P as each upwards step
has length at most k.) These chords combine with P to give a cycle of length at
least (1 − 2ε −1 × 2ε 2 )k = (1 − 4ε)k, as shown in Figure 6.6.

11
00 11
00 11
00 11
00 11
00 11
00 11
00 11
00 11
00 11
00
00
11
00
11 00
11
00
11 00
11
00
11 00
11
00
11 00
11
00
11 00
11
00
11 00
11
00
11 00
11
00
11 00
11
00
11 00
11
00
11
v0 v1 u0 v2 u1 v3 u2 v4 u3 u4

Figure 6.6 The path P, with the root off to the right. Each chord {vi , ui } has length
at least εk (and at most k); from ui to vi+1 is at most 2ε 2 k steps back along P. The
chords and the thick part of P form a cycle.

6.6 Spanning Subgraphs

Consider a fixed sequence H (d) of graphs where n = |V (H (d) )| → ∞. In partic-
ular, we consider a sequence Qd of d-dimensional cubes where n = 2d and a
sequence of 2-dimensional lattices Ld of order n = d 2 . We ask when Gn,p or
Gn,m contains a copy of H = H (d) w.h.p.
We give a condition that can be proved in quite an elegant and easy way. This
proof is from Alon and Füredi [25].
Theorem 6.18 Let H be fixed sequence of graphs with n = |V (H)| → ∞ and
maximum degree ∆, where (∆2 + 1)2 < n. If
10 logbn/(∆2 + 1)c
p∆ > , (6.15)
bn/(∆2 + 1)c
then Gn,p contains an isomorphic copy of H w.h.p.
Proof To prove this we first apply the Hajnal-Szemerédi Theorem to the
104 Spanning Subgraphs

square H 2 of our graph H.

Recall that we square a graph if we add an edge between any two vertices of our
original graph which are at distance at most two. The Hajnal-Szemerédi Theo-
rem states that every graph with n vertices and maximum vertex degree at most
d is d + 1-colorable with all color classes of size bn/(d + 1)c or dn/(d + 1)e,
i.e, the (d + 1)-coloring is equitable.
Since the maximum degree of H 2 is at most ∆2 , there exists an equitable ∆2 +1-
coloring of H 2 which induces a partition of the vertex set of H, say U = U(H),
into ∆2 + 1 pairwise disjoint subsets U1 ,U2 , . . . ,U∆2 +1 , so that each Uk is an
independent set in H 2 and the cardinality of each subset is either bn/(∆2 + 1)c
or dn/(∆2 + 1)e.
Next, partition the set V of vertices of the random graph Gn,p into pairwise
disjoint sets V1 ,V2 , . . . ,V∆2 +1 , so that |Uk | = |Vk | for k = 1, 2, . . . , ∆2 + 1.
We define a one-to-one function f : U 7→ V , which maps each Uk onto Vk re-
sulting in a mapping of H into a an isomorphic copy of H in Gn,p . In the first
step, choose an arbitrary mapping of U1 onto V1 . Now U1 is an independent
subset of H and so Gn,p [V1 ] trivially contains a copy of H[U1 ]. Assume, by
induction, that we have already defined
f : U1 ∪U2 ∪ . . . ∪Uk 7→ V1 ∪V2 ∪ . . . ∪Vk ,
and that f maps the induced subgraph of H on U1 ∪U2 ∪ . . . ∪Uk into a copy of
it in V1 ∪V2 ∪ . . . ∪Vk . Now, define f on Uk+1 , using the following construction.
 first that Uk+1 = {u1 , u2 , . . . , um } and Vk+1 = {v1 , v2 , . . . , vm } where
Suppose
m ∈ bn/(∆2 + 1)c, dn/(∆2 + 1)e .
(k)
Next, construct a random bipartite graph Gm,m,p∗ with a vertex set V = (X,Y ),
where X = {x1 , x2 , . . . , xm } and Y = {y1 , y2 , . . . , ym } and connect xi and y j with
an edge if and only if in Gn,p the vertex v j is joined by an edge to all vertices
f (u), where u is a neighbor of ui in H which belongs to U1 ∪ U2 ∪ . . . ∪ Uk .
Hence, we join xi with y j if and only if we can define f (ui ) = v j .
Note that for each i and j, the edge probability p∗ ≥ p∆ and that edges of
(k)
Gm,m,p∗ are independent of each other, since they depend on pairwise disjoint
sets of edges of Gn,p . This follows from the fact that Uk+1 is independent in H 2 .
Assuming that the condition (6.15) holds and that (∆2 + 1)2 < n, then by Theo-
(k)
rem 6.1, the random graph Gm,m,p∗ has a perfect matching w.h.p. Moreover, we
(k)
can conclude that the probability that there is no perfect matching in Gm,m,p∗ is
1
at most (∆2 +1)n . It is here that we have used the extra factor 10 in the RHS of
(6.15). We use a perfect matching in G(k) (m, m, p∗ ) to define f , assuming that
if xi and y j are matched then f (ui ) = v j . To define our mapping f : U 7→ V we
have to find perfect matchings in all G(k) (m, m, p∗ ), k = 1, 2, . . . , ∆2 + 1. The
 6.7 Exercises 105

probability that we can succeed in this is at least 1 − 1/n. This implies that
Gn,p contains an isomorphic copy of H w.h.p.

Corollary 6.19 Let n = 2d and suppose that d → ∞ and p ≥ 12 +od (1), where
od (1) is a function that tends to zero as d → ∞. Then w.h.p. Gn,p contains a
copy of a d-dimensional cube Qd .

log n
Corollary 6.20 Let n = d 2 and p = ω(n)
n1/4
, where ω(n), d → ∞. Then w.h.p.
Gn,p contains a copy of the 2-dimensional lattice Ld .

6.7 Exercises
6.7.1 Consider the bipartite graph process Γm , m = 0, 1, 2, . . . , n2 where we add
the n2 edges in A × B in random order, one by one. Show that w.h.p. the
hitting time for Γm to have a perfect matching is identical with the hitting
time for minimum degree at least one.
6.7.2 Show that

0

 if cn → −∞
−c−e −c
lim
n→∞
P(G n,p has a near perfect matching) = e if cn → c
n odd

if c → ∞.

1
n

A near pefect matching is one of size bn/2c.

6.7.3 Show that if p = log n+(k−1)nlog log n+ω where k = O(1) and ω → ∞ then
w.h.p. Gn,n,p contains a k-regular spanning subgraph.
6.7.4 Consider the random bipartite graph G with bi-partition A, B where |A| =
|B| = n. Each vertex a ∈ A independently chooses d2 log ne random neigh-
bors in B. Show that w.h.p. G contains a perfect matching.
6.7.5 Show that if p = log n+(k−1)nlog log n+ω where k = O(1) and ω → ∞ then
w.h.p. Gn,p contains bk/2c edge disjoint Hamilton cycles. If k is odd,
show that in addition there is an edge disjoint matching of size bn/2c.
(Hint: Use Lemma 6.4 to argue that after “peeling off” a few Hamilton
cycles, we can still use the arguments of Sections 6.1, 6.2).
6.7.6 Let m∗k denote the first time that Gm has minimum degree at least k. Show
that w.h.p. in the graph process (i) Gm∗1 contains a perfect matching and
(ii) Gm∗2 contains a Hamilton cycle.
 106 Spanning Subgraphs

6.7.7 Show that if p = log n+logn log n+ω where ω → ∞ then w.h.p.Gn,n,p con-
tains a Hamilton cycle. (Hint: Start with a 2-regular spanning subgraph
from (ii). Delete an edge from a cycle. Argue that rotations will always
produce paths beginning and ending at different sides of the partition.
Proceed more or less as in Section 6.2).
6.7.8 Show that if p = log n+logn log n+ω where n is even and ω → ∞ then w.h.p.
Gn,p contains a pair of vertex disjoint n/2-cycles. (Hint: Randomly par-
tition [n] into two sets of size n/2. Then move some vertices between
parts to make the minimum degree at least two in both parts).
6.7.9 Show that if three divides n and np2  log n then w.h.p. Gn,p contains
n/3 vertex disjoint triangles. (Hint: Randomly partition [n] into three sets
A, B,C of size n/3. Choose a perfect matching M between A and B and
then match C into M).
6.7.10 Let G = (X,Y, E) be an arbitrary bipartite graph where the bi-partition
X,Y satisfies |X| = |Y | = n. Suppose that G has minimum degree at least
3n/4. Let p = K logn
n
where K is a large constant. Show that w.h.p. G p
contains a perfect matching.
6.7.11 Let p = (1+ε) logn n for some fixed ε > 0. Prove that w.h.p. Gn,p is Hamil-
ton connected i.e. every pair of vertices are the endpoints of a Hamilton
path.
6.7.12 Show that if p = (1+ε)n log n for ε > 0 constant, then w.h.p. Gn,p contains a
copy of a caterpillar on n vertices. The diagram below is the case n = 16.

6.7.13 Show that for any fixed ε > 0 there exists cε such that if c ≥ cε then Gn,p
2
contains a cycle of length (1 − ε)n with probability 1 − e−cε n/10 .
6.7.14 Let p = (1 + ε) logn n for some fixed ε > 0. Prove that w.h.p. Gn,p is pan-
cyclic i.e. it contains a cycle of length k for every 3 ≤ k ≤ n.
(See Cooper and Frieze [207] and Cooper [201], [203]).
6.7.15 Show that if p is constant then

P(Gn,p is not Hamiltonian) = O(e−Ω(np) ).

6.7.16 Let T be a tree on n vertices and maximum degree less than c1 log n. Sup-
pose that T has at least c2 n leaves. Show that there exists K = K(c1 , c2 )
such that if p ≥ K log
n
n
then Gn,p contains a copy of T w.h.p.
 6.7 Exercises 107

6.7.17 Let p = 1000 n and G = Gn,p . Show that w.h.p. any red-blue coloring of
n
the edges of G contains a mono-chromatic path of length 1000 . (Hint:
Apply the argument of Section 6.3 to both the red and blue sub-graphs
of G to show that if there is no long monochromatic path then there is a
pair of large sets S, T such that no edge joins S, T .)
This question is taken from Dudek and Pralat [263]
6.7.18 Suppose that p = n−α for some constant α > 0. Show that if α > 13 then
w.h.p. Gn,p does not contain a maximal spanning planar subgraph i.e. a
planar subgraph with 3n − 6 edges. Show that if α < 31 then it contains
one w.h.p. (see Bollobás and Frieze [143]).
6.7.19 Show that the hitting time for the existence of k edge-disjoint spanning
trees coincides w.h.p. with the hitting time for minimum degree k, for
k = O(1). (See Palmer and Spencer [630]).
6.7.20 Let p = nc where c > 1 is constant. Consider the greedy algorithm for
constructing a large independent set I: choose a random vertex v and put
v into I. Then delete v and all of its neighbors. Repeat until there are
no vertices left. Use the differential equation method (see Section 6.4)
and show that w.h.p. this algorithm chooses an independent set of size at
least logc c n.
6.7.21 Consider the modified greedy matching algorithm where you first choose
a random vertex x and then choose a random edge {x, y} incident with x.
Show that applied to Gn,m , with m = cn, that w.h.p. it produces a match-
−2c

)
ing of size 1
2 + o(1) − log(2−e
4c n.
n

6.7.22 Let X1 , X2 , . . . , N = 2 be a sequence of independent Bernouilli random
variables with common probability p. Let ε > 0 be sufficiently small.
(See [523]).
7 log n
(a) Let p = 1−ε
n and let k = ε 2 . Show that w.h.p. there is no interval I
of length kn in [N] in which at least k of the variables take the value
1.
εn2
(b) Let p = 1+ε
n and let N0 = 2 . Show that w.h.p.
 
 N0 ε(1 + ε)n 
 ∑ Xi −  ≤ n2/3 .

i=1 2 

1−ε
6.7.23 Use the result of Exercise 6.7.21(a) to show that if p = n then w.h.p.
the maximum component size in Gn,p is at most 7 logε2
n
.
1+ε
6.7.24 Use the result of Exercise 6.7.21(b) to show that if p = n then w.h.p
ε2n
Gn,p contains a path of length at least 5 .
108 Spanning Subgraphs

6.8 Notes
Hamilton cycles
Multiple Hamilton cycles
There are several results pertaining to the number of distinct Hamilton cycles
in Gn,m . Cooper and Frieze [206] showed that in the graph process Gm∗2 con-
tains (log n)n−o(n) distinct Hamilton cycles w.h.p. This number was improved
n
by Glebov and Krivelevich [376] to n!pn eo(n) for Gn,p and loge n eo(n) at
time m∗2 . McDiarmid [576] showed that for Hamilton cycles, perfect match-
ings, spanning trees the expected number was much higher. This comes from
the fact that although there is a small probability that m∗2 is of order n2 , most
of the expectation comes from here. (m∗k is defined in Exercise 6.7.5).
Bollobás and Frieze [142] (see Exercise 6.7.4) showed that in the graph
process, Gm∗k contains bk/2c edge disjoint Hamilton cycles plus another edge
disjoint matching of size bn/2c if n is odd. We call this property Ak . This was
the case k = O(1). The more difficult case of the occurrence of Ak at m∗k , where
k → ∞ was verified in two papers, Krivelevich and Samotij [520] and Knox,
Kühn and Osthus [500].

Conditioning on minimum degree

Suppose that instead of taking enough edges to make the minimum degree in
Gn,m two very likely, we instead condition on having minimum degree at least
≥k denote G
two. Let Gδn,m n,m conditioned on having minimum degree at least
k = O(1). Bollobás, Fenner and Frieze [140] proved that if
 
n log n
m= + k log log n + ω(n)
2 k+1
≥k has A w.h.p.
then Gδn,m k
≥k has prop-
Bollobás, Cooper, Fenner and Frieze [137] prove that w.h.p. Gδn,cn
erty Ak−1 w.h.p. provided 3 ≤ k = O(1) and c ≥ (k + 1) . For k = 3, Frieze
3
≥3 is Hamiltonian w.h.p. for c ≥ 10.
[334] showed that Gδn,cn
≥k
The k-core of a random graphs is distributed like Gδν,µ for some (random)
ν, µ. Krivelevich, Lubetzky and Sudakov [519] prove that when a k-core first
appears, k ≥ 15, w.h.p. it has b(k − 3)/2c edge disjoint Hamilton cycles.

Algorithms for finding Hamilton cycles

Gurevich and Shelah [396] and Thomason [728] gave linear expected time
algorithms for finding a Hamilton cycle in a sufficiently dense random graph
i.e. Gn,m with m  n5/3 in the Thomason paper. Bollobás, Fenner and Frieze
 6.8 Notes 109

[139] gave an O(n3+o(1) ) time algorithm that w.h.p. finds a Hamilton cycle in
the graph Gm∗2 . Frieze and Haber [336] gave an O(n1+o(1) ) time algorithm for
≥3 for c sufficiently large.
finding a Hamilton cycle in Gδn,cn

Long cycles
A sequence of improvements, Bollobás [127]; Bollobás, Fenner and Frieze
[141] to Theorem 6.8 in the sense of replacing O(log c/c) by something smaller
led finally to Frieze [327]. He showed that w.h.p. there is a cycle of length
n(1 − ce−c (1 + εc )) where εc → 0 with c. Up to the value of εc this is best
possible.
Glebov, Naves and Sudakov [377] prove the following generalisation of
(part of) Theorem 6.5. They prove that if a graph G has minimum degree at
k+ωk (1)
least k and p ≥ log k+log log
k then w.h.p. G p has a cycle of length at least
k + 1.

Spanning Subgraphs
Riordan [662] used a second moment calculation to prove the existence of
a certain (sequence of) spanning subgraphs H = H (i) in Gn,p . Suppose that
we denote the number of vertices in a graph H by |H| and the number of
edges by e(H). Suppose that |H| = n. For k ∈ [n] we let eH (k) = max{e(F) :
H (k)
F ⊆ H, |F| = k} and γ = max3≤k≤n ek−2 . Riordan proved that if the follow-
ing conditions hold, then Gn,p contains a copy of H w.h.p.: (i) e(H) ≥ n, (ii)
N p, (1 − p)n1/2 → ∞, (iii) npγ /∆(H)4 → ∞.
This for example replaces the 12 in Corollary 6.19 by 41 .

Spanning trees
Gao, Pérez-Giménez and Sato [365] considered the existence of k edge disjoint
spanning trees in Gn,p . Using a characterisation ofmNash-Williams
[618] they
were able to show that w.h.p. one can find min δ , n−1 edge disjoint spanning
trees. Here δ denotes the minimum degree and m denotes the number of edges.
When it comes to spanning trees of a fixed structure, Kahn conjectured that
the threshold for the existence of any fixed bounded degree tree T , in terms
of number of edges, is O(n log n). For example, a comb consists of a path P
of length n1/2 with each v ∈ P being one endpoint of a path Pv of the same
length. The paths Pv , Pw being vertex disjoint for v 6= w. Hefetz, Krivelevich
and Szabó [412] proved this for a restricted class of trees i.e. those with a linear
number of leaves or with an induced path of length Ω(n). Kahn, Lubetzky and
Wormald [464], [465] verified the conjecture for combs. Montgomery [597],
110 Spanning Subgraphs

[598] sharpened the result for combs, replacing m = Cn log n by m = (1 +

ε)n log n and proved that any tree can be found w.h.p. when m = O(∆n(log n)5 ),
where ∆ is the maximum degree of T . More recently, Montgomery [600] im-
proved the upper bound on m to the optimal, m = O(∆n(log n)).

Large Matchings
Karp and Sipser [486] analysed a greedy algorithm for finding a large matching
in the random graph Gn,p , p = c/n where c > 0 is a constant. It has a much
better performance than the algorithm described in Section 6.4. It follows from
their work that if µ(G) denotes the size of the largest matching in G then w.h.p.
µ(Gn,p ) γ ∗ + γ∗ + γ ∗ γ∗
≈ 1−
n 2c
where γ∗ is the smallest root of x = c exp {−ce−x } and γ ∗ = ce−γ∗ .
Later, Aronson, Frieze and Pittel [39] tightened their analysis. This led to
≥2 . Frieze and
the consideration of the size of the largest matching in Gδn,m=cn
Pittel [354] showed that w.h.p. this graph contains a matching of size n/2 − Z
where Z is a random variable with bounded expectation. Frieze [332] proved
that in the bipartite analogue of this problem, a perfect matching exists w.h.p.
Building on this work, Chebolu, Frieze and Melsted [178] showed how to find
an exact maximum sized matching in Gn,m , m = cn in O(n) expected time.

H-factors
By an H-factor of a graph G, we mean a collection of vertex disjoint copies of a
fixed graph H that together cover all the vertices of G. Some early results on the
existence of H-factors in random graphs are given in Alon and Yuster [32] and
Ruciński [678]. For the case of when H is a tree, Łuczak and Ruciński [559]
found the precise threshold. For general H, there is a recent breakthrough paper
of Johansson, Kahn and Vu [460] that gives the threshold for strictly balanced
H and good estimates in general. See Gerke and McDowell [364] for some
further results.
 7
Extreme Characteristics

This chapter is devoted to the extremes of certain graph parameters. We look

first at the diameter of random graphs i.e. the extreme value of the shortest
distance between a pair of vertices. Then we look at the size of the largest
independent set and the the related value of the chromatic number. We decribe
an important recent result on “interpolation” that proves certain limits exist.
We end the chapter with the likely values of the first and second eigenvalues of
a random graph.

7.1 Diameter
In this section we will first discuss the threshold for Gn,p to have diameter d,
when d ≥ 2 is a constant. The diameter of a connected graph G is the maximum
over distinct vertices v, w of dist(v, w) where dist(v, w) is the minimum number
of edges in a path from v to w. The theorem below was proved independently
by Burtin [167], [168] and by Bollobás [125]. The proof we give is due to
Spencer [709].

Theorem 7.1 Let d ≥ 2 be a fixed positive integer. Suppose that c > 0 and
pd nd−1 = log(n2 /c).
Then
(
e−c/2 if k = d
lim P(diam(Gn,p ) = k) =
n→∞ 1 − e−c/2 if k = d + 1.
Proof (a): w.h.p. diam(G) ≥ d.
Fix v ∈ V and let
Nk (v) = {w : dist(v, w) = k}. (7.1)
It follows from Theorem 3.4 that w.h.p. for 0 ≤ k < d,
|Nk (v)| ≤ ∆k ≈ (np)k ≈ (n log n)k/d = o(n). (7.2)
(b) w.h.p. diam(G) ≤ d + 1

111
112 Extreme Characteristics

Fix v, w ∈ [n]. Then for 1 ≤ k < d, define the event

   
np k
Fk = |Nk (v)| ∈ Ik = , (2np)k
.
2
Then for k ≤ dd/2e we have
P(F¯k |F1 , . . . , Fk−1 ) =
! !
k−1
= P Bin n − ∑ |Ni (v)|, 1 − (1 − p)|Nk−1 (v)| ∈
/ Ik
i=0
    
3  np k−1 np k
≤ P Bin n − o(n), p ≤
4 2 2
   
5
+ P Bin n − o(n), (2np)k−1 p ≥ (2np)k
4
n  o
k
≤ exp −Ω (np)
= O(n−3 ).
So with probability 1 − O(n−3 ),
 np bd/2c  np dd/2e
|Nbd/2c (v)| ≥ and |Ndd/2e (w)| ≥ .
2 2

111
000 000
111
Y 00
11 00
11
000
111
000
111 000
111
000
111 00
11
00
11 00
11
00
11
000
111
000111
111 000
111
000
111 00
11
00
11 00
11
00
11
0000
1111
0000
1111 0001111
0000
1111 1111
0000
0000
1111000
111
000
1110000
1111
0000
1111000
111
1111 000
111
0000 1110000 1111111
00000000000111
1111
0000000000
1111
0000 0000000
1111
1111111
0000000 1111
00001111111111
000
111
000
0000
1111 000
111
0000 111
1111 0000
1111 0000
1111000
111
0000111
1111 0000
1111
0000111
0001111
1111111000
0000
1111 000
000
1110000
1111 0000
11110000000000
111
0000
1111
0000
1111 0001111
1110000
0000
1111 0000
1111
0000
1111000
111
000
1110000
1111
0000
1111
000
111
000
111
0000 111
1111 000
000
1110000
1111 0000111
1111 0000111
0001111
1111111000
1111
0000
1111
0000 0001111
111
000
111
0000
0000
1111
1111
1111
0000
1111
0000000
111
0000000
1111
0000
000
111
000
111
000
111
0000 111
1111 0000000
000
111
0000
1111 000
111
0000111
1111000
1110000111
00
11
1111000
00
11
111
000
000
111
0000
1111 000
111
111
000 000
111
0000
1111 00
11
0000000
1111 00
11
111
000
000
111
000
111 000
111
000
111 000
111
0000
1111
000
111 00
11
00
11 00
11
00
11
000
111 000
111 000
111 00
11 00
11

v X w

If X = Nbd/2c (v) and Y = Ndd/2e (w) then, either

X ∩Y 6= 0/ and dist(v, w) ≤ bd/2c + dd/2e = d,
or since the edges between X and Y are unconditioned by our construction,
   
( np d
) np d
P(6 ∃ an X : Y edge ) ≤ (1 − p) 2 ≤ exp − p
2
 7.1 Diameter 113

≤ exp{−(2 − o(1))np log n} = o(n−3 ).

So
P(∃v, w : dist(v, w) > d + 1) = o(n−1 ).
We now consider the probability that d or d + 1 is the diameter. We will use
Janson’s inequality, see Section 22.6. More precisely, we will use the earlier
inequality, Corollary 22.14, from Janson, Łuczak and Ruciński [440].
We will first use this to estimate the probability of the following event: Let
v 6= w ∈ [n] and let
Av,w = {v, w are not joined by a path of length d}.
For x = x1 , x2 , . . . , xd−1 let
Bv,x,w = {(v, x1 , x2 , . . . , xd−1 , w) is a path in Gn,p }.
Let
Z = ∑ Zx ,
x

where
(
1 if Bv,x,w occurs
Zx =
0 otherwise.
Janson’s inequality allows us to estimate the probability that Z = 0, which is
precisely the probability of Av,w .
Now
 2   
n 1
µ = E Z = (n − 2)(n − 3) · · · (n − d)pd = log 1+O .
c n
Let x = x1 , x2 , · · · , xd−1 , y = y1 , y2 , . . . , yd−1 and
∆= ∑ P(Bx ∩ By )
x,y:x6=y
v,x,w and v,y,w
share an edge
d−1  
d 2(d−1)−t 2d−t
≤ ∑ n p , t is the number of shared edges,
t=1 t
!
d−1 2d−t
2(d−1)−t− d−1
d (2d−t)
=O ∑n (log n) d
t=1
!
d−1
=O ∑ n−t/d+o(1)
t=1

= o(1).
114 Extreme Characteristics

Applying Corollary 22.14), P(Z = 0) ≤ e−µ+∆ , we get

c + o(1)
P(Z = 0) ≤ .
n2
On the other hand the FKG inequality (see Section 22.3) implies that
 (n−2)(n−3)···(n−d) c + o(1)
P(Z = 0) ≥ 1 − pd = .
n2
So
c + o(1)
P(Av,w ) = P(Z = 0) = .
n2
So
c + o(1)
E(#v, w : Av,w occurs) =
2
and we should expect that

P(6 ∃ v, w : Av,w occurs) ≈ e−c/2 . (7.3)

Indeed if we choose v1 , w1 , v2 , w2 , . . . , vk , wk , k constant, we will find that

 c k
P Av1 ,w1 , Av2 ,w2 , . . . , Avk ,wk ≈ 2 (7.4)
n
and (7.3) follows from the method of moments.
The proof of (7.3) is just a more involved version of the proof of the special
case k = 1 that we have just completed. We now let
k
Bx = Bvi ,x,wi
[

i=1

and re-define
Z = ∑ Zx ,
x

where now
(
1 if Bx occurs
Zx =
0 otherwise.

i=1 Avi ,wi .

Tk
Then we have {Z = 0} is equivalent to
Now,
n2
   
d 1
E Z = k(n − 2)(n − 3) · · · (n − d)p = k log 1+O .
c n
 7.1 Diameter 115

We need to show that the corresponding ∆ = o(1). But,

∆≤∑ ∑ P(Bvr ,x,wr ∩ Bvs ,y,ws )
r,s x,y:x6=y
vr ,x,wr and vs ,y,ws
share an edge
d−1  
2 d 2(d−1)−t 2d−t
≤k ∑ n p
t=1 t
= o(1).
This shows that
 k
2 /c+o(1) c + o(1)
P(Z = 0) ≤ e−k log(n = .
n2
On the other hand, the FKG inequality (see Section 22.3) shows that
k
P Av1 ,w1 , Av2 ,w2 , . . . , Avk ,wk ≥ ∏ P (Avi ,wi ) .


i=1

This verifies (7.4) and completes the proof of Theorem 7.1.

We turn next to a sparser case and prove a somewhat weaker result.
ω log n
Theorem 7.2 Suppose that p = n where ω → ∞. Then
log n
diam(Gn,p ) ≈ w.h.p.
log np
Proof Fix v ∈ [n] and let Ni = Ni (v) be as in (7.1). Let N≤k =
S
i≤k Ni . Using
the proof of Theorem 3.4(b) we see that we can assume that
(1 − ω −1/3 )np ≤ deg(x) ≤ (1 + ω −1/3 )np for all x ∈ [n]. (7.5)
It follows that if γ = ω −1/3 and
log n − log 3 log n
k0 = ≈
log np + γ log np
then w.h.p.
2n
|N≤k0 | ≤ ∑ ((1 + γ)np)k ≤ 2((1 + γ)np)k0 = 3 + o(1)
k≤k0
log n
and so the diameter of Gn,p is at least (1 − o(1)) log np .
We can assume that np = no(1) as larger p are dealt with in Theorem 7.1.
Now fix v, w ∈ [n] and let Ni be as in the previous paragraph. Now consider a
Breadth First Search (BFS) that constructs N1 , N2 , . . . , Nk1 where
3 log n
k1 = .
5 log np
116 Extreme Characteristics

It follows that if (7.5) holds then for k ≤ k1 we have

|Ni≤k | ≤ n3/4 and |Nk |p ≤ n−1/5 . (7.6)

Observe now that the edges from Ni to [n] \ N≤i are unconditioned by the BFS
up to layer k and so for x ∈ [n] \ N≤k ,

P(x ∈ Nk+1 | N≤k ) = 1 − (1 − p)|Nk | ≥ |Nk |p(1 − |Nk |p) ≥

ρk = |Nk |p(1 − n−1/5 ).

The events x ∈ Nk+1 are independent and so |Nk+1 | stochastically dominates

the binomial Bin(n − n3/4 , ρk ). Assume inductively that |Nk | ≥ (1 − γ)k (np)k
for some k ≥ 1. This is true w.h.p. for k = 1 by (7.5). Let Ak be the event that
(7.6) holds. It follows that

E(|Nk+1 | | Ak ) ≥ np|Nk |(1 − O(n−1/5 )).

It then follows from the Chernoff bounds (Theorem 22.6) that

 2 
γ
P(|Nk+1 | ≤ ((1 − γ)np)k+1
≤ exp − |Nk |np = o(n−anyconstant ).
4
There is a small point to be made about conditioning here. We can condition
on (7.5) holding and then argue that this only multiplies small probabilities by
1 + o(1) if we use P(A | B) ≤ P(A)/ P(B).
It follows that if
log n log n
k2 = ≈
2(log np + log(1 − γ) 2 log np
then w.h.p. we have
|Nk2 | ≥ n1/2 .

Analogously, if we do BFS from w to create Nk0 , i = 1, 2, . . . , k2 then |Nk0 2 | ≥

0
n1/2 . If N≤k2 ∩ N≤k 6= 0/ then dist(v, w) ≤ 2k2 and we are done. Otherwise, we
2
observe that the edges Nk2 : Nk0 2 between Nk2 and Nk0 2 are unconditioned (except
for (7.5)) and so
1/2 ×n1/2
P(Nk2 : Nk0 2 = 0)
/ ≤ (1 − p)n ≤ n−ω .

If Nk2 : Nk0 2 6= 0/ then dist(v, w) ≤ 2k2 +1 and we are done. Note that given (7.5),
all other unlikely events have probability O(n−anyconstant ) of occurring and so
we can inflate these latter probabilities by n2 to account for all choices of v, w.
This completes the proof of Theorem 7.2.
 7.2 Largest Independent Sets 117

7.2 Largest Independent Sets

Let α(G) denote the size of the largest independent set in a graph G.

Dense case
The following theorem was first proved by Matula [568].

1
Theorem 7.3 Suppose 0 < p < 1 is a constant and b = 1−p . Then w.h.p.

α(Gn,p ) ≈ 2 logb n.

Proof Let Xk be the number of independent sets of order k.

(i) Let
k = d2 logb ne

Then,
 
n k
E Xk = (1 − p)(2)
k
 k
ne k/2
≤ (1 − p)
k(1 − p)1/2
 k
e
≤
k(1 − p)1/2
= o(1).

(ii) Let now

k = b2 logb n − 5 logb log nc.

Let
∆= ∑ P(Si , S j are independent in Gn,p ),
i, j
Si ∼S j

where S1 , S2 , . . . , S(n) are all the k-subsets of [n] and Si ∼ S j iff |Si ∩ S j | ≥ 2.
k
By Janson’s inequality, see Theorem 22.13,

(E Xk )2
 
P(Xk = 0) ≤ exp − .
2∆
Here we apply the inequality in the context of Xk being the number of k-cliques
118 Extreme Characteristics

in the complement of Gn,p . The set [N] will be the edges of the complete graph
and the sets Di will the edges of the k-cliques. Now
n
(2k) ∑k n−k
k
(2k)−(2j)
∆ k (1 − p) j=2 k− j j (1 − p)
= 2
(E Xk )2 (2k)


n
k (1 − p)
k n−k
k

j
= ∑
(1 − p)−(2)
k− j j
n
j=2 k
k
= ∑ u j.
j=2

Notice that for j ≥ 2,

u j+1 k− j k− j
= (1 − p)− j
uj n − 2k + j + 1 j + 1
k (1 − p)− j
   2
logb n
≤ 1+O .
n n( j + 1)
Therefore,
 2  j−2
uj k 2(1 − p)−( j−2)( j+1)/2
≤ (1 + o(1))
u2 n j!
 2  j−2
2k e − j+1
≤ (1 + o(1)) (1 − p) 2 ≤ 1.
nj
So
(E Xk )2 1 n2 (1 − p)
≥ ≥ .
∆ ku2 k5
Therefore
2 /(log n)5 )
P(Xk = 0) ≤ e−Ω(n . (7.7)

Matula used the Chebyshev inequality and so he was not able to prove an
exponential bound like (7.7). This will be important when we come to discuss
the chromatic number.

Sparse Case
We now consider the case where p = d/n and d is a large constant. Frieze
[331] proved
 7.2 Largest Independent Sets 119

Theorem 7.4 Let ε > 0 be a fixed constant. Then for d ≥ d(ε) we have that
w.h.p.
 
α(Gn,p )) − 2n (log d − log log d − log 2 + 1) ≤ εn .
 
 d  d
Dani and Moore [232] have recently given an even sharper result.
In this section we will prove that if p = d/n and d is sufficiently large then
w.h.p.
 
α(Gn,p ) − 2 log d n ≤ ε log d n.
 
(7.8)
 d  d
This will follow from the following. Let Xk be as defined in the previous sec-
tion. Let
(2 − ε/4) log d (2 + ε/4) log d
k0 = n and k1 = n.
d d
Then,
(log d)2
     
  ε log d
P α(Gn,p ) − E(α(Gn,p )) ≥ n ≤ exp −Ω n .
4d d2
(7.9)
(log d)2
   
P(Xk1 > 0) ≤ exp −Ω n . (7.10)
d
( ! )
(log d)3/2
P(Xk0 > 0) ≥ exp −O n . (7.11)
d2

Let us see how (7.8) follows from these two. Indeed, (7.9) and (7.11) imply
that
ε log d
k0 ≤ E(α(Gn,p )) + n. (7.12)
4d
Furthermore (7.9) and (7.10) imply that
ε log d
k1 ≥ E(α(Gn,p )) − n. (7.13)
4d
It follows from (7.12) and (7.13) that
ε log d
|k0 − E(α(Gn,p ))| ≤ n.
2d
We obtain (7.8) by applying (7.9) once more.

Proof of (7.9): This follows directly from the Azuma-Hoeffding inequality –

see Section 22.7, in particular Lemma 22.17. If Z = α(Gn,p ) then we write
Z = Z(Y2 ,Y3 , . . . ,Yn ) where Yi is the set of edges between vertex i and vertices
120 Extreme Characteristics

[i − 1] for i ≥ 2. Y2 ,Y3 , . . . ,Yn are independent and changing a single Yi can

change Z by at most one. Therefore, for any t > 0 we have

t2
 
P(|Z − E(Z)| ≥ t) ≤ exp − .
2n − 2
ε log d
Setting t = 2d n yields (7.9).

Proof of (7.10): The first moment method gives

 k1 !k1
d (2) d (k1 −1)/2
   
n ne
Pr(Xk1 > 0) ≤ 1− ≤ · 1−
k1 n k1 n
k1
(log d)2
    
de −(1+ε/5)
≤ ·d = exp −Ω n .
2 log d d

Proof of (7.11): Now, after using Lemma 22.1(g),

k0 n−k0 k0



1 E(Xk20 ) k0 − j j
(1 − p)−(2)
j
≤ =∑ n

P(Xk0 > 0) E(Xk0 )2 j=0 k0
k0    2  j
k0 e jd jd
≤∑ · exp +O ×
j=0 j 2n n2
n − k0 k0 − j
 j  
k0
(7.14)
n n− j
k0    2  j
k0 e k0 jd jd
≤∑ · · exp +O ×
j=0 j n 2n n2
(k0 − j)2
 
exp −
n− j
k0    j  2
k0 e k0 jd 2k0 k
≤b ∑ · · exp + × exp − 0
j=0 j n 2n n n
k0
= ∑ v j. (7.15)
j=0

(The notation A ≤b B is shorthand for A = O(B) when the latter is considered

to be ugly looking).
We observe first that (A/x)x ≤ eA/e for A > 0 implies that

k0 e k0 j
   2
k
· × exp − 0 ≤ 1.
j n n
 7.3 Interpolation 121

So,

(log d)3/4 j2 d 2 jk0

 
j ≤ j0 = n =⇒ v j ≤ exp +
d 3/2 2n n
( ! )
(log d)3/2
= exp O n . (7.16)
d2
Now put
α log d 1 ε
j= n where 1/2 1/4
< α < 2− .
d d (log d) 2
Then
   
k0 e k0 jd 2k0 4e log d α log d 4 log d
· · exp + ≤ · exp +
j n 2n n αd 2 d
 
4e 4 log d
= exp
αd 1−α/2 d
< 1.

To see this note that if f (α) = αd 1−α/2 then f increases between d −1/2 and
2/ log d after which it decreases. Then note that
 
n
−1/2
o 4 log d
min f (d ), f (2 − ε) > 4e exp .
d
Thus v j < 1 for j ≥ j0 and (7.11) follows from (7.16).

7.3 Interpolation
The following theorem is taken from Bayati, Gamarnik and Tetali [64]. Note
that it is not implied by Theorem 7.4. This paper proves a number of other
results of a similar flavor for other parameters. It is an important paper in that
it verifies some very natural conjectures about some graph parameters, that
have not been susceptible to proof until now.
Theorem 7.5 There exists a function H(d) such that
E(α(Gn,bdnc ))
lim = H(d).
n→∞ n
(A)
Proof For this proof we use the model Gn,m of Section 1.3. This is proper
since we we know that w.h.p.
(A) (A)
|α(Gn,m ) − α(Gn,m )| ≤ ||E(Gn,m )| − m| ≤ log n.
122 Extreme Characteristics

We will prove that for every 1 ≤ n1 , n2 ≤ n − 1 such that n1 + n2 = n,

(A) (A) (A)
E(α(Gn,bdnc )) ≥ E(α(Gn1 ,m1 )) + E(α(Gn2 ,m2 )) (7.17)

where mi = Bin(bdnc , ni /n), i = 1, 2.  

Assume (7.17). We have E(|m j − dn j |) = O(n1/2 ). This and (7.17) and
the fact that adding/deleting one edge changes α by at most one implies that
(A) (A) (A)
E(α(Gn,bdnc )) ≥ E(α(Gn )) + E(α(Gn ,bdn c )) − O(n1/2 ). (7.18)
1 ,bdn1 c 2 2

(A)
Thus the sequence un = E(α(Gn,bdnc )) satisfies the conditions of Lemma 7.6
below and the proof of Theorem 7.5 follows.
Proof of (7.17): We begin by constructing a sequence of graphs interpolat-
(A) (A) (A)
ing between Gn,bdnc and a disjoint union of Gn1 ,m1 and Gn2 ,m2 . Given n, n1 , n2
such that n1 + n2 = n and any 0 ≤ r ≤ m = bdnc, let G(n, m, r) be the ran-
dom (pseudo-)graph on vertex set [n] obtained as follows. It contains pre-
cisely m edges. The first r edges e1 , e2 , . . . , er are selected randomly from [n]2 .
The remaining m − r edges er+1 , . . . , em are generated as follows. For each
j = r +1, . . . , m, with probability n j /n, e j is selected randomly from M1 = [n1 ]2
and with probability n2 /n, e j is selected randomly from M2 =[n1 + 1, n]2 . Ob-
serve that when r = m we have G(n, m, r) = G(A) (n, m) and when r = 0 it is
(A) (A)
the disjoint union of Gn1 ,m1 and Gn2 ,m2 where m j = Bin(m, n j /n) for j = 1, 2.
We will show next that
E(α(G(n, m, r))) ≥ E(α(G(n, m, r − 1))) for r = 1, . . . , m. (7.19)
It will follow immediately that
(A)
E(α(Gn,m )) = E(α(G(n, m, m))) ≥
(A) (A)
E(α(G(n, m, 0))) = E(α(Gn1 ,m1 )) + E(α(Gn2 ,m2 ))
which is (7.17).
Proof of (7.19): Observe that G(n, m, r − 1) is obtained from
G(n, m, r) by deleting the random edge er and then adding an edge from M1
or M2 . Let G0 be the graph obtained after deleting er , but before adding its
replacement. Remember that
G(n, m, r) = G0 + er .
We will show something stronger than (7.19) viz. that
E(α(G(n, m, r)) | G0 ) ≥ E(α(G(n, m, r − 1)) | G0 ) for r = 1, . . . , m. (7.20)
Now let O∗ ⊆ [n] be the set of vertices that belong to every largest independent
 7.4 Chromatic Number 123

set in G0 . Then for er = (x, y), α(G0 + e) = α(G0 ) − 1 if x, y ∈ O∗ and α(G0 +

e) = α(G0 ) if x ∈/ O∗ or y ∈/ O∗ . Because er is randomly chosen, we have
 ∗ 2
|O |
E(α(G0 + er ) | G0 ) − E(α(G0 )) = − .
n
By a similar argument
E(α(G(n, m, r − 1) | G0 ) − α(G0 )
n1 |O∗ ∩ M1 | 2 n2 |O∗ ∩ M2 | 2
   
=− −
n n1 n n2
n1 |O ∩ M1 | n2 |O ∩ M2 | 2
∗ ∗
 
≤− +
n n1 n n2
 ∗ 2
|O |
=−
n
= E(α(G0 + er ) | G0 ) − E(α(G0 )),
completing the proof of (7.20).
The proof of the following lemma is left as an exercise.
Lemma 7.6 Given γ ∈ (0, 1), suppose that the non-negative sequence un , n ≥
1 satisfies
un ≥ un1 + un2 − O(nγ )
for every n1 , n2 such that n1 + n2 = n. Then limn→∞ unn exists.

7.4 Chromatic Number

Let χ(G) denote the chromatic number of a graph G, i.e., the smallest number
of colors with which one can properly color the vertices of G. A coloring is
proper if no two adjacent vertices have the same color.

Dense Graphs
We will first describe the asymptotic behavior of the chromatic number of
dense random graphs. The following theorem is a major result, due to Bollobás
[133]. The upper bound without the 2 in the denominator follows directly from
Theorem 7.3. An intermediate result giving 3/2 instead of 2 was already proved
by Matula [569].
124 Extreme Characteristics
1
Theorem 7.7 Suppose 0 < p < 1 is a constant and b = 1−p . Then w.h.p.
n
χ(Gn,p ) ≈ .
2 logb n
Proof (i) By Theorem 7.3
n n
χ(Gn,p ) ≥ ≈ .
α(Gn,p ) 2 logb n
n
(ii) Let ν = (logb n)2
and k0 = 2 logb n − 4 logb logb n. It follows from (7.7) that

P(∃S : |S| ≥ ν, S does not contain an independent set of order ≥ k0 )

ν2
    
n
≤ exp −Ω (7.21)
ν (log n)5
= o(1).
So assume that every set of order at least ν contains an independent set of order
at least k0 . We repeatedly choose an independent set of order k0 among the set
of uncolored vertices. Give each vertex in this set a new color. Repeat until
the number of uncolored vertices is at most ν. Give each remaining uncolored
vertex its own color. The number of colors used is at most
n n
+ν ≈ .
k0 2 logb n

It should be noted that Bollobás did not have the Janson inequality available
to him and he had to make a clever choice of random variable for use with the
Azuma-Hoeffding inequality. His choice was the maximum size of a family
of edge independent independent sets. Łuczak [552] proved the corresponding
result to Theorem 7.7 in the case where np → 0.

Concentration

Theorem 7.8 Suppose 0 < p < 1 is a constant. Then

 2
t
P(|χ(Gn,p ) − E χ(Gn,p )| ≥ t) ≤ 2 exp −
2n
Proof Write
χ = Z(Y1 ,Y2 , . . . ,Yn ) (7.22)
where
Y j = {(i, j) ∈ E(Gn,p ) : i < j}.
 7.4 Chromatic Number 125

Then
|Z(Y1 ,Y2 , . . . ,Yn ) − Z(Y1 ,Y2 , . . . , Ŷi , . . . ,Yn )| ≤ 1

and the theorem follows from the Azuma-Hoeffding inequality, see Section
22.7, in particular Lemma 22.17.

Greedy Coloring Algorithm

We show below that a simple greedy algorithm performs very efficiently. It
uses twice as many colors as it “should” in the light of Theorem 7.7. This
algorithm is discussed in Bollobás and Erdős [138] and by Grimmett and Mc-
Diarmid [392]. It starts by greedily choosing an independent set C1 and at
the same time giving its vertices color 1. C1 is removed and then we greedily
choose an independent set C2 and give its vertices color 2 and so on, until all
vertices have been colored.

Algorithm GREEDY

• k is the current color.

• A is the current set of vertices that might get color k in the current round.
• U is the current set of uncolored vertices.

begin
k ←− 0, A ←− [n], U ←− [n], Ck ←− 0./
while U 6= 0/ do
k ←− k + 1 A ←− U
while A 6= 0/
begin
Choose v ∈ A and put it into Ck
U ←− U \ {v}
A ←− A \ ({v} ∪ N(v))
end
end

1
Theorem 7.9 Suppose 0 < p < 1 is a constant and b = 1−p . Then w.h.p.
algorithm GREEDY uses approximately n/ logb n colors to color the vertices
of Gn,p .
126 Extreme Characteristics

Proof At the start of an iteration the edges inside U are un-examined. Sup-
pose that
n
|U| ≥ ν = .
(logb n)2
We show that approximately logb n vertices get color k i.e. at the end of round
k, |Ck | ≈ logb n.
Each iteration chooses a maximal independent set from the remaining uncol-
ored vertices. Let k0 = logb n − 5 logb logb n. Then

P(∃ T : |T | ≤ k0 , T is maximally independent in U)

k0  
n t
n −t
≤∑ (1 − p)(2) 1 − (1 − p)t 0
t=1 t
k0 
ne t−1 t
 t
≤∑ (1 − p) 2 e−(n0 −t)(1−p)
t=1 t
k0 
t t
 t
≤ ∑ ne1+(1−p) e−n0 (1−p)
t=1

k0 3 1 3
≤ k0 ne2 e−(logb n) ≤ e− 2 (logb n) .

So the probability that we fail to use at least k0 colors while |U| ≥ ν is at most
1 3
ne− 2 (logb ν) = o(1).

So w.h.p. GREEDY uses at most

n n
+ν ≈ colors.
k0 logb n
We now put a lower bound on the number of colors used by GREEDY. Let

k1 = logb n + 2 logb logb n.

Consider one round. Let U0 = U and suppose u1 , u2 , . . . ∈ Ck and Ui+1 = Ui \

({ui }) ∪ N(ui )).
Then
E( |Ui+1 | |Ui ) ≤ |Ui |(1 − p),

and so, for i = 1, 2, ..

E |Ui | ≤ n(1 − p)i .

So
1
P(k1 vertices colored in one round) ≤ ,
(logb n)2
 7.4 Chromatic Number 127

and
1
P(2k1 vertices colored in one round) ≤ .
n
So let
(
1 if at most k1 vertices are colored in round i
δi =
0 otherwise

We see that
1
P(δi = 1|δ1 , δ2 , . . . , δi−1 ) ≥ 1 − .
(logb n)2
So the number of rounds that color more than k1 vertices is stochastically
dominated by a binomial with mean n/(logb n)2 . The Chernoff bounds imply
that w.h.p. the number of rounds that color more than k1 vertices is less than
2n/(logb n)2 . Strictly speaking we need to use Lemma 22.24 to justify the use
of the Chernoff bounds. Because no round colors more than 2k1 vertices we
see that w.h.p. GREEDY uses at least
n − 4k1 n/(logb n)2 n
≈ colors.
k1 logb n

Sparse Graphs
We now consider the case of sparse random graphs. We first state an important
conjecture about the chromatic number.
Conjecture: Let k ≥ 3 be a fixed positive integer. Then there exists dk >
0 such that if ε is an arbitrary positive constant and p = dn then w.h.p. (i)
χ(Gn,p ) ≤ k for d ≤ dk − ε and (ii) χ(Gn,p ) ≥ k + 1 for d ≥ dk + ε.
In the absence of a proof of this conjecture, we present the following result
due to Łuczak [553]. It should be noted that Shamir and Spencer [699] had
already proved six point concentration.

Theorem 7.10 If p < n−5/6−δ , δ > 0, then the chromatic number of Gn,p is
w.h.p. two point concentrated.

Proof To prove this theorem we need three lemmas.

Lemma 7.11
128 Extreme Characteristics

(a) Let 0 < δ < 1/10, 0 ≤ p < 1 and d = np. Then w.h.p. each subgraph H of
Gn,p on less than nd −3(1+2δ ) vertices has less than (3/2 − δ )|H| edges.
(b) Let 0 < δ < 1.0001 and let 0 ≤ p ≤ δ /n. Then w.h.p. each subgraph H of
Gn,p has less than 3|H|/2 edges.
The above lemma can be proved easily by the first moment method, see
Exercise 7.6.6. Note also that Lemma 7.11 implies that each subgraph H sat-
isfying the conditions of the lemma has minimum degree less than three, and
thus is 3-colorable, due to the following simple observation (see Bollobás [134]
Theorem V.1)

Lemma 7.12 Let k = maxH⊆G δ (H), where the maximum is taken over all
induced subgraphs of G. Then χ(G) ≤ k + 1.
Proof This is an easy exercise in Graph Theory. We proceed by induction on
|V (G)|. We choose a vertex of minimum degree v, color G − v inductively and
then color v.
The next lemma is an immediate consequence of the Azuma-Hoeffding in-
equality, see Section 22.7, in particular Lemma 22.17.

Lemma 7.13 9 Let k = k(n) be such that

1
P (χ(Gn,p ) ≥ k) > . (7.23)
log log n
Then w.h.p. all but at most n1/2 log n vertices of Gn,p can be properly colored
using k colors.
Proof Let Z be the maximum number of vertices in Gn,p that can be properly
colored with k colors. Write Z = Z(Y1 ,Y2 , . . . ,Yn ) as in (7.22). Then we have
 2
1 t
P(Z = n) > and P(|Z − E Z)| ≥ t) ≤ 2 exp − . (7.24)
log log n 2n
Putting t = 21 n1/2 log n into (7.24) shows that E Z ≥ n−t and the lemma follows
after applying the concentration inequality in (7.24) once again.
Now we are ready to present Łuczak’s ingenious argument to prove Theorem
7.10. Note first that when p is such that np → 0 as n → ∞, then by Theorem 2.1
Gn,p is a forest w.h.p. and so its chromatic number is either 1 or 2. Furthermore,
for 1/ log n < d < 1.0001 the random graph Gn,p w.h.p. contains at least one
edge and no subgraph with minimal degree larger than two (see Lemma 7.11),
which implies that χ(Gn,p ) is equal to 2 or 3 (see Lemma 7.12). Now let us
 7.5 Eigenvalues 129

assume that the edge probability p is such that 1.0001 < d = np < n1/6−δ .
Observe that in this range of p the random graph Gn,p w.h.p. contains an odd
cycle, so χ(Gn,p ) ≥ 3.
Let k be as in Lemma 7.13 and let U0 be a set of size at most u0 = n1/2 log n
such that [n] \ U0 can be properly colored with k colors. Let us construct a
nested sequence of subsets of vertices U0 ⊆ U1 ⊆ . . . ⊆ Um of Gn,p , where we
define Ui+1 = Ui ∪ {v, w}, where v, w 6∈ Ui are connected by an edge and both
v and w have a neighbor in Ui . The construction stops at i = m if such a pair
{v, w} does not exist.
Notice that m can not exceed m0 = n1/2 log n, since if m > m0 then a subgraph
of Gn,p induced by vertices of Um0 would have

|Um0 | = u0 + 2m0 ≤ 3n1/2 log n < nd −3(1+2δ )

vertices and at least 3m0 ≥ (3/2 − δ )|Um0 | edges, contradicting the statement
of Lemma 7.11.
As a result, the construction produces a set Um in Gn,p , such that its size is
smaller than nd −3(1+2δ ) and, moreover, all neighbors N(Um ) of Um form an
independent set, thus “isolating” Um from the “outside world”.
Now, the coloring of the vertices of Gn,p is an easy task. Namely, by Lemma
7.13, we can color the vertices of Gn,p outside the set Um ∪ N(Um ) with k
colors. Then we can color the vertices from N(Um ) with color k +1, and finally,
due to Lemmas 7.11 and 7.12, the subgraph induced by Um is 3-colorable and
we can color Um with any three of the first k colors.

7.5 Eigenvalues
Separation of first and remaining eigenvalues
The following theorem is a weaker version of a theorem of Füredi and Komlós
[362], which was itself a strengthening of a result of Juhász [463]. See also
Coja–Oghlan [192] and Vu [740]. In their papers, 2ω log n is replaced by 2 +
o(1) and this is best possible.

Theorem 7.14 Suppose that ω → ∞, ω = o(log n) and ω 3 (log n)2 ≤ np ≤

n − ω 3 (log n)2 . Let A denote the adjacency matrix of Gn,p . Let the eigenvalues
of A be λ1 ≥ λ2 ≥ · · · ≥ λn . Then w.h.p.

(i) λ1 ≈ np
p
(ii) |λi | ≤ 2ω log n np(1 − p) for 2 ≤ i ≤ n.
130 Extreme Characteristics

The proof of the above theorem is based on the following lemma.

In the following |x| denotes the Euclidean norm of x ∈ R.
Lemma 7.15 Let J be the all 1’s matrix and M = pJ − A. Then w.h.p.
p
kMk ≤ 2ω log n np(1 − p)
where
kMk = max |Mx| = max {|λ1 (M)|, |λn (M)|} .
|x|=1

We first show that the lemma implies the theorem. Let e denote the all 1’s
vector.
Suppose that |ξ | = 1 and ξ ⊥e. Then Jξ = 0 and
p
|Aξ | = |Mξ | ≤ kMk ≤ 2ω log n np(1 − p).
Now let |x| = 1 and let x = αu + β y where u = √1 e and y⊥e and |y| = 1. Then
n

|Ax| ≤ |α||Au| + |β ||Ay|.

We have, writing A = pJ + M, that
1 1
|Au| = √ |Ae| ≤ √ (np|e| + kMk|e|)
n n
p
≤ np + 2ω log n np(1 − p)
p
|Ay| ≤ 2ω log n np(1 − p)
Thus
p
|Ax| ≤ |α|np + (|α| + |β |)2ω log n np(1 − p)
p
≤ np + 3ω log n np(1 − p).
This implies that λ1 ≤ (1 + o(1))np.
But
|Au| ≥ |(A + M)u| − |Mu|
= |pJu| − |Mu|
p
≥ np − 2ω log n np(1 − p),
implying λ1 ≥ (1 + o(1))np, which completes the proof of (i).
Now
|Aξ |
λ2 = min max
η 06=ξ ⊥η |ξ |
 7.5 Eigenvalues 131
|Aξ |
≤ max
06=ξ ⊥u |ξ |
|Mξ |
≤ max
06=ξ ⊥u |ξ |
p
≤ 2ω log n np(1 − p)
λn = min ξ T Aξ ≥ min ξ T Aξ − pξ T Jξ
|ξ |=1 |ξ |=1
T
p
= min −ξ Mξ ≥ −kMk ≥ −2ω log n np(1 − p).
|ξ |=1

This completes the proof of (ii).

Proof of Lemma 7.15:
As in previously mentioned papers, we use the trace method of Wigner [746].
Putting M̂ = M − pIn we see that
kMk ≤ kM̂k + kpIn k = kM̂k + p
and so we bound kM̂k.
Letting mi j denote the (i, j)th entry of M̂ we have
(i) E mi j = 0
(ii) Var mi j ≤ p(1 − p) = σ 2
(iii) mi j , mi0 j0 are independent, unless (i0 , j0 ) = ( j, i),
in which case they are identical.
Now let k ≥ 2 be an even integer.
n
Trace(M̂ k ) = ∑ λi (M̂ k )
i=1
n o
≥ max λ1 (M̂ k ), λn (M̂ k )
= kM̂ k k.
We estimate
kM̂k ≤ Trace(M̂ k )1/k ,
where k = ω log n.
Now,
E(Trace(M̂ k )) = ∑ E(mi0 i1 mi1 i2 · · · mik−2 ik−1 mik−1 i0 ).
i0 ,i1 ,...,ik−1 ∈[n]

Recall that the i, jth entry of M̂ k is the sum over all products
mi,i1 mi1 ,i2 · · · mik−1 j .
 132 Extreme Characteristics

Continuing, we therefore have

k
E kM̂kk ≤ ∑ En,k,ρ
ρ=2

where
 !
 k−1 
En,k,ρ = ∏ mi j i j+1  .
 
∑ E
i0 ,i1 ,...,ik−1 ∈[n] j=0

|{i0 ,i1 ,i2 ,...,ik−1 }|=ρ

Note that as mii = 0 by construction of M̂ we have that En,k,1 = 0

Each sequence i = i0 , i1 , . . . , ik−1 , i0 corresponds to a walk W (i) on the graph
Kn with n loops added. Note that
!
k−1
E ∏ mi j i j+1 =0 (7.25)
j=0

if the walk W (i) contains an edge that is crossed exactly once, by condition (i).
On the other hand, |mi j | ≤ 1 and so by conditions (ii), (iii),
 !
 k−1 
E ∏ mi j i j+1  ≤ σ 2(ρ−1)
 
j=0
 

if each edge of W (i) is crossed at least twice and if |{i0 , i1 , . . . , ik−1 }| = ρ.

Let Rk,ρ denote the number of (k, ρ) walks i.e closed walks of length k that
visit ρ distinct vertices and do not cross any edge exactly once. We use the
following trivial estimates:

(i) ρ > 2k + 1 implies Rk,ρ = 0. (ρ this large will invoke (7.25)).

(ii) ρ ≤ 2k + 1 implies Rk,ρ ≤ nρ kk ,

where nρ bounds from above the the number of choices of ρ distinct vertices,
while kk bounds the number of walks of length k.
We have
1 k+1 1 k+1
2 2 1
2(ρ−1)
k
E kM̂k ≤ ∑ Rk,ρ σ ≤ ∑ nρ kk σ 2(ρ−1) ≤ 2n 2 k+1 kk σ k .
ρ=2 ρ=2

Therefore,

E kM̂kk
 
1 k
 1
 
P kM̂k ≥ 2kσ n 2 = P kM̂kk ≥ 2kσ n 2 ≤
1 k

2kσ n 2
 7.5 Eigenvalues 133
1
!k k
2n 2 k+1 kk σ k (2n)1/k

1
≤   = = + o(1) = o(1).
1 k 2 2
2kσ n 2
p
It follows that w.h.p. kM̂k ≤ 2σ ω(log n)n1/2 ≤ 2ω log n np(1 − p) and com-
pletes the proof of Theorem 7.14.

Concentration of eigenvalues
We show here how one can use Talagrand’s inequality, Theorem 22.18, to show
that the eigenvalues of random matrices are highly concentrated around their
median values. The result is from Alon, Krivelevich and Vu [29].

Theorem 7.16 Let A be an n × n random symmetric matrix with independent

entries ai, j = a j,i , 1 ≤ i ≤ j ≤ n with absolute value at most one. Let its eigen-
values be λ1 (A) ≥ λ2 (A) ≥ · · · ≥ λn (A). Suppose that 1 ≤ s ≤ n. Let µs denote
the median value of λs (A) i.e. µs = infµ {P(λs (A) ≤ µ) ≥ 1/2}. Then for any
t ≥ 0 we have
2 /32s2
P(|λs (A) − µs | ≥ t) ≤ 4e−t .

The same estimate holds for the probability that λn−s+1 (A) deviates from its
median by more than t.

Proof We will use Talagrand’s inequality, Theorem 22.18. We let m = n+1

2
and let Ω = Ω1 × Ω2 × · · · × Ωm where for each 1 ≤ k ≤ m we have Ωk = ai, j
for some i ≤ j. Fix a positive integer s and let M,t be real numbers. Let A be
the set of matrices A for which λs (A) ≤ M and let B be the set of matrices for
which λs (B) ≥ M + t. When applying Theorem 22.18 it is convenient to view
A as an m-vector.
Fix B ∈ B and let v(1) , v(2) , . . . , v(s) be an orthonormal set of eigenvectors
(k) (k) (k)
for the s largest eigenvalues of B. Let v(k) = (v1 , v2 , . . . , vn ),
s
(k) 2
αi,i = ∑ (vi ) for 1 ≤ i ≤ n
k=1

and
s s
s s
(k) (k) 2
αi, j = 2 ∑ (vi )2 ∑ (v j ) for 1 ≤ i < j ≤ n.
k=1 k=1

Lemma 7.17
∑ αi,2 j ≤ 2s2 .
1≤i≤ j≤n
134 Extreme Characteristics

Proof
!2 !
n s s s
(k) (k) (k)
∑ αi,2 j =∑ (vi )2
∑ +4 ∑ ∑ (vi )2 (v j )2 ∑
1≤i≤ j≤n i=1 k=1 1≤i< j≤n k=1 k=1
!2 !2
n s s n
(k) (k)
≤2 ∑∑ (vi )2 =2 ∑∑ (vi )2 = 2s2 ,
i=1 k=1 k=1 i=1

where we have used the fact that each v(k) is a unit vector.

Lemma 7.18 For every A = (ai, j ) ∈ A and B = (bi, j ) ∈ B,

∑ αi, j ≥ t/2.
1≤i≤ j≤n:ai, j 6=bi, j

Fix A ∈ A . Let u = ∑sk=1 ck v(k) be a unit vector in the span S of the vectors
(k)
v , k = 1, 2, . . . , s which is orthogonal to the eigenvectors of the (s − 1) largest
eigenvalues of A. Recall that v(k) , k = 1, 2, . . . , s are eigenvectors of B. Then
∑sk=1 c2k = 1 and ut Au ≤ λs (A) ≤ M, whereas ut Bu ≥ minv∈S vt Bv = λs (B) ≥
M + t. Recall that all entries of A and B are bounded in absolute value by 1,
implying that |bi, j − ai, j | ≤ 2 for all 1 ≤ i, j ≤ n. It follows that if X is the set
of ordered pairs (i, j) for which ai, j 6= bi, j then
!t
s s
(k) (k)
t ≤ ut (B − A)u = ∑ (bi, j − ai, j ) ∑ ck vi ∑ ck v j
(i, j)∈X k=1 k=1
  
 s  s 
(k)   (k) 
≤ 2 ∑  ∑ ck vi   ∑ ck v j 

(i, j)∈X k=1
  k=1 
s s ! s s !
s s  2 s s  2
(k) (k)
≤2 ∑ ∑ c2k ∑ vi ∑ c2k ∑ v j
(i, j)∈X k=1 k=1 k=1 k=1

=2 ∑ αi, j
(i, j)∈X

as claimed. (We obtained the third inequality by use of the Cauchy-Schwarz

inequality).

By the above two lemmas, and by Theorem 22.18 for every M and every
t >0
2 /(32s2 )
P(λs (A) ≤ M) P(λs (B) ≥ M + t) ≤ e−t . (7.26)

If M is the median of λs (A) then P(λs (A) ≤ M) ≥ 1/2, by definition, implying

 7.6 Exercises 135

that
2 /(32s2 )
P(λs (A) ≥ M + t) ≤ 2e−t .

Similarly, by applying (7.26) with M + t being the median of λs (A) we con-

clude that
2 /(32s2 )
P(λs (A) ≤ M − t) ≤ 2e−t .

This completes the proof of Theorem 7.16 for λs (A). The proof for λn−s+1
follows by applying the theorem to s and −A.

7.6 Exercises
7.6.1 Let p = d/n where d is a positive constant. Let S be the set of vertices
2 log n
of degree at least 3 log log n . Show that w.h.p., S is an independent set.
7.6.2 Let p = d/n where d is a large positive constant. Use the first moment
method to show that w.h.p.
2n
α(Gn,p ) ≤ (log d − log log d − log 2 + 1 + ε)
d
for any positive constant ε.
7.6.3 Complete the proof of Theorem 7.4.
Let m = d/(log d)2 and partition [n] into n0 = mn sets S1 , S2 , . . . , Sn0 of
size m. Let β (G) be the maximum size of an independent set S that
satisfies |S ∩ Si | ≤ 1 for i = 1, 2, . . . , n0 . Use the proof idea of Theorem
7.4 to show that w.h.p.
2n
β (Gn,p ) ≥ k−ε = (log d − log log d − log 2 + 1 − ε).
d
7.6.4 Prove Theorem7.4 using Talagrand’s inequality, Theorem 22.22.
(Hint: Let A = α(Gn,p ) ≤ k−ε − 1 ).
7.6.5 Prove Lemma 7.6.
7.6.6 Prove Lemma 7.11.
7.6.7 Prove that if ω = ω(n) → ∞ then there exists an interval I of length
ωn1/2 / log n such that w.h.p. χ(Gn,1/2 ) ∈ I. (See Scott [697]).
7.6.8 A topological clique of size s is a graph obtained from the complete
graph Ks by subdividing edges. Let tc(G) denote the size of the largest
topological clique contained in a graph G. Prove that w.h.p. tc(Gn,1/2 ) =
Θ(n1/2 ).
 136 Extreme Characteristics

7.6.9 Suppose that H is obtained from Gn,1/2 by planting a clique C of size m

= n1/2 log n inside it. describe a polynomial time algorithm that w.h.p.
finds C. (Think that an adversary adds the clique without telling you
where it is).
7.6.10 Show that if d > 2k log k for a positive integer k ≥ 2 then w.h.p. G(n, d/n)
is not k-colorable. (Hint:Consider the expected number of proper k-
coloring’s).
7.6.11 Let p = K log n/n for some large constant K > 0. Show that w.h.p. the
diameter of Gn,p is Θ(log n/ log log n).
7.6.12 Suppose that 1 + ε ≤ np = o(log n), where ε > 0 is constant. Show that
given A > 0, there exists B = B(A) such that
 
log n
P diam(K) ≥ B ≤ n−A ,
log np
where K is the giant component of Gn,p .
7.6.13 Let p = d/n for some constant d > 0. Let A be the adjacency matrix of
Gn,p . Show that w.h.p. λ1 (A) ≈ ∆1/2 where ∆ is the maximum degree in
Gn,p . (Hint: the maximum eigenvalue of the adjacency matrix of K1,m is
m1/2 ).
7.6.14 A proper 2-tone k-coloring of a graph G = (V, E) is an assignment of
pairs of colors Cv ⊆ [k], |Cv | = 2 such that (i) |Cv ∩Cw | < d(v, w) where
d(v, w) is the graph distance from v to w. If χ2 (G) denotes the mini-
mum k for which there exists a 2-tone coloring of G, show that w.h.p.
χ2 (Gn,p ) ≈ 2χ(Gn,p ). (This question is taken from [46]).
7.6.15 The set chromatic number χs (G) of a graph G = (V, E) is defined as
follows: Let C denote a set of colors. Color each v ∈ V with a color
f (v) ∈ C. Let Cv = { f (w) : {v, w} ∈ G}. The coloring is proper if Cv 6=
Cw whenever {v, w} ∈ E. χs is the minimum size of C in a proper coloring
of G. Prove that if 0 < p < 1 is  constant then w.h.p. χs (Gn,p ) ≈ r log2 n
2
where r = log 1/s and s = min q + (1 − q` )2 : ` = 1, 2, . . . where q =
2`
2
1 − p. (This question is taken from Dudek, Mitsche and Pralat [262]).

7.7 Notes
Chromatic number
There has been a lot of progress in determining the chromatic number of sparse
random graphs. Alon and Krivelevich [26] extended the result in [553] to
the range p ≤ n−1/2−δ . A breakthrough came when Achlioptas and Naor [5]
 7.7 Notes 137

identified the two possible values for np = d where d = O(1): Let kd be the
smallest integer k such that d < 2k log k. Then w.h.p. χ(Gn,p ) ∈ {kd , kd + 1}.
This implies that dk , the (conjectured) threshold for a random graph to have
chromatic number at most k, satisfies dk ≥ 2k log k − 2 logk −2 + ok (1) where
ok (1) → 0 as k → ∞. Coja–Oghlan, Panagiotou and Steger [194] extended the
result of [5] to np ≤ n1/4−ε , although here the guaranteed range is three val-
ues. More recently, Coja–Oghlan and Vilenchik [195] proved the following.
Let dk,cond = 2k log k − log k − 2 log 2. Then w.h.p. dk ≥ dk,cond − ok (1). On the
other hand Coja–Oghlan [193] proved that dk ≤ dk,cond + (2 log 2 − 1) + ok (1).
It follows from Chapter 2 that the chromatic number of Gn,p , p ≤ 1/n is
w.h.p. at most 3. Achlioptas and Moore [3] proved that in fact χ(Gn,p ) ≤ 3
w.h.p. for p ≤ 4.03/n. Now a graph G is s-colorable iff it has a homomorphism
ϕ : G → Ks . (A homomorphism from G to H is a mapping ϕ : V (G) → V (H)
such that if {u, v} ∈ E(G) then (ϕ(u), ϕ(v)) ∈ E(H)). It is therefore of interest
in the context of coloring, to consider homomorphisms from Gn,p to other
graphs. Frieze and Pegden [353] show that for any ` > 1 there is an ε > 0
such that with high probability, Gn, 1+ε either has odd-girth < 2` + 1 or has
n
a homomorphism to the odd cycle C2`+1 . They also showed that w.h.p. there
is no homomorphism from Gn,p , p = 4/n to C5 . Previously, Hatami [405] has
shown that w.h.p. there is no homomorphism from a random cubic graph to
C7 .
Alon and Sudakov [31] considered how many edges one must add to Gn,p
in order to significantly increase the chromatic number. They show that if
n−1/3+δ ≤ p ≤ 1/2 for some fixed δ > 0 then w.h.p. for every set E of
2−12 ε 2 n2
(log (np))2
edges, the chromatic number of Gn,p ∪ E is still at most 2 (1+ε)n
log (np) .
b b
Let Lk be an arbitrary function that assigns to each vertex of G a list of k col-
ors. We say that G is Lk -list-colorable if there exists a proper coloring of the
vertices such that every vertex is colored with a color from its own list. A graph
is k-choosable, if for every such function Lk , G is Lk -list-colorable. The mini-
mum k for which a graph is k-choosable is called the list chromatic number,
or the choice number, and denoted by χL (G). The study of the choice number
of Gn,p was initiated in [20], where Alon proved that a.a.s., the choice number
of Gn,1/2 is o(n). Kahn then showed (see [21]) that a.a.s. the choice number of
Gn,1/2 equals (1+o(1))χ(Gn,1/2 ). In [513], Krivelevich showed that this holds
for p  n−1/4 , and Krivelevich, Sudakov, Vu, and Wormald [525] improved
this to p  n−1/3 . On the other hand, Alon, Krivelevich, Sudakov [27] and
Vu [739] showed that for any value of p satisfying 2 < np ≤ n/2, the choice
number is Θ(np/ log(np)). Krivelevich and Vu [526] generalized this to hyper-
graphs; they also improved the leading constants and showed that the choice
138 Extreme Characteristics

number for C/n ≤ p ≤ 0.9 (where C is a sufficiently large constant) is at most

a multiplicative factor of 2 + o(1) away from the chromatic number, the best
known factor for p ≤ n−1/3 .

Algorithmic questions
We have seen that the Greedy algorithm applied to Gn,p generally produces a
coloring that uses roughly twice the minimum number of colors needed. Note
also that the analysis of Theorem 7.9, when k = 1, implies that a simple greedy
algorithm for finding a large independent set produces one of roughly half
the maximum size. In spite of much effort neither of these two results have
been significantly improved. We mention some negative results. Jerrum [457]
showed that the Metropolis algorithm was unlikely to do very well in finding
an independent set that was significantly larger than GREEDY. Other earlier
negative results include: Chvátal [188], who showed that for a significant set
of densities, a large class of algorithms will w.h.p. take exponential time to find
the size of the largest independent set and McDiarmid [573] who carried out a
similar analysis for the chromatic number.
Frieze, Mitsche, Pérez-Giménez and Pralat [351] study list coloring in an
on-line setting and show that for a wide range of p, one can asymptotically
match the best known constants of the off-line case. Moreover, if pn ≥ logω n,
then they get the same multiplicative factor of 2 + o(1).

Randomly Coloring random graphs

A substantial amount of research in Theoretical Computer Science has been
associated with the question of random sampling from complex distributions.
Of relevance here is the following: Let G be a graph and k be a positive in-
teger. Then let Ωk (G) be the set of proper k-coloring’s of the vertices of G.
There has been a good deal of work on the problem of efficiently choosing a
(near) random member of Ωk (G). For example, Vigoda [737] has described an
algorithm that produces a (near) random sample in polynomial time provided
k > 11∆(G)/6. When it comes to Gn,p , Dyer, Flaxman, Frieze and Vigoda
[270] showed that if p = d/n, d = O(1) then w.h.p. one can sample a ran-
dom coloring if k = O(log log n) = o(∆). The bound on k was reduced to
k = O(d O(1) ) by Mossell and Sly [607] and then to k = O(d) by Efthymiou
[276].

Diameter of sparse random graphs

The diameter of the giant component of Gn,p , p = λ /n, λ > 1 was considered
by Fernholz and Ramachandran [303] and by Riordan and Wormald [667]. In
 7.7 Notes 139
log n
particular, [667] proves that w.h.p. the diameter is log λ + 2 loglog n
1/λ ∗ +W where
λ ∗ < 1 and λ ∗ e −λ ∗
= λ e−λ and W = O p (1) i.e. is bounded in probability for
λ = O(1) and O(1) for λ → ∞. In addition, when λ = 1 + ε where ε 3 n → ∞
i.e. the case of the emerging giant, [667] shows that w.h.p. the diameter is
log ε 3 n log ε 3 n 3
log λ + 2 log 1/λ ∗ + W where W = O p (1/ε). If λ = 1 − ε where ε n → ∞ i.e.
the sub-critical case, then Łuczak [555] showed that w.h.p. the diameter is
log(2ε 3 n)+O p (1)
− log λ .
 8
Extremal Properties

A typical question in extremal combinatorics can be viewed as “how many

edges of the complete graph (or hypergraph) on n vertices can a graph have
without having some property P”. In recent years research has been carried
out where the complete graph is replaced by a random graph.

8.1 Containers
Ramsey theory and the Turán problem constitute two of the most important
areas in extremal graph theory. For a fixed graph H we can ask how large
should n be so that in any r-coloring of the edges of Kn can we be sure of
finding a monochromatic copy of H – a basic question in Ramsey theory. Or
we can ask for the maximum α > 0 such that we take an α proportion of the
edges of Kn without including a copy of H – a basic question related to the
Turán problem. Both of these questions have analogues where we replace Kn
by Gn,p .
There have been recent breakthroughs in transferring extremal results to the
context of random graphs and hypergraphs. Conlon and Gowers [200], Schacht
[693], Balogh, Morris and Samotij [55] and Saxton and Thomason [691] have
proved general theorems enabling such transfers. One of the key ideas being
to bound the number of independent sets in carefully chosen hypergraphs. Our
presentation will use the framework of [691] where it could just as easily have
used [55]. The use of containers is a developing field and seems to have a
growing number of applications.
In this section, we present a special case of Theorem 2.3 of [691] that will
enable us to deal with Ramsey and Turán properties of random graphs. For a
graph H with e(H) ≥ 2 we let
e(H 0 ) − 1
m2 (H) = max . (8.1)
H 0 ⊆H,e(H 0 )>1 v(H 0 ) − 2

Next let
ex(n, H)
π(H) = lim n
(8.2)
n→∞
2

140
 8.2 Ramsey Properties 141

where as usual, ex(n, H) is the maximum number of edges in an H-free sub-

graph of Kn .
Theorem 8.1 Let H be a graph with e(H) ≥ 2 and let ε be a positive con-
stant. For some constant h = h(H, ε) > 0 and n sufficiently large, there exists
a collection C of graphs on vertex set [n] such that the following holds. The
graphs C are the containers:
(a) For every H-free graph Γ there exists T ⊆ Γ ⊆ C(T ) ∈ C such that e(T ) ≤
hn2−1/m2 (H) .
(b) C contains at most εnv(H) copies of H and e(C) ≤ (π(H) + ε) n2 for every

C ∈ C.

We prove Theorem 8.1 in Section 8.4. We have extracted just enough from
Saxton and Thomason [691] and [692] to give a complete proof. But first we
give a couple of examples of the use of this theorem

8.2 Ramsey Properties

The investigation of the Ramsey properties of Gn,p was initiated by Łuczak,
Ruciński and Voigt [560]. Later, Rödl and Ruciński [669], [672] proved that
the following holds w.h.p. for some constants 0 < c < C. Here H is some fixed
graph containing at least one cycle. Suppose that the edges of Gn,m are colored
with r colors. If m < cn2−1/m2 (H) then w.h.p. there exists an r-coloring without
a mono-chromatic copy of H, while if m > Cn2−1/m2 (H) then w.h.p. in every
r-coloring there is a monochromatic copy of H.
We will give a proof of the 1-statement based on Theorem 8.1. We will
closely follow the argument in a recent paper of Nenadov and Steger [620].
The notation G → (H)er means that in every r-coloring of the edges of G there
is a copy of H with all edges the same color. Rödl and Ruciński [672] proved
the following
Theorem 8.2 For any graph H with e(H) ≥ v(H) and r ≥ 2, there exist
c0 , c1 > 0 such that
(
e o(1) p ≤ c0 n−1/m2 (H)
P(Gn,p → (H)r ) =
1 − o(1) p ≥ c1 n−1/m2 (H)

The density p0 = n−1/m2 (H) is the threshold for every edge of Gn,p to be
contained in a copy of H. When p ≤ cp0 for small c, the copies of H in Gn,p
will be spread out and the associated 0-statement is not so surprising. We will
142 Extremal Properties

use Theorem 8.1 to prove the 1-statement for p ≥ c1 p0 . The proof of the 0-
statement follows [620] and is given in Exercises 8.5.1 to 8.5.6.
We begin with a couple of lemmas:

Lemma 8.3 For every graph H and r ≥ 2 there exist constants α > 0 and n0
such that for all n ≥ n0 every r-coloring of the edges of Kn contains at least
αnv(H) monochromatic copies of H.

Proof From Ramsey’s theorem we know that there exists N = N(H, r) such
that every r-coloring of the edges of KN contains a monochromatic copy of H.
Thus, in any r-coloring of Kn , every N-subset of the vertices of Kn contains
at least one monochromatic copy of H. As every copy of H is contained in at
n−v(H)

most N−v(H) N-subsets, the theorem follows with α = 1/N v(H) .
From this we get

Corollary 8.4 For every graph H and every positive integer r there exist
constants n0 and δ , ε > 0 such that the following is true: If n ≥ n0 , then for
any E1 , E2 , . . . , Er ⊆ E(Kn ) such that for all 1 ≤ i ≤ r the set Ei contains at
most εnv(H) copies of H, we have

|E(Kn ) \ (E1 ∪ E2 ∪ · · · ∪ Er )| ≥ δ n2 .

Proof Let α and n0 be as given in Lemma 8.3 for H and r + 1. Further, let
Er+1 = E(Kn ) \ (E1 ∪ E2 ∪ · · · ∪ Er ), and consider the coloring f : E(Kn ) →
[r + 1] given by f (e) = mini∈[r+1] {e ∈ Ei }. By Lemma 8.3 there exist at least
αnv(H) monochromatic copies of H under coloring f , and so by our assumption
on the sets Ei , 1 ≤ i ≤ r, Er+1 must contain at least αnv(H) copies. As every
edge is contained in at most e(H)nv(H)−2 copies and E1 ∪ E2 ∪ · · · Er contains at
most rεnv(H) copies of H, we see that Er+1 contains at least (α − rεe(H))nv(H)
v(H)
copies of H. It follows that |Er+1 | ≥ (α−rεe(H))n
e(H)nv(H)−2
and so the corollary follows
α−re(H)ε α
with δ = e(H) . Here we take ε ≤ 2re(H) .

Proof We can now proceed to the proof of the 1-statement of Theorem 8.2.
If Gn,p 6→ (H)er then there must exist a coloring f : E(Gn,p ) → [r] such that for
all 1 ≤ i ≤ r the set Ei = f −1 (i) does not contain a copy of H. By Theorem 8.1
we have that for every such Ei there exists Ti and a container Ci such that Ti ⊆
Ei ⊆ Ci . The crucial observation is that Gn,p completely avoids E0 = E(Kn ) \
(C1 ∪C2 ∪ · · · ∪Cr ), which by Corollary 8.4 and a choice of ε has size at least
δ n2 .
Therefore, we can bound P(Gn,p 6→ (H)er ) by the probability that there exist
 8.3 Turán Properties 143

T = {T1 , . . . , Tr } and C = {Ci = C(Ti ) : i = 1, 2, . . . , r} such that E0 is edge-

disjoint from Gn,p . Thus,
P((Gn,p 6→ (H)er ) ≤ ∑ P(Ti ⊆ Gn,p , 1 ≤ i ≤ r ∧ E(Gn,p ) ∩ E0 = 0).
/
Ti ,1≤i≤r

Note that the two events in the above probability are independent and can thus
be bounded by pa (1 − p)b where a = | i Ti | and b = δ n2 . The sum can be
S

bounded by first deciding on a ≤ rhn2−1/m2 (H) (h from Theorem 8.1) and then
n

choosing a edges ( (a2) choices) and then deciding for every edge in which Ti
it appears (ra choices). Thus,
rhn2−1/m2 (H)  n 


2
P((Gn,p 6→ (H)er ) ≤ e−δ n p ∑ 2 (rp)a
a=0 a
rhn2−1/m2 (H)  2 a
−δ n2 p en rp
≤e ∑ .
a=0 2a

Recall that p = c1 n−1/m2 (H) . By choosing c1 sufficiently large with respect to

h we get

rhn2−1/m2 (H)  a
en2 rp  erc (rh/c1 )n2 p
1
∑ ≤ n2
a=0 2a 2rh
 2
ec 2rh/c1 δ δ n p/2

2 1 2 p/2
=n ≤ eδ n ,
2h
and thus P((Gn,p 6→ (H)er ) = o(1) as desired. Recall that (eA/x)x is unimodal
with a maximum at x = A and that c1 is large. This implies that n2 rp/2 >
1 2rh/c1 δ
rhn2−1/m2 (H) , giving the first inequality and ec


2h < e, giving the second
inequality.

8.3 Turán Properties

Early success on the Turán problem for random graphs was achieved by Haxell,
Kohayakawa and Łuczak [407], [408], Kohayakawa, Kreuter and Steger [502],
Kohayakawa, Łuczak and Rödl [503], Gerke, Prömel, Schickinger and Steger
[369], Gerke, Schickinger and Steger [370], Łuczak [556]. It is only recently
that Turán’s theorem in its full generality has been transferred to Gn,p .
From its definition, every H-free graph with n vertices will have
n

(π(H) + o(1)) 2 edges. In this section we prove a corresponding result for
random graphs. Our proof is taken from [691], although Conlon and Gowers
144 Extremal Properties

[200] gave a proof for 2-balanced H and Schacht [693] gave a proof for general
H.
Theorem 8.5 Suppose that 0 < γ < 1. Then there exists A > 0 such that if
p ≥ An−1/m2 (H) and n is sufficiently large then the following event occurs with
3 n
probability at least 1 − e−γ (2) p/384 :
 
n
Every H-free subgraph of Gn,p has at most (π(H) + γ) p edges.
2
To prove the theorem, we first prove the following lemma:
Lemma 8.6 Given 0 < η < 1 and h ≥ 1, there is a constant ϕ = ϕ(η, h) such
that the following holds: Let M be a set, |M| = N and let I ⊆ 2M . Let t ≥ 1,
ϕt/N ≤ p
≤ 1 and let ηN/2 ≤ d ≤ N. Suppose there exists C : 2M → 2M and
T ⊆ ≤t M
such that for each I ∈ I there exists TI ∈ T such that TI ⊆ I and
CI = C(TI ) ⊆ M, where |CI | ≤ d. Let X ⊆ M be a random subset where each
element is chosen independently with probability p. Then
2 d p/24
P(∃I ∈ I : |CI ∩ X| > (1 + η)pd and I ⊆ X) ≤ e−η . (8.3)
Proof For T ∈ T let ET be the event that
T ⊆ X and |C(T ) ∩ X| ≥ (1 + η)pd.
The event ET is contained in FT ∩ GT where FT is the event that T ⊆ X and
GT is the event that |(C(T ) \ T ) ∩ X| ≥ (1 + η)d p − |T |. Since FT and GT
are independent, P(ET ) ≤ P(FT ) P(GT ). Now |T | ≤ t ≤ N p/ϕ ≤ 2d p/ϕη ≤
ηd p/2 if ϕ is large. So by the Chernoff bound, see Lemma 22.6,
2 d p/12
P(GT ) ≤ P(Bin(d, p) ≥ (1 + η/2)d p) ≤ e−η .
Note that P(FT ) = p|T | . Let x = N p/t ≥ ϕ, so that t ≤ N p/x ≤ 2d p/ηx. If ϕ
is large we may assume that p(N − t) > t. So
t  
eN p t
 
N i 2
∑ P(FT ) ≤ ∑ i p ≤ 2 = 2(xe)t ≤ 2(xe)2d p/ηx ≤ eη d p/24 ,
T i=0 t
if ϕ, and therefore x, is large. If there exists I ⊆ X, I ∈ I with |C(TI ) ∩ X| ≥
(1 + η)d p then the event ET holds. Hence the probability in (8.3) is bounded
by
2 2 2
∑ P(FT ) P(GT ) ≤ eη d p/24 e−η d p/12 = e−η d p/24 .
T

With this lemma in hand, we can complete the proof of Theorem 8.5.
[n]

Let I be the set of H-free graphs on vertex set [n]. We take M = 2
 8.4 Containers and the proof of Theorem 8.1 145

and X = E(Gn,p ) and N = n2 . For I ∈ I , let TI and h = h(H, ε) be given

by Theorem 8.1. Each H-free graph I ∈ I is contained in CI and so if Gn,p

contains an H-free subgraph with (π(H) + γ)N p edges then there exists I such
that |X ∩CI | ≥ (π(H) + γ)N p. Our aim is to apply Lemma 8.6 with
γ  γ
η = , d = π(H) + N, t = hn2−1/m2 (H) .
2 4
The conditions of Lemma 8.6 then hold after noting that d ≥ ηN/2 and that
p ≥ An−1/m2 (H) ≥ ϕt/N if A is large enough. Note also that |CI | ≤ d. Now (1 +
η)d p ≤ (π(H) + γ)N p, and so the probability that the event in the statement
of the theorem fails to occur is bounded by
 3 
2 γ Np
e−η d p/24 ≤ exp −
384
completing the proof.

8.4 Containers and the proof of Theorem 8.1

An `-graph or `-uniform hypergraph G = (V, E) has a set of vertices V and a
collection of edges E ⊆ V` , the set of `-element subsets of V . The following

theorem generalises Theorem 8.1 to `-graphs.

Theorem 8.7 Let H be an `-graph with e(H) ≥ 2 and let ε > 0. For some
h > 0 and for every N ≥ h, there exists a collection C of `-graphs on vertex
set [N] such that

(a) for every H-free `-graph I on vertex set [N], there exists C ∈ C with I ⊆ C,
(b) for every `-graph C ∈ C , the number of copies of H in C is at most εN v(H) ,
N

and e(C) ≤ (π(H) + ε) ` ,
(c) moreover, for every I in (a), there exists T ⊆ I, e(T ) ≤ hN `−1/m(H) , such that
C = C(T ).

The degree d(σ ) of a subset σ , where |σ | ≤ r, is the number of edges of G

that contain σ and d ( j) (σ ) = max{d(σ 0 ) : σ ⊆ σ 0 ∈ [n]( j) }. We write d ( j) (v)
instead of d ( j) ({v}).

Definition 8.8 Let G be an r-graph of order n and average degree d. Let

S ⊆ V (G). The degree measure µ(S) of S is defined by
1
µ(S) = ∑ d(u).
nd u∈S
 146 Extremal Properties

Thus µ is a probability measure on V (G). We note the following inequality,

in which G is an r-graph of order n and average degree d:
1 µ(S)nd
e(G[S]) ≤ ∑ d(v) = = µ(S)e(G). (8.4)
r v∈S r

We now state the main theorem. An independent set of an `-graph is a set I

such that e ∈ E(G) implies e 6⊆ I.

Theorem 8.9 Let r ∈ N. Let G be an r-graph with average degree d and

vertex set [n]. Suppose that we can choose 0 < c, τ < 1 such that

d(σ ) ≤ cdτ |σ |−1 holds for all σ , |σ | ≥ 2. (8.5)

Then there is a function C : P[n] → P[n], such that, for every independent set
I ⊆ [n] there exists T ⊆ I with

(a) I ⊆ C(T ),
(b) µ(T ) ≤ τ,
(c) |T | ≤ τn, and
(d) µ(C(T )) ≤ 1 − c.

Corollary 8.10 Let r ∈ N and let ε > 0. Let G be an r-graph of average

degree d on vertex set [n]. Suppose that we can choose 0 < c, τ < 1 such that
(8.5) holds. Then there is a function C : P[n] → P[n], such that, for every
independent set I ⊆ [n] there exists T ⊆ I with

(a) I ⊆ C(T ),
(b) |T | ≤ τn, and
(c) e(G[C]) ≤ εe(G).

The algorithm
We now describe an algorithm which given independent set I, constructs the
quantities in Theorem 8.9 and Corollary 8.10. It runs in two modes, prune
mode, builds T ⊆ I and build mode, which constructs C ⊇ I.
The algorithm builds multigraphs Ps , s ∈ [r] and then we define the degree
of σ in the multigraph Ps to be

ds (σ ) = |{e ∈ E(Ps ) : σ ⊆ e}| ,

where we are counting edges with multiplicity in the multiset E(Ps ). (Naturally
we may write ds (v) instead of ds ({v}) if v ∈ [n].)
The algorithm uses a threshold function which makes use of a real number δ .
 8.4 Containers and the proof of Theorem 8.1 147

Definition 8.11 For s = 2, . . . , r and σ ∈ [n](≤s) , the threshold functions θs

are given as follows, where δ is the minimum real number such that d(σ ) ≤
δ dτ |σ |−1 holds for all σ , |σ | ≥ 2.

θs (σ ) = τ r−s d(v) for σ = {v}, i.e. |σ | = 1

θs (σ ) = δ dτ r−s+|σ |−1 for |σ | ≥ 2

The container algorithm is set out in Table 8.1.

I NPUT
an r-graph G on vertex set [n], with average degree d
parameters τ, ζ > 0
in prune mode a subset I ⊆ [n]
in build mode a subset T ⊆ [n]
O UTPUT
in prune mode a subset T ⊆ [n]
in build mode a subset C ⊆ [n]
I NITIALISATION
put B = {v ∈ [n] : d(v) < ζ d}
evaluate the thresholds θs (σ ), σ ∈ [n](≤s) , 1 ≤ i ≤ r
A: put Pr = E(G), Ps = 0,/ Γs = 0,/ s = 1, 2, . . . , r − 1
in prune mode put T = 0/
in build mode put C = [n]

for v = 1, 2, . . . , n do:
for s = 1, 2, . . . , r − 1 do:
let Fv,s = { f ∈ [v + 1, n](s) : {v} ∪ f ∈ Ps+1 , and 6 ∃ σ ∈ Γs , σ ⊆ f }
[here Fv,s is a multiset with multiplicities inherited from Ps+1 ]
if v ∈/ B, and |Fv,s | ≥ ζ τ r−s−1 d(v) for some s
in prune mode if v ∈ I, add v to T
in build mode if v ∈ / T , remove v from C
if v ∈ T then for s = 1, 2, . . . , r − 1 do:
add Fv,s to Ps
for each σ ∈ [v + 1, n](≤s) , if ds (σ ) ≥ θs (σ ), add σ to Γs

Table 8.1 The container algorithm

Note that C = C(T ) here, as opposed to C = C(I). Then observe that

Ps consists of sets {u1 , u2 , . . . , us } such that there exist

v1 , v2 , . . . , vr−s ∈ T where {v1 , . . . , vr−s , u1 , . . . , us } ∈ E(G). (8.6)

This is clearly true for s = r via line A. If we add f = {u1 , u2 , . . . , us } ∈ Fv,s to

148 Extremal Properties

Ps then we can inductively assert that {v1 , . . . , vr−s−1 , v, u1 , . . . , us } ∈ E(G) for

some v1 , . . . , vr−s−1 .
Note also that T ∩ Γ1 = 0/ else (8.6) implies that T and hence I contains an
edge.
We keep the degree in Ps of each set σ close to its target degree θs (σ ) and
Γs comprises those σ that have reached their target degree in Ps . After which,
we add no more to ds (u). This keeps Ps small. The multiset Fv,s is the poten-
tial contribution of v to Ps ; it is the edges of Ps+1 that contain v (with v then
removed), but which don’t contain anything from Γs . If Fv,s is large for some s
then v makes a substantial contribution to that Ps , and we place v in T , updating
all Ps and Γs accordingly. Because Ps is small, this tends to keep T small.

Analysis of the algorithm

Proof of Theorem 8.9

Lemma 8.12 For 1 ≤ s ≤ r, we have

ds (u) ≤ τ r−s (d(u) + rδ d) for all u ∈ [n], and

r−s−|σ |−1
ds (σ ) ≤ rδ dτ for all σ ⊆ [n], 2 ≤ |σ | ≤ r.

Proof We prove the bounds by induction on r − s; in fact we show ds (σ ) ≤

(r − s + 1)δ dτ r−s+|σ |−1 for |σ | ≥ 2. For s = r the bounds hold by the definition
of δ in Definition 8.11. If σ ∈ Γs then σ entered Γs after some vertex v was
inspected and the set Fv,s was added to Ps . Before this addition, ds (σ ) < θs (σ )
was true. The increase in ds (σ ) resulting from the addition is the number of s-
sets in Fv,s that contain σ . By definition of Fv,s , these come from edges of Ps+1
that contain both v and σ ; the number of these is at most ds+1 ({v} ∪ σ ). The
value of ds (σ ) remains unchanged after the addition, and so at the end we have
ds (σ ) ≤ θs (σ ) + ds+1 ({v} ∪ σ ) (for some v depending on σ ). This inequality
trivially holds if σ ∈/ Γs , and so it holds for all σ ∈ [n](≤s) . So for |σ | ≥ 2 we
have, by applying the induction hypothesis to {v} ∪ σ ,

ds (σ ) ≤ δ dτ r−s+|σ |−1 + (r − s)δ dτ r−s−1+|σ | = (r − s + 1)δ dτ r−s+|σ |−1

as claimed. For σ = {u} we apply the induction hypothesis to σ = {v, u} to

obtain

ds (u) ≤ τ r−s d(u) + (r − s)δ dτ r−s−1+2−1 ≤ τ r−s (d(u) + rδ d)

again as claimed. This completes the proof.

Lemma 8.13 Let G be an r-graph on vertex set [n] with average degree d.
 8.4 Containers and the proof of Theorem 8.1 149

Let Pr = E(G) and let Pr−1 , . . . , P1 be the multisets constructed during the al-
gorithm, either in build mode or in prune mode. Then

∑ ds (u) ≤ (µ(U) + rδ ) τ r−s nd

u∈U

holds for all subsets U ⊆ [n] and for 1 ≤ s ≤ r.

Proof The inequalities

∑ ds (u) ≤ ∑ τ r−s (d(u) + rδ d) ≤ (µ(U) + rδ ) τ r−s nd

u∈U u∈U

follow immediately from Lemma 8.12 and the definition of µ.

Lemma 8.14 Let T be produced by the algorithm in prune mode. Then
µ(T ) ≤ (r − 1)(τ/ζ )(1 + rδ ).
Proof For 1 ≤ s ≤ r − 1, let Ts = {v ∈ T : |Fv,s | ≥ ζ τ r−s−1 d(v)}. From the
operation of the algorithm we see that T ⊆ T1 ∪ · · · ∪ Tr−1 (the sets here need
not be disjoint). For each s, the sets Fv,s for v ∈ Ts are added to Ps and, because
Ps is a multiset, we obtain
1 1
ζ τ r−s−1 ndµ(Ts ) = ζ τ r−s−1 ∑ d(v) ≤ |Ps | = s ∑ ds (u) ≤ s τ r−s nd(1 + rδ )
v∈Ts u∈[n]

by Lemma 8.13 with U = [n]. Thus µ(Ts ) ≤ (τ/ζ )(1+rδ ), and µ(T ) ≤ µ(T1 )+
· · · + µ(Tr−1 ) ≤ (r − 1)(τ/ζ )(1 + rδ ).
Lemma 8.15 Let C be the set produced by the algorithm in build mode. Let
D = ([n] \C) ∪ T ∪ B. Define es by the equation |Ps | = es τ r−s nd for 1 ≤ s ≤ r.
Then
es+1 ≤ r2s es + µ(D) + ζ + 2rδ , for r − 1 ≥ s ≥ 1.
Proof The way the algorithm builds C means that T ∪ B ⊆ C. Let C0 = C \
(T ∪ B), so D = [n] \ C0 . For v ∈ [n] let fs+1 (v) be the number of sets in Ps+1
for which v is the first vertex in the vertex ordering. Then
|Ps+1 | = ∑ fs+1 (v) = ∑ fs+1 (v) + ∑ fs+1 (v) for 1 ≤ s < r. (8.7)
v∈[n] v∈C0 v∈D

By definition of |Fv,s |, of the fs+1 (v) sets in Ps+1 beginning with v, fs+1 (v) −
|Fv,s | of them contain some σ ∈ Γs . If v ∈ C0 then v ∈/ B and v ∈
/ T and so, since
v ∈ C, we have |Fv,s | < ζ τ r−s−1 d(v). Therefore, writing PΓ for the multiset of
edges in Ps+1 that contain some σ ∈ Γs , we have

∑0 ( fs+1 (v) − ζ τ r−s−1 d(v)) < |PΓ| ≤ ∑ ds+1 (σ ). (8.8)

v∈C σ ∈Γs
150 Extremal Properties

By definition, if σ ∈ Γs and |σ | ≥ 2, then ds (σ ) ≥ θs (σ ) = δ dτ r−s+|σ |−1 .

Using Lemma 8.12, we then see that ds+1 (σ ) ≤ rδ dτ r−s+|σ |−2 ≤ (r/τ)ds (σ ).
Similarly, if σ = {u} ∈ Γs then ds (σ ) ≥ τ r−s d(u) and ds+1 (σ ) ≤ τ r−s−1 (d(u)+
rδ d) ≤ (1/τ)ds (σ ) + rδ dτ r−s−1 . Therefore, for s ≥ 1, we obtain
r r
∑ ds+1 (σ ) ≤ ∑ ds (σ ) + rδ dτ r−s−1 ≤ 2s |Ps | + rδ ndτ r−s−1 .
∑
σ ∈Γs τ
σ ∈Γs {u}∈Γs
τ
(8.9)
Finally, making use of (8.7) and (8.8) together with Lemma 8.13, we have

es+1 τ r−s−1 nd = |Ps+1 | = ∑ fs+1 (v) + ∑ fs+1 (v)

v∈C0 v∈D

≤ ∑0 ζ τ r−s−1 d(v) + ∑ ds+1 (σ ) + ∑ ds+1 (v)

v∈C σ ∈Γs v∈D
r−s−1
≤ ζτ nd + ∑ ds+1 (σ ) + τ r−s−1 nd(µ(D) + rδ ) ,
σ ∈Γs

The bound (8.9) for ∑σ ∈Γs ds+1 (σ ) now gives the result claimed.

Proof of Theorem 8.9 We begin by choosing the constant c. Let γ = 2r1 r2

25r 2
and c = γ r . Let G be as in the theorem and let τ be chosen so that (8.5) is
√
satisfied. Let ζ = 2rγ. For later use, we note c ≤ γ ≤ ζ /2r ≤ 2rζ < 1.
As might be expected, we prove the theorem by using the containers C and
the sets T supplied by the algorithm. However, the input parameters we supply
to the algorithm are not τ and ζ as just defined, but instead τ∗ = γτ and ζ .
We therefore remind the reader that the values of τ and ζ appearing in the
lemmas above are those values input to the algorithm. Hence in the present
case, where we are using inputs τ ∗ and ζ , the conclusions of the lemmas hold
with τ ∗ in place of τ. Again, as highlighted earlier, the value of δ in the lemmas
is that supplied by Definition 8.11 with τ ∗ in place of τ. Explicitly, δ is (by
definition) minimal such that d(σ ) ≤ δ dτ ∗(|σ |−1) for all σ . Now τ was chosen
to satisfy (8.5), so we know that d(σ ) ≤ cdτ (|σ |−1) . Since h = γ r this implies
we know, for all σ , that d(σ ) ≤ γ r dτ (|σ |−1) ≤ γdτ ∗(|σ |−1) , because γ ≤ 1 and
|σ | ≤ r. Consequently, by the minimality of δ , we have δ ≤ γ.
What remains is to verify the claims of the theorem. Condition (a) follows
from the general properties of the algorithm.
Now cτ r−1 = γτ∗r−1 ≤ (ζ /r)τ∗r−1 , and cτ r = τ∗r ≤ (r/ζ )τ∗r . So, by Lemma 8.14,
µ(T ) ≤ (rτ∗ /ζ )(1 + rδ ) ≤ 2rτ∗ /ζ = 2rγτ/ζ = ζ τ, easily establishing condi-
tion (b). Moreover T ∩ B = 0, / so |T |ζ d ≤ ∑v∈T d(v) = ndµ(T ) ≤ ndζ τ, giv-
ing condition (c). To show that condition (d) holds, note that 2rδ ≤ 2rγ ≤ ζ ,
and so by Lemma 8.15 we comfortably have es+1 ≤ r2s es + µ(D) + 2ζ for
s+1
r − 1 ≥ s ≥ 1. Dividing the bound for e by rs+1 2( 2 ) and adding over
s+1
 8.4 Containers and the proof of Theorem 8.1 151

s = 1, . . . , r − 1, we obtain
 
er 1 1 1 1 1 2
r ≤ (µ(D) + 2ζ ) 2
+ 3 3 + 4 6 + · · · ≤ (µ(D) + 2ζ ) 2 .
(
rr 2 2 ) r r 2 r 2 r

r
Recall that er nd = |Pr | = e(G) = nd/r so er = 1/r. Hence µ(D)+2ζ ≥ r−r 2−(2) =
5γ 1/2 2r/2 ≥ 5ζ . So µ(D) ≥ 3ζ . By definition, D = [n] − (C − (T ∪ B)). Thus
µ(C) ≤ 1 − µ(D) + µ(T ) + µ(B). We showed previously that µ(T ) ≤ ζ τ ≤ ζ .
Moreover µ(B) ≤ ζ by definition of B. Therefore µ(C) ≤ 1 − 3ζ + ζ + ζ =
1 − ζ ≤ 1 − c, completing the proof.

We finish with a proof of Corollary 8.10.

Proof of Corollary 8.10 Write c∗ for the constant c from Theorem 8.9. We
prove the corollary with c = ε`−r c∗ , where ` = d(log ε)/ log(1 − c∗ )e. Let G, I
and τ be as stated in the corollary. We shall apply Theorem 8.9 several times.
Each time we apply the theorem, we do so with with τ∗ = τ/` in place of τ,
with the same I, but with different graphs G, as follows (we leave it till later
to check that the necessary conditions always hold). Given I, apply the theo-
rem to find T1 ⊆ I and I ⊆ C1 = C(T1 ), where |T1 | ≤ τ∗ n and µ(C1 ) ≤ 1 − c∗ .
It is easily shown that e(G[C1 ]) ≤ µ(C1 )e(G) ≤ (1 − c∗ )e(G) see (8.4). Now
I is independent in the graph G[C1 ] so apply the theorem again, to the r-
graph G[C1 ], to find T2 ⊆ I and a container I ⊆ C2 . We have |T2 | ≤ τ∗ |C1 |,
and e(G[C2 ]) ≤ (1 − c∗ )e(G[C1 ]) ≤ (1 − c∗ )2 e(G). We note that, in the first
application, the algorithm in build mode would have constructed C1 from in-
put T1 ∪ T2 , and would likewise have constructed C2 from input T1 ∪ T2 in the
second application. Thus C2 is a function of T1 ∪ T2 . We repeat this process
k times until we obtain the desired container C = Ck with e(G[C]) ≤ εe(G).
Since e(G[C]) ≤ (1 − c∗ )k e(G) this occurs with k ≤ `. Put T = T1 ∪ · · · ∪ Tk .
Then C is a function of T ⊆ I.
We must check that the requirements of Theorem 8.9 are fulfilled at each
application. Observe that, if d j is the average degree of G[C j ] for j < k, then
|C j |d j = re(G[C j ]) > rεe(G) = εnd, and since |C j | ≤ n we have d j ≥ εd. The
conditions of Corollary 8.10 mean that d(σ ) ≤ cdτ |σ |−1 = ε`−r c∗ dτ |σ |−1 <
|σ |−1
c∗ d j τ∗ ; since the degree of σ in G[C j ] is at most d(σ ), this means that
(8.5) is satisfied every time Theorem 8.9 is applied.
Finally condition (c) of the theorem implies |T j | ≤ τ∗ |C j | ≤ τ∗ n = τn/`, and
so |T | ≤ kτn/` ≤ τn, giving condition (b) of the corollary and completing the
proof.
152 Extremal Properties

H-free graphs
In this section we prove Theorem 8.7. We will apply the container theorem
given by Corollary 8.10 to the following hypergraph, whose independent sets
correspond to H-free `-graphs on vertex set [N].

Definition 8.16 Let H
be an `-graph. Let r = e(H).
The r-graph GH has

vertex set V (GH ) = [N]
` , where B = {v1 , ..., vr } ∈
V (GH )
r is an edge whenever
B, considered as
an `-graph with vertices in [N] and with r edges, is isomorphic
to H. So B ∈ Mr where M = [N]


` .

All that remains before applying the container theorem to GH is to verify

(8.5).

Lemma 8.17 Let H be an `-graph with r = e(H) ≥ 2 and let τ = 2`!v(H)!N −1/m(H) .
N

Let N be sufficiently large. Suppose that e(GH ) = αH v(H) where αH ≥ 1 de-
re(GH ) rαH N v(H)
pends only on H. The average degree d in GH satisfies d = = .
v(GH ) (N` )
Then,
1
d(σ ) ≤ dτ |σ |−1 , holds for all σ , |σ | ≥ 2.
αr
Proof Consider σ ⊆ [N](`) (so σ is both a set of vertices of GH and an `-
graph on vertex set [N]). The degree of σ in GH is at most the number of
ways of extending σ to an `-graph isomorphic to H. If σ as an `-graph is
not isomorphic to any subgraph of H, then clearly d(σ ) = 0. Otherwise, let
v(σ ) be the number of vertices in σ considered as an `-graph, so there exists
V ⊆ [N], |V | = v(σ ) with σ ⊆ V (`) . Edges of GH containing σ correspond
to copies of H in [N](`) containing σ , each such copy given by a choice of
v(H) − v(σ ) vertices in [N] −V and a permutation of the vertices of H. Hence
for N sufficiently large,
 
N − v(σ )
d(σ ) ≤ v(H)! ≤ v(H)!N v(H)−v(σ )
v(H) − v(σ )
Now for σ ≥ 2 we have
d(σ ) N v(H)−v(σ ) 1 −v(σ )+`+ |σm(H)
|−1 1
≤ = N ≤ .
dτ |σ |−1 αH rN v(H)−`−(|σ |−1)/m(H) αH r αH r

A well-known supersaturation theorem bounds the number of edges in con-

tainers.

Proposition 8.18 (Erdős and Simonovits [285]) Let H be an `-graph and

 8.5 Exercises 153

let ε > 0. There exists N0 and η > 0 such that if C is an `-graph on N ≥
N0
vertices containing at most ηN v(H) copies of H then e(C) ≤ (π(H) + ε) N` .

Proof of Theorem 8.7 Let η = η(ε, H) be given by Proposition 8.18, and let
β = min{ε, η}. Recall that r = e(H). Apply Corollary 8.10 to GH with c = α1H r
and τ = 2`!v(H)!N −1/m(H) and with β playing the role of ε in the corollary.
The conditions of Corollary 8.10 are satisfied; denote by c̃ the constant c ap-
pearing in the corollary. The collection of containers C satisfies the following.

• For every independent set I there exists some C ∈ C with I ⊆ C. This implies
condition (a) of the theorem,
• For each C ∈ C , we
have e(GH [C]) ≤ β N v(H) . Proposition 8.18 implies
e(C) ≤ (π(H) + ε) N` , because we chose β ≤ η. This gives condition (b).
• Finally, for
every set I as above, there exists T ⊆ I such that C = C(T ),
|T | ≤ c̃τ N` . This implies condition (c).

8.5 Exercises
8.5.1 An edge e of G is H-open if it is contained in at most one copy of H
and H-closed otherwise. The H-core ĜH of G is obtained by repeatedly
deleting H-open edges. Show that G → (H)e2 implies that ĜH 0 → (H 0 )e2
for every H 0 ⊆ H. (Thus one only needs to prove the 0-statement of
Theorem 8.2 for strictly 2-balanced H. A graph H is strictly 2-balanced
if H 0 = H is the unique maximiser in (8.1)).
8.5.2 A subgraph G0 of the H-core is H-closed if it contains at least one copy
of H and every copy of H in ĜH is contained in G0 or is edge disjoint
from G0 . Show that the edges of ĜH can be partitioned into inclusion
minimal H-closed subgraphs.
8.5.3 Show that there exists a sufficiently small c > 0 and a constant L =
L(H, c) such that if H is 2-balanced and p ≤ cn−1/m2 (H) then w.h.p. every
inclusion minimal H-closed subgraph of Gn,p has size at most L. (Try
c = o(1) first here).
8.5.4 Show that if e(G)/v(G) ≤ m2 (H) and m2 (H) > 1 then G 6→ (H)e2 .
8.5.5 Show that if H is 2-balanced and p = cn−1/m2 (H) then w.h.p. every sub-
graph G of Gn,p with v(G) ≤ L = O(1) satisfies e(G)/v(G) ≤ m2 (H).
8.5.6 Prove the 0-statement of Theorem 8.2 for m2 (H) > 1.
154 Extremal Properties

8.6 Notes
The largest triangle-free subgraph of a random graph
Babai, Simonovits and Spencer [44] proved that if p ≥ 1/2 then w.h.p. the
largest triangle-free subgraph of Gn,p is bipartite. They used Szemerédi’s reg-
ularity lemma in the proof. Using the sparse version of this lemma, Brightwell,
Panagiotou and Steger [160] improved the lower bound on p to n−c for some
(unspecified) positive constant c. DeMarco and Kahn [237] improved the lower
bound to p ≥ Cn−1/2 (log n)1/2 , which is best possible up to the value of the
constant C. And in [238] they extended their result to Kr -free graphs.

Anti-Ramsey Property
Let H be a fixed graph. A copy of H in an edge colored graph G is said to
be rainbow colored if all of its edges have a different color. The study of rain-
bow copies of H was initiated by Erdős, Simonovits and Sós [284]. An edge-
coloring of a graph G is said to be b-bounded if no color is used more than b
times. A graph is G said to have property A (b, H) if there is a rainbow copy
of H in every b-bounded coloring. Bohman, Frieze, Pikhurko and Smyth [118]
studied the threshold for Gn,p to have property A (b, H). For graphs H con-
taining at least one cycle they prove that there exists b0 such that if b ≥ b0 then
there exist c1 , c2 > 0 such that
(
0 p ≤ c1 n−1/m2 (H)
lim P(Gn,p ∈ A (b, H)) = . (8.10)
n→∞ 1 p ≥ c2 n−1/m2 (H)
A reviewer of this paper pointed out a simple proof of the 1-statement. Given
a b-bounded coloring of G, let the edges colored i be denoted ei,1 , ei,2 , . . . , ei,bi
where bi ≤ b for all i. Now consider the auxiliary coloring in which edge ei, j
is colored with j. At most b colors are used and so in the auxiliary coloring
there will be a monochromatic copy of H. The definition of the auxiliary col-
oring implies that this copy of H is rainbow in the original coloring. So the
1-statement follows directly from the results of Rödl and Ruciński [672], i.e.
Theorem 8.2.
Nenadov, Person, Škorić and Steger [619] gave further threshold results on
both Ramsey and Anti-Ramsey theory of random graphs. In particular they
proved that in many cases b0 = 2 in (8.10).
 9
Resilience

Sudakov and Vu [720] introduced the idea of the local resilience of a monotone
increasing graph property P. Suppose we delete the edges of some graph H
on vertex set [n] from Gn,p . Suppose that p is above the threshold for Gn,p to
have the property. What can we say about the value ∆ so that w.h.p. the graph
G = ([n], E(Gn,p ) \ E(H)) has property P for all H with maximum degree at
most ∆? We will denote the maximum ∆ by ∆P
In this chapter we discuss the resilience of various properties. In Section
9.1 we discuss the resilience of having a perfect matching. In Section 9.2 we
discuss the resileince of having a Hamilton cycle. In Section 9.3 we discuss
the resilience of the chromatic number.

9.1 Perfect Matchings

Sudakov and Vu [720] proved that if M denotes the property of having a per-
fect matching then

Theorem 9.1
Suppose that n = 2m
is even and that np  log n. Then w.h.p.
in Gn,p , 12 − ε np ≤ ∆M ≤ 12 + ε np for any positive constant ε.

Proof The upper bound ∆M ≤ 12 + ε np is easy to prove. Randomly parti-

tion [n] into two subsets X,Y of sizes m + 1 and m − 1 respectively. Now delete
all edges inside X so that X becomes an independent set. Clearly, the remain-
ing graph contains no perfect matching. The Chernoff bounds, Corollary 22.7,
imply that we have deleted ≈ np/2 edges incident with each vertex.
The lower bound requires a little more work. Theorem
3.4 implies that w.h.p.
the minimum degree in G is at least (1 − o(1)) 21 + ε np. We randomly parti-
tion [n] into two sets X,Y of size m. We have that w.h.p.

PM1 dY (x) & 14 + ε2 np for all x ∈ X and dX (y) & 14 + ε2 np for all y ∈ Y .

PM2 e(S, T ) ≤ 1 + ε3 np


4 |S| for all S ⊆ X, T ⊆ Y, |S| = |T | ≤ n/4.

Property PM1 follows immediately
from the Chernoff bounds, Corollary 22.7,
and the fact that dG (v) & 12 + ε np  log n.

155
156 Resilience

Property PM2 is derived as follows:

ε  np  n/4 n 2 −4ε 2 snp/27

   
Pr ∃S, T : e(S, T ) ≥ |S| 1 + ≤∑ e
3 4 s=1 s
2
!s
n/4
n2 e2−4ε np/27
≤∑ = o(1).
s=1 s2

Given, PM1, PM2, we see that if there exists S ⊆ X, |S| ≤ n/4 such that
|NX (S)| ≤ |S| then for T = NX (S),
 
1 ε  ε  np
+ np|S| . e(S, T ) ≤ 1 + |S|,
4 2 3 4
contradiction. We finish the proof that Hall’s condition holds, i.e. deal with
|S| > n/4 just as we did for |S| > n/2 in Theorem 6.1.

9.2 Hamilton Cycles

Sudakov and Vu [720] proved that if np  log
4 n and H denotes the Hamil-
tonicity property, then w.h.p. ∆H = 21 − o(1) np. This is the optimal value
for ∆H but a series of papers culminating in the following theorem due to Lee
and Sudakov.

Theorem 9.2 If np  log n, then w.h.p. ∆H = 12 − o(1) np.

Going even further, Montgomery [599] and Nenadov, Steger and Trujić
[623] have given tight hitting time versions. The proofs in these papers rely
on the use of Pósa rotations, as in Chapter 6. Some recent papers have intro-
duced the use of the absorbing method from extremal combinatorics to related
problems. The method was initiated by Rödl, Ruciński and Szemerédi [670].
Our purpose in this section is to give an example of this important technique.
Our exposition closely follows the paper of Ferber, Nenadov, Noever, Peter
and Trujić [301]. They consider the resilience of Hamiltonicity in the context
of random digraphs, but their proof can be adapted and simplified when con-
sidering graphs. Their proof in turn utilises ideas from Montgomery [599].
10
Theorem 9.3 Suppose that p ≥ logn n . Then w.h.p. in Gn,p ,
   
1 1
− ε np ≤ ∆H ≤ + ε np
2 2
for any positive constant ε.
 9.2 Hamilton Cycles 157

From our previous remarks, we can see that log10 n is not optimal. The proof
we give can be tightened, but probably not down to log n. The proof of Theorem
9.3 takes up the remainder of this section.
The proof of the upper bound is essentially the same as for Theorem 9.1.
After making X independent, there is no possibility of a Hamilton cycle.
The lower bound requires more work.

A pseudo-random condition
We say that a graph G = (V, E) with |V | = n is (n, α, p)-pseudo-random if
Q1 dG (v) ≥ 21 + α np for all v ∈ V (G).


2
Q2 eG (S) ≤ |S| log3 n for all S ⊆ V, |S| ≤ 10 log
p
n
.
α

log2 n
Q3 eG (S, T ) ≤ 1 + 4 |S||T |p for all disjoint S, T ⊆ V , |S|, |T | ≥ p .

Lemma 9.4 Let α be an arbitrary small positive constant. Suppose that p ≥

log10 n 1


n . Let H be a subgraph of G = Gn,p with maximum degree 2 − 3α np
and let G = G − H. Then w.h.p. G is (n, α, p)-pseudo-random.
Proof Q1: This follows from the fact that w.h.p. every vertex of Gn,p has
degree (1 + o(1))np, see Theorem 3.4(ii).
Q2: We show that this is true w.h.p. in G and hence in G. Indeed,

10 log2 n
 
P ∃S : eG (S) ≥ |S| log3 n and |S| ≤ ≤
p
10p−1 log2 n   s
 10p−1 log2 n  log3 n !s
n 2 s log3 n ne sep
∑ p ≤ ∑ ≤
s=log n s s log3 n s=log n s 2 log3 n
10p−1 log2 n  3 !s
5e log n

ne
∑ = o(1).
s=log n s log n

Q3: We show that this is true w.h.p. in G and hence in G. We first note that
the Chernoff bounds, Corollary 22.7, imply that
  α  2
P eG (S, T ) ≥ 1 + |S||T |p ≤ e−α |S||T |p/50 .
4
So,

log2 n
 
P ∃S, T : |S|, |T | ≥ and eG (S, T ) ≥ (1 + α)|S||T |p ≤
p
2
!s 2
!t
n n
ne1−α t p/100 ne1−α sp/100
  
n n −α 2 st p/50
∑ e ≤ ∑ ≤
−1 2 s t −1 2 s t
s,t=p log n s,t=p log n
158 Resilience
2 log2
!s 2 log2 n/100
!t
n
ne1−α n/100 ne1−α
∑ ≤
s t
s,t=p−1 log2 n

2 log2 n/100
!s 2
n
ne1−α  = o(1).
 ∑ s
s=p−1 log2 n

Pseudo-random implies Hamiltonian

The rest of this section is devoted to the proof that if G = ([n], E) is (n, α, p)-
pseudo-random then G is Hamiltonian.
We randomly partition [n] into sets let Vi , i = 1, 2, . . . , 5 such that
4 log3 n
 
αn
|V1 | = , |Vi | = , i = 2, 3, 4
p 5(1 + 2α)
so that
5 + 7α n 5 + 7α
|V5 | = n− 2
≈ n.
5 + 10α log n 5 + 10α
The number of neighbors of v in Vi is distributed as the binomial
Bin(dG (v), |Vi |/n). Thus for all v ∈ [n] we have, using our assumption that
np ≥ log10 n,
   
|Vi | 1 1 np
E(dVi (v)) = dG (v) ≥ + α |Vi |p ≥ +α  log n
n 2 2 log2 n
and so the Chernoff bounds imply that w.h.p.
 
1
dVi (v) & + α |Vi |p for all v ∈ [n] (9.1)
2
The proof now rests on two lemmas: the following quantities are fixed for the
remainder of the proof:
4 log3 n
 
` = 12 dlog ne + 3 and t = (9.2)
p
Lemma 9.5 [Connecting Lemma] Let {ai , bi } , i = 1, 2, . . . ,t be a family of
pairs of vertices from [n] with ai 6= a j and bi 6= b j for every distinct i, j ∈ [t],
(ai = bi is allowed). Let L = ti=1 {ai , bi }. Assume that K ⊆ [n] \ L is such that
S

C1 |K|  `t logt.
1


C2 For every v ∈ K ∪ L we have |dK (v)| & 2 + α p|K|.
 9.2 Hamilton Cycles 159

Then there exist t internally disjoint paths P1 , P2 , . . . , Pt such that Pi connects ai

to bi and V (Pi ) \ {ai , bi } ⊆ K. Furthermore, each path is of length `.

Lemma 9.6 [Absorbing Lemma] There is a path P∗ with V (P∗ ) ⊆ V2 ∪V3 ∪V4
such that for every W ⊆ V1 there is a path PW∗ such that V (PW∗ ) = V (P∗ ) ∪ W
and such that P∗ and PW∗ have the same endpoints.

With these two lemmas in hand, we can show that G is Hamiltonian. Let P∗
be as in Lemma 9.6 and let U = (V2 ∪V3 ∪V4 ∪V5 ) \V (P∗ ). If v ∈ U then
 
1
dU (v) ≥ dV5 (v) & + α |V5 |p
2
   
1 5 + 7α 1 3α
& +α np ≥ + |U|p. (9.3)
2 5 + 10α 2 4
5n
j k j k
Next let k = |U| log
n and s = logn5 n . Randomly choose disjoint sets
S1 , S2 , . . . , Sk ⊆ U of size s and let S = ki=1 Si and S0 = U \ S. It follows from
S

(9.3) and the Chernoff bounds and the fact that

 
|Si |dU (v) 1 3α
E(dSi (v)) = & + |Si |p  log n
|U| 2 4
that w.h.p.
 
1 α
dSi (v) ≥ + |Si |p for all i ∈ [k], v ∈ U. (9.4)
2 2
Claim 6 Assuming (9.4), we see that there is a perfect matching Mi between
Si , Si+1 for 1 ≤ i < k.

We prove the claim below. The matchings M1 , M2 , . . . , Mk−1 combine to give

us s vertex disjoint paths Qi from S1 to Sk for i = 1, 2, . . . , s. Together with a 0-
length path for each v ∈ |S0 | we have t 0 = s + |S0 | ≤ 2s internally disjoint paths
Qi from xi to yi , i = 1, 2, . . . ,t 0 that cover U. Now let t = t 0 + 1 and suppose
that xt , yt are the endpoints of P∗ . Applying Lemma 9.5 with K = V1 and ai =
yi , bi = xi+1 for i ∈ [t] we obtain a cycle C = (Q1 , P1 , Q2 , P2 , . . . , Qt 0 , P∗ , Pt )
(here xt+1 = x1 ) that covers V2 ∪ V3 ∪ V4 ∪ V5 , see Figure 9.1. Putting W =
V1 \ V (C) and using the fact that P∗ ⊆ C, we can use Lemma 9.6 to extend C
to a Hamilton cycle of G.
Proof of Claim 6
Fix i and now consider Hall’s theorem. We have to show that
|N(X, Si+1 )| ≥ |X| for X ⊆ Si .
Case 1: |X| ≤ 10p−1 log2 n. Let Y = N(X, Si+1 ) and suppose that |Y | < |X|.
160 Resilience

We now have
|X ∪Y | log5 n
 
1 α
eG (X ∪Y ) ≥ + |X||Si |p & ,
2 2 4
which contradicts Q2.
Case 2: 10p−1 log2 n ≤ |X| ≤ s/2. In this case we have
   
1 α 1 α
eG (X,Y ) ≥ + |Si ||X|p ≥ + |X||Y |p,
2 2 2 2
which contradicts Q3. (If |Y | < p−1 log2 n then we can add arbitrary vertices
from Si+1 \Y to Y so that we can apply Q3.)
The case |X| > s/2 is dealt with just as we did for |S| > n/2 in Theorem 6.1.
This completes the proof of Claim 6.
End of proof of Claim 6

Proof of Lemma 9.5

We begin with some lemmas on expansion. For sets X,Y and integer `, let
NG` (X,Y ) be the set of vertices y ∈ Y for which there exists x ∈ X and a path P
of length ` from x to y such that V (P) \ {x} ⊆ Y . We let NG (X,Y ) = NG1 (X,Y )
denote the set of neighbors of X in Y . The sets X,Y need not be disjoint in this
definition.
Lemma 9.7j Suppose that X,Y ⊆ [n] are (not necessarily disjoint) sets such
log2 n
k 3n
, |Y | ≥ 3 log and that |NG (x,Y )| ≥ 21 + α2 p|Y | for all


that |X| = p αp
x ∈ X. Then
 
1 α
|NG (X,Y )| ≥ + |Y |. (9.5)
2 3
Proof Let Z = X ∪ NG (X,Y ). We have
 
1 α
eG (Z) ≥ ∑ NG (x,Y ) − eG (X) ≥ + p|X||Y | − |X| log3 n ≥
x∈X 2 2
   
3 3 3 3 |X|
+ |X| log3 n = + log3 n · |Z|
2α 2 2α 2 |Z|
10 log2 n 3 3
If |Z| < p then we see that eG (Z) & 20α log n. This contradicts Q2 for
2
10 log n
small α and so we can assume that |Z| ≥ p and therefore |NG (X,Y )| ≥
9 log2 n
p . On the other hand, if (9.5) fails then from Q3 we have
 9.2 Hamilton Cycles 161

Q1
x1 y1

P1

Q2
x2 y2

P2

Qs
Pt xs ys

Pt 0 −1
xs+1 xt 0 = yt 0

Qt 0

P∗
xt yt

Figure 9.1 Cycle C = (Q1 , P1 , Q2 , P2 , . . . , Qt 0 , P∗ , Pt )

 
1 α
+ p|X||Y | ≤ eG (X,Y \ X) + 2eG (X) ≤
2 2
 
3
 α 1 α
eG (X,Y \ X) + 2|X| log n ≤ 1 + |X| + |Y |p + 2|X| log3 n,
4 2 3
contradiction.
Lemma 9.8 Suppose that X,Y ⊆ [n] are disjoint sets such that
3 log3 n
D1 |Y | ≥ αp .
2 log2 n
D2 |NG (X,Y )| ≥ p .
1 α

log2 n
D3 |NG (S,Y )| ≥ 2 + 4 |Y | for all S ⊆ Y, |S| ≥ p .
162 Resilience
1
Then there exists x ∈ X such that |NG` (x,Y )| ≥


2 + α8 |Y |
2 log2 n
Proof We first show that there exists x ∈ X such that |NG`−1 (x,Y )| ≥ p .
For this we use the following claim:
2 log2 n
Claim 7 Let i < ` and A ⊆ X be such that |NGi (A,Y )| ≥ p . Then there
2 log2 n
exists A0 ⊆ A such that |A0 | ≤ d|A|/2e and |NGi+1 (A0 ,Y )| ≥ p .

We prove the claim below. Using D2 and the claim ` − 2 times, we obtain a set
2
X 0 ⊆ X such that |X 0 | ≤ |X|/2`−2 and |NG`−1 (X 0 ,Y )| ≥ 2 logp n . But ` − 2 ≥
 

log2 n mand so we have |X 0 | = 1. Let X 0 = {x} and M ⊆ NG (x,Y ) be of size

l
log2 n
p .
By definition, there is a path Pw of length ` − 1 from x to each w ∈ M. Let
V ∗ = ( w∈M V (Pw )) \ {x}. Then D2 and D3 imply
S

   
1 α 1 α
|NG` (x,Y )| ≥ |NG (M,Y \V ∗ )| ≥ + |Y | − `|M| ≥ + |Y |.
2 4 2 8

Proof of Claim 7
First note that if A1 , A2 is a partition of A with |A1 | = d|A|/2e then we have
2 log2 n
|NGi (A1 ,Y )| + |NGi (A2 ,Y )| ≥ |NGi (A,Y )| ≥ .
p
We can assume therefore that there exists A0 ⊆ A, |A0 | ≤ d|A|/2e such that
2
|NGi (A0 ,Y )| ≥ logp n .
l 2 m
Choose B ⊆ NGi (A0 ,Y ), |B| = logp n . Then D3 implies that |NG (B,Y )| ≥ 21 + α4 |Y |.

Each v ∈ B is the endpoint of a path Pv of length i from a vertex in A0 . Let

V ∗ = v∈B V (Pv ). Then,
S

2 log2 n
 
i+1 0 ∗ 1 α
|NG (A ,Y )| ≥ |NG (B,Y )| − |V | ≥ + |Y | − `|B| ≥ .
2 4 p
End of proof of Claim 7
We are now ready to prove an approximate version of Lemma 9.5. Let t, ` be
as in (9.2).

Lemma 9.9 Let {ai , bi }, i = 1, 2, . . . ,t be a family of pairs of vertices from [n]

with ai 6= a j and bi 6= b j for every distinct i, j ∈ [t]. Furthermore, let RA , RB ⊆
[n] \ ti=1 {ai , bi } be disjoint and such that
S

48t`
E1 |RA |, |RB | ≥ α .
 9.2 Hamilton Cycles 163

E2 For Z = A, B, |NG (S, RZ )| ≥ 21 + α4 |RZ | for all S ⊆ RA ∪RB ∪ ti=1 {ai , bi }

S

log2 n
such that |S| ≥ p .

Then there exists a set I ⊆ [t], |I| = bt/2c and internally disjoint paths Pi , i ∈ I
such that Pi connects ai to bi and V (Pi ) \ {ai , bi } ⊆ RA ∪ RB .
Proof We prove this by induction. Assume that we have found s < bt/2c
paths Pi from ai to bi for i ∈ J ⊆ I, |J| = s. Then let
R0A = RA \ V (Pi ), R0B = RB \
[ [
V (Pi ).
i∈J i∈J
Choose hA , hB ≥ 2 log n so that hA + hB + 1 = `.
Claim 8 There exists i ∈ K = [t] \ J such that
   
1 α 1 α
|NGhA (ai , R0A )| ≥ + |R0A | and |NGhB (bi , R0B )| ≥ + |R0B |.
2 8 2 8
We verify Claim 8 below. Assume its truth for now. Let S = NGhA (ai , R0A ). Then
1 α 47t` log2 n
   
1 α
|S| ≥ + (RA − s`) ≥ + ≥ .
2 8 2 8 α p
Now from E2 we obtain

|NGhA +1 (ai , R0B )| ≥ |NG (S, R0B )| − ` ≥ |NG (S, RB )| − t`

     
1 α 1 α 1 α
≥ + |RB | − t` ≥ + |RB | ≥ + |R0B |.
2 4 2 8 2 8
Now from Claim 8 we have that |NGhB (bi , R0B )| ≥ 12 + α8 |R0B | and so

NGhA +1 (ai , R0B ) ∩ NGhB (bi , R0B ) 6= 0/ and there is a path as claimed.
It only remains to prove Claim 8. Assume inductively that
we have found
v1 , v2 , . . . , vk ∈ {ai : i ∈ K} such that |NGhA (vi , R0A )| ≥ 12 + 16α
|R0A | for i ∈ [k].
The base case is k = 0. We apply Lemma 9.8 with Y = R0A and X = {ai : i ∈ K}\
{v1 , v2 , . . . , vk }. We check that the lemma’s conditions are satisfied. |R0A | ≥
48t` 47` log3 n
α − t` ≥ αp and so D1 is satisfied. On the other hand E2 implies that if
2
S ⊆ R0A ∪ {ai : i ∈ K} is of size at least logp n then

log2 n
   
0 1 α 1 α
|NG (S, RA )| ≥ |NG (S, RA )|−t` ≥ + |RA |−t` ≥ + |RA |  .
2 4 2 8 p
2
So, D3 is satisfied and also D2 if |X| ≥ logp n i.e. if k ≤ t/2, completing the
induction. So, we obtain IA ⊆ [t], |IA | = bt/2c + 1 such that
 
0 1 α
hA
|N (ai , RA )| ≥ + |R0A | for i ∈ IA .
2 8
164 Resilience

A similar argument proves
the existence of IB ⊆ [t], |IB | = bt/2c + 1 such that
|N hB (bi , R0B )| ≥ 12 + α8 |R0B | for i ∈ IB and the claim follows, since IA ∩ IB 6= 0/
and we can therefore choose i ∈ IA ∩ IB .

Completing the proof of Lemma 9.5

We now define some parameters that will be used for the remainder of the
proof:
|K|
m = dlog2 te + 1, si = 2t, i ∈ [2m], s2m+1 = s2m+2 = , k = 2m + 2.
4
We randomly choose disjoint sets Si ⊆ K, |Si | = si , i ∈ [k]. The Chernoff bounds
and C2 imply that w.h.p.
 
1 α
|NG (v, Si )| ≥ + psi for v ∈ K ∪ L, i ∈ [k]. (9.6)
2 2
We prove the following lemma at the end of this section:
 
Lemma 9.10 Given sets [t] = I1 ⊇ I2 ⊇ · · · Im such that |I j | = |I j−1 |/2 , G
contains complete binary trees TA (i), TB (i), i ∈ [t] such that

F1 The depth of TA (i), TB (i), i ∈ Is is s − 1.

F2 TA (i) is rooted at ai and TB (i) is rooted at bi for i ∈ [t].
F3 The vertices TA (i, j) at depth j ∈ [0, m] in TA (i) are contained in S j .
F4 The vertices TB (i, j) at depth j ∈ [0, m] in TB (i) are contained in Sm+ j .
F5 The trees are vertex disjoint.

Assuming the truth of the lemma, we proceed as follows. We repeatedly use

Lemma 9.9 to find vertex disjoint paths. We first find bt/2c paths P1 of length
` from ai to bi for i ∈ J1 . We then let I2 = I1 \ J1 and construct the trees
TA (i), TB (i) for i ∈ I2 and then use Lemma 9.9 once more to find |I2 | vertex
disjoint paths Q2 of length ` − 2 from TA (i, 1) to TB (i, 1) for i ∈ I2 . We can now
select at least half of the Q2 to make d|I2 |/2e paths P2 from ai to bi for i ∈ J2
and then let I3 = I2 \ J2 . We repeat this process until we have constructed the
required set of paths.
We now check that Lemma 9.9 can be applied as claimed. To simplify notation
we use the convention that
j
[
TA (i, j) = TB (i, j) = 0/ for i ∈ Il .
l=1
 9.2 Hamilton Cycles 165

With this convention, we let

! !!
 t s−1
[ [ s−1
[ [
Ms = K (TA (i, j) ∪ TB (i, j)) ∪ V (Q)
i=1 j=1 i=1 Q∈Qi

and then in round s we apply Lemma 9.9 with ai , bi , i = 1, 2, . . . ,t replaced by

x j , y j , j = 1, 2, . . . ,t where the x j are made up of the TA (i, s) for i ∈ Is and the
y j are made up of the TB (i, s) for i ∈ Is and
RA = S2m+1 \ Ms and RB = S2m+2 \ Ms .
Thus
!
t s−1
|K| t |K| |K|
|RA | ≥
4
−O ∑ ∑ 2 j · 2 j ` = 4 − O(`t logt) ≥ 5  `t logt,
i=1 j=1

and similarly for |SB | and so E1 holds.

2
Now suppose that S ⊆ RA ∪ RB ∪ ti=1 x j , y j with |S| ≥ logp n . We can apply
S 

Lemma 9.7 with X = S and Y = RA because E1 holds and because of Assump-

tion C2 and (9.6). It follows that
 
1 α
|NG (S, RA )| ≥ + |RA |
2 3
and similarly for RB and so E2 holds. It now follows from Lemma 9.9 that
there are bt/2c indices i for which there is a path from x j to y j and these yield
at least dt/2s e indices Is and a path from ai to bi for i ∈ Is .
It remains only to prove Lemma 9.10.
Proof Let VA (s) denote the endpoints of Qs in Ss . Because of (9.6), we can
reduce
 this to the following: given a bipartite graph Γ with bipartion X =
x1 , x2 , . . . , xbt/2c ⊆
Ss \VA (s), B = Ss+1 = {y1 , y2 , . . . , y2t } and minimum de-
gree at least 12 + α2 psi+1 ≥ 4 log3 n and such that Q2, Q3 hold,  there exists
a
partition of B into t pairs {zi,1 , zi,2 } , i ∈ [t] such that both edges xi , zi,l , l =
1, 2 exist in Γ. (The reader can check that after each round, there are t/2 ver-
tices that are leaves of the current active trees, and need two neighbors to grow
the tree. We say that a tree is active if its root is not the endpoint of a path in
Qs .)
We need to verify the following condition:
S ⊆ A implies |NΓ (S, B)| ≥ 2|S|. (9.7)
2 log2 n
Case 1: |S| ≤ 3p .
2 log2 n
Let T = NΓ (S, B). If (9.7) fails then |S ∪ T | ≤ p and eG (S ∪ T ) > |S| log3 n,
violating Q2.
166 Resilience

sx x tx

s1x tx1 s2x tx2 six txi s2k

x tx2k

Figure 9.2 The absorber for k1 = 3. The cycle Cx is drawn with solid lines. The
dashed lines represent paths. The part inside the rectangle can be repeated to make
larger absorbers.

2
Case 2: |S| > 2 log n
3p .
If (9.7) fails then
 
1 α  α 2
+ si |S|p ≤ eG (S, T ) ≤ 2 1 + |S| p. (9.8)
2 2 4
The lower bound is from (9.6) and the upper bound in (9.8) is from Q3. Equa-
tion (9.8) is a contradiction, because si = 2t ≥ 4|S|.

Proof of Lemma 9.6

Let `x be an integer and Ax a graph with |V (Ax )| = `x + 1. Ax is called an ab-
sorber if there are vertices x, sx ,tx such that Ax contains paths Px , Px0 of lengths
/ V (Px0 ).
`x , `x − 1 from sx to tx such that x ∈

Let k, ` be integers and consider the graph Ax with 3 + 2k(` + 1) vertices con-
structed as follows:

S1 Ax contains a cycle Cx of length 4k + 3, the solid lines in Figure 9.2.

S2 Ax contains 2k pairwise disjoint six ,txi paths, P1 , P2 , . . . , P2k , each of which
is of length `, the dashed lines in Figure 9.2.

Lemma 9.11 Ax is an absorber for vertex x.

Proof We take

Px = sx , x, sx1 , P1 ,t1x , s2x , P2 , . . . ,tx2k ,tx .

Px0 = sx , s2x , P2 ,tx2 , s4x , P4 , . . . , s2k 2k 1 1 3 2k−1
x , P2k ,tx , sx , P1 ,tx , sx , P3 , . . . ,tx ,tx .
 9.3 The chromatic number 167

We first apply Lemma 9.5 to find Cx , x ∈ V1 . We let L = V1 and let ai = bi =

x, x ∈ V1 . We let K = V2 and k = 3 dlog ne so that ` = 4k + 3. The lemma is
αn
applicable as |K| = 5(1+2α)  `t logt and (9.1) implies that C2 holds.
We next apply Lemma 9.5 to connect ai = six to bi = txi by a path of length ` for
αn
i ∈ [2k], x ∈ V1 . We let K = V3 and note that |K| = 5(1+2α)  `t logt and that
(9.1) implies that C2 holds, once more.
At this point we have paths Px , Px0 for x ∈ V1 . We finally construct P∗ , us-
ing Lemma 9.5 to connect the paths Px0 , x ∈ V1 . We let t = h − 1 where V1 =
{x1 , x2 , . . . , xh }. We take ai = txi and bi = sxi+1 for i ∈ [h − 1] and K = V4 .
It is easy to see that this construction has the desired property. Where neces-
sary, we can absorb x ∈ V1 by replacing Px0 by Px .

9.3 The chromatic number

In this section we consider adding the edges of a fixed graph H to Gn,p . We
examine the case where p is constant and where ∆ = ∆(H) is sufficiently small.
We will see that under some circumstances, we can add quite a few edges
without increasing the chromatic number by very much.

Theorem 9.12 Suppose that H is a graph on vertex set [n] with maximum
degree ∆ = no(1) . Let p be constant and let G = Gn,p + H. Then w.h.p. χ(G) ≈
χ(Gn,p ), for all choices of H.

Proof We first observe that

n
W.h.p. every set of t ≤ vertices, span fewer than
10 log2 n
2npt
edges of Gn,p . (9.9)
log2 n
Indeed,

n/10 log2 n
  t

n 2
Pr(∃ a set negating (9.9)) ≤ ∑ 2
2 p2npt/ log n
t 2npt/ log n
t=2np/ log2 n
n/10 log2 n  ne t  t 2 ep log2 n 2npt/ log2 n
≤ ∑ ≤
t 4npt
t=2np/ log2 n
!t
n/10 log2 n 2n
ne  te 2np/ log
∑ · = o(1).
t 4n
t=2np/ log2 n
168 Resilience

We let s = 20∆ log2 n and randomly partition [n] into s sets V1 ,V2 , . . . ,Vs of size
n/s. Let Y denote the number of edges of H that have endpoints in the same
set of the partition. Then
E(Y ) ≤ |E(H)| s ≤ ∆n2s . It follows from the Markov
∆n 1
inequality that Pr Y ≥ s ≤ 2 and so there exists a partition for which Y ≤
∆n n
s = 20 log2 n . Furthermore, it follows from (7.21), that w.h.p. the subgraphs Gi
of Gn,p induced by each Vi have chromatic number ≈ 2 logn/s(n/s) .
b
Given V1 ,V2 , . . . ,Vs , we color G as follows: we color the edges of Gi with ≈
n/s n n
2 logb (n/s) colors, using different colors for each set and ≈ 2 logb (n/s) ≈ 2 logb (n) ≈
χ(Gn,p ) colors overall. We must of course deal with the at most Y edges that
could be improperly colored. Let W denote the endpoints of these edges. Then
n
|W | ≤ 10 log 2 n . It follows from (9.9) that we can write W = {w1 .w2 , . . . , wm }
2np
such that wi has at most log2 n
+ ∆ neighbors in {w1 , w2 , . . . , wi−1 } i.e. the col-
2np
oring number of the subgraph of Gn,p induced by W is at most log2 n
+ ∆. It
2np
follows that we can re-color the Y badly colored edges using at most log2 n
+
∆ + 1 = o(χ(Gn,p )) new colors.

9.4 Exercises
9.4.1 Prove that if p ≥ (1+η)n log n for a postive constant η then ∆C ≥ 12 − ε np,

where C denotes connectivity. (See Haller and Trujić [399].)

9.4.2 Show that for every ε > 0 there exists cε > 0 such that the following is
true w.h.p. If c ≥ cε and p = c/n and we remove any set of at most (1 −
ε)cn/2 edges from Gn,p , then the remaining graph contains a component
of size at least εn/2.

9.5 Notes
Sudakov and Vu [720] were the first to discuss local resilience in the context
of random graphs. Our examples are taken from this paper except that we have
given a proof of hamiltonicity that introduces the absorbing method.

Hamiltonicity
4
Sudakov and Vu proved local resilience for p ≥ logn n and ∆H = (1−o(1))np
2 . The
expression for ∆H is best posible, but the needed value for p has been lowered.
Frieze and Krivelevich [341] showed that there exist constants K, α such that
 9.5 Notes 169

w.h.p. ∆H ≥ αnp for p ≥ K log n

n . Ben-Shimon, Krivelevich and Sudakov [76]
1−ε
improved this to α ≥ 6 holds w.h.p. and then in [77] they obtained a result
on resilience for np − (log n + log log n) → ∞, but with K close to 13 . (Vertices
np
of degree less than 100 can lose all but two incident edges.) Lee and Sudakov
[533] proved the sought after result that for every positive ε there exists C =
C(ε) such that w.h.p. ∆H ≥ (1−ε)np 2 holds for p ≥ C log n
n . Condon, Espuny
Dı́az, Kim, Kühn and Osthus [198] refined [533]. Let H be a graph with degree
sequence d1 ≥ d2 ≥ · · · ≥ dn where di ≤ (n − i)p − εnp for i < n/2. They say
that G is ε-Pósa-resilient if G − H is Hamiltonian for all such H. Given ε > 0
there is a constant C = C(ε) such that if p ≥ C logn
n
then Gn,p is ε-Pósa-resilient
w.h.p. The result in [533] has now been improved to give a hitting time result,
see Montgomery [599] and Nenadov, Steger and Trujić [623]. The latter paper
also proves the optimal resilience of the 2-core when p = (1+ε) 3n
log n
.
Fischer, Škorić, Steger and Trujić [305] have shown that there exists C > 0 such
3n
that if p ≥ C nlog
1/2 then not only is there the square of a Hamilton cycle w.h.p.,
but containing a square is resilient to the deletion of not too many triangles
incident with each vertex.
Krivelevich, Lee and Sudakov [514] proved that G = Gn,p , p  n−1/2 remains
pancyclic w.h.p. if a subgraph H of maximum degree ( 12 − ε)np is deleted, i.e.
pancyclicity is locally resilient. The same is true for random regular graphs
when r  n1/2 .
Hefetz, Steger and Sudakov [413] began the study of the resilience of Hamil-
tonicity for random digraphs. They showed that if p  log n1/2
n
then w.h.p. the
1
Hamiltonicity of Dn,p is resilient to the deletion of up to ( 2 − o(1))np edges
8
incident with each vertex. The value of p was reduced to p  logn n by Ferber,
Nenadov, Noever, Peter and Škorić [301]. Finally, Montgomery [601] proved
that in the random digraph process, at the hitting time for Hamiltonicity, the
property is resilient w.h.p.
 P ART TWO

BASIC MODEL EXTENSIONS

 10
Inhomogeneous Graphs

Thus far we have concentrated on the properties of the random graphs Gn,m
and Gn,p . We first consider a generalisation of Gn,p where the probability of
edge (i, j) is pi j is not the same for all pairs i, j. We call this the generalized
binomial graph . Our main result on this model concerns the probability that
it is connected. For this model we concentrate on its degree sequence and the
existence of a giant component. After this we move onto a special case of this
model, viz. the expected degree model. Here pi j is proportional to wi w j for
weights wi . In this model, we prove results about the size of the largest com-
ponents. We finally consider another special case of the generalized binomial
graph, viz. the Kronecker random graph.

10.1 Generalized Binomial Graph

Consider the following natural generalisation of the binomial random graph
Gn,p , first considered by Kovalenko [511].
Let V = {1, 2, . . . , n} be the vertex set. The random graph Gn,P has vertex set
V and two vertices i and j from V , i 6= j, are joined by an edge with probability
pi j = pi j (n), independently of all other edges. Denote by
P = [pi j ]
the symmetric n × n matrix of edge probabilities, where pii = 0. Put qi j =
1 − pi j and for i, k ∈ {1, 2, . . . , n} define
n n
Qi = ∏ qi j , λn = ∑ Qi .
j=1 i=1

Note that Qi is the probability that vertex i is isolated and λn is the expected
number of isolated vertices. Next let
Rik = min qi j1 · · · qi jk .
1≤ j1 < j2 <···< jk ≤n

Suppose that the edge probabilities pi j are chosen in such a way that the fol-
lowing conditions are simultaneously satisfied as n → ∞:
max Qi → 0, (10.1)
1≤i≤n

173
174 Inhomogeneous Graphs

lim λn = λ = constant, (10.2)

n→∞

and
!k
n/2 n
1 Qi
lim ∑ ∑ Rik = eλ − 1. (10.3)
k=1 k!
n→∞
i=1

The next two theorems are due to Kovalenko [511].

We will first give the asymptotic distribution of the number of isolated vertices
in Gn,P , assuming that the above three conditions are satisfied. The next theo-
rem is a generalisation of the corresponding result for the classical model Gn,p
(see Theorem 3.1(ii)).

Theorem 10.1 Let X0 denote the number of isolated vertices in the random
graph Gn,P . If conditions (10.1) (10.2) and (10.3) hold, then

λ k −λ
lim P(X0 = k) = e
n→∞ k!
for k = 0, 1, . . ., i.e., the number of isolated vertices is asymptotically Poisson
distributed with mean λ .

Proof Let
(
1 with prob. pi j
Xi j =
0 with prob. qi j = 1 − pi j .

Denote by Xi , for i = 1, 2, . . . n, the indicator of the event that vertex i is isolated

in Gn,P . To show that X0 converges in distribution to the Poisson random vari-
able with mean λ one has to show (see Corollary 21.11) that for any natural
number k
!
λk
E ∑ Xi1 Xi2 · · · Xik → (10.4)
1≤i1 <i2 <...<i ≤n k!
k

as n → ∞. But

k

E Xi1 Xi2 · · · Xik = ∏ P Xir = 1|Xi1 = . . . = Xir−1 = 1 , (10.5)
r=1

where in the case of r = 1 we condition on the sure event.

Since the LHS of (10.4) is the sum of E Xi1 Xi2 · · · Xik over all i1 < · · · < ik ,
we need to find matching upper and lower bounds for this expectation. Now
 10.1 Generalized Binomial Graph 175


P Xir = 1|Xi1 = . . . = Xir−1 = 1 is the unconditional probability that ir is not
adjacent to any vertex j 6= i1 , . . . , ir−1 and so

∏nj=1 qir j
P Xir = 1|Xi1 = . . . = Xir−1 = 1 = r−1 .
∏s=1 qir is
Hence

Qir Qir
Qir ≤ P Xir = 1|Xi1 = . . . = Xir−1 = 1 ≤ ≤ .
Rir ,r−1 Rir k
It follows from (10.5) that

Qi Qi
Qi1 · · · Qik ≤ E Xi1 · · · Xik ≤ 1 · · · k . (10.6)
Ri1 k Rik k
Applying conditions (10.1) and (10.2) we get that

1
∑ Qi1 · · · Qik = ∑ Qi1 · · · Qik ≥
1≤i1 <···<ik ≤n k! 1≤i
1 6=···6=ir ≤n
!
1 k n
Qi1 · · · Qik − ∑ Q2i Qi1 · · · Qik−2
k! 1≤i1 ∑
,...,i ≤n k! i=1 ∑
1≤i1 ,...,ik−2 ≤n
k

λnk λk λk
≥ − (max Qi )λnk−1 = n − (max Qi )λnk−1 → , (10.7)
k! i k! i k!
as n → ∞.
Now,
n n
Qi
∑ Rik ≥ λn = ∑ Qi ,
i=1 i=1
 k
n/2
and if lim sup ∑ni=1 RQi > λ then lim sup ∑k=1 k!1 ∑ni=1 RQi > eλ − 1, which
ik ik
n→∞ n→∞
contradicts (10.3). It follows that
n
Qi
lim
n→∞
∑ Rik = λ .
i=1

Therefore
!k
n
1 Qi λk
∑ Qi1 · · · Qik ≤ ∑ Rik → .
1≤i1 <...<ik ≤n k! i=1 k!
as n → ∞.
Combining this with (10.7) gives us (10.4) and completes the proof of Theorem
10.1.
176 Inhomogeneous Graphs

One can check that the conditions of the theorem are satisfied when
log n + xi j
pi j = ,
n
where xi j ’s are uniformly bounded by a constant.
The next theorem shows that under certain circumstances, the random graph
Gn,P behaves in a similar way to Gn,p at the connectivity threshold.

Theorem 10.2 If the conditions (10.1), (10.2) and (10.3) hold, then
lim P(Gn,P is connected) = e−λ .
n→∞

Proof To prove the this we will show that if (10.1), (10.2) and (10.3) are
satisfied then w.h.p. Gn,P consists of X0 + 1 connected components, i.e., Gn,P
consists of a single giant component plus components that are isolated vertices
only. This, together with Theorem 10.1, implies the conclusion of Theorem
10.2.
Let U ⊆ V be a subset of the vertex set V . We say that U is closed if Xi j = 0
for every i and j, where i ∈ U and j ∈ V \ U. Furthermore, a closed set U is
called simple if either U or V \U consists of isolated vertices only. Denote the
number of non-empty closed sets in Gn,P by Y1 and the number of non-empty
simple sets by Y . Clearly Y1 ≥ Y .
We will prove first that

lim inf EY ≥ 2eλ − 1. (10.8)

n→∞

Denote the set of isolated vertices in Gn,P by J. If V \ J is not empty then

Y = 2X0 +1 − 1 (the number of non-empty subsets of J plus the number of their
complements, plus V itself). If V \ J = 0/ then Y = 2n − 1. Now, by Theorem
10.1, for every fixed k = 0, 1, . . . ,
λk
lim P(Y = 2k+1 − 1) = e−λ .
n→∞ k!
Observe that for any ` ≥ 0,
`
EY ≥ ∑ (2k+1 − 1) P(Y = 2k+1 − 1)
k=0

and hence
`
λ k e−λ
lim inf EY ≥
n→∞
∑ (2k+1 − 1) k!
.
k=0
 10.1 Generalized Binomial Graph 177

So,
`
λ k e−λ
lim inf EY ≥ lim
n→∞ `→∞
∑ (2k+1 − 1) k!
= 2eλ − 1
k=0

which completes the proof of (10.8).

We will show next that

lim sup EY1 ≤ 2eλ − 1. (10.9)

n→∞

To prove (10.9) denote by Zk the number of closed sets of order k in Gn,P so

that Y1 = ∑nk=1 Zk . Note that

Zk = ∑ Zi1 ...ik ,
i1 <...<ik

where Zi1 ,...ik indicates whether set Ik = {i1 . . . ik } is closed. Then

E Zi1 ,...ik = P(Xi j = 0, i ∈ Ik , j 6∈ Ik ) = ∏ qi j .

i∈Ik , j6∈Ik

Consider first the case when k ≤ n/2. Then

∏i∈Ik ,1≤ j≤n qi j Qi Qi
∏ qi j = =∏ ≤∏ .
i∈Ik , j6∈Ik ∏i∈Ik , j∈Ik qi j i∈Ik ∏ j∈Ik q i j R
i∈Ik ik

Hence
!k
n
Qi 1 Qi
E Zk ≤ ∑ ∏ ≤ ∑ Rik .
i1 <...<ik i∈Ik Rik k! i=1

Now, (10.3) implies that

n/2
lim sup ∑ E Zk ≤ eλ − 1.
n→∞ k=1

To complete the estimation of E Zk (and thus for EY1 ) consider the case when
k > n/2. For convenience let us switch k with n − k, i.e, consider E Zn−k , when
0 ≤ k < n/2. Notice that E Zn = 1 since V is closed. So for 1 ≤ k < n/2

E Zn−k = ∑ ∏ qi j .
i1 <...<ik i∈Ik , j6∈Ik

But qi j = q ji so, for such k, E Zn−k = E Zk . This gives

lim sup EY1 ≤ 2(eλ − 1) + 1,

n→∞

where the +1 comes from Zn = 1 This completes the proof of (10.9).

178 Inhomogeneous Graphs

Now,
P(Y1 > Y ) = P(Y1 −Y ≥ 1) ≤ E(Y1 −Y ).

Estimates (10.8) and (10.9) imply that

lim sup E(Y1 −Y ) ≤ 0,

n→∞

which in turn leads to the conclusion that

lim P(Y1 > Y ) = 0.

n→∞

i.e., asymptotically, the probability that there is a closed set that is not simple,
tends to zero as n → ∞. It is easy to check that X0 < n w.h.p. and therefore
Y = 2X0 +1 − 1 w.h.p. and so w.h.p. Y1 = 2X0 +1 − 1. If Gn,P has more than X0 + 1
connected components then the graph after removal of all isolated vertices
would contain at least one closed set, i.e., the number of closed sets would
be at least 2X0 +1 . But the probability of such an event tends to zero and the
theorem follows.
We finish this section by presenting a sufficient condition for Gn,P to be
connected w.h.p. as proven by Alon [22].

Theorem 10.3 For every positive constant b there exists a constant

c = c(b) > 0 so that if, for every non-trivial S ⊂ V ,

∑ pi j ≥ c log n,
i∈S, j∈V \S

then probability that Gn,P is connected is at least 1 − n−b .

Proof In fact Alon’s result is much stronger. He considers a random subgraph

G pe of a multi-graph G on n vertices, obtained by deleting each edge e inde-
pendently with probability 1 − pe . The random graph Gn,P is a special case of
G pe when G is the complete graph Kn . Therefore, following in his footsteps,
we will prove that Theorem 10.3 holds for G pe and thus for Gn,P .
So, let G = (V, E) be a loopless undirected multigraph on n vertices, with
probability pe , 0 ≤ pe ≤ 1 assigned to every edge e ∈ E and suppose that
for any non-trivial S ⊂ V the expectation of the number ES of edges in a cut
(S,V \ S) of G pe satisfies

E ES = ∑ pe ≥ c log n. (10.10)
e∈(S,V \S)

Create a new graph G0 = (V, E 0 ) from G by replacing each edge e by k =

 10.1 Generalized Binomial Graph 179

c log n parallel copies with the same endpoints and giving each copy e0 of e a
probability p0e0 = pe /k.
Observe that for S ⊂ V
E ES0 = ∑ p0e0 = E ES .
e0 ∈(S,V \S)

Moreover, for every edge e of G, the probability that no copy e0 of e survives

in a random subgraph G0p0 is (1 − pe /k)k ≥ 1 − pe and hence the probability
that G pe is connected exceeds the probability of G0p0 being connected, and so
e
in order to prove the theorem it suffice to prove that
P(G0p0e is connected) ≥ 1 − n−b . (10.11)
To prove this, let E10 ∪E20 ∪. . .∪Ek0 be a partition of the set E 0 of the edges of G0 ,
such that each Ei0 consists of a single copy of each edge of G. For i = 0, 1, . . . , k
define G0i as follows. G00 is the subgraph of G0 that has no edges, and for all
i ≥ 1, G0i is the random subgraph of G0 obtained from G0i−1 by adding to it
each edge e0 ∈ Ei0 independently, with probability p0e0 .
Let Ci be the number of connected components of G0i . Then we have C0 = n
and we have G0k ≡ G0p0 . Let us call the stage i, 1 ≤ i ≤ k, successful if either
e
Ci−1 = 1 (i.e., G0i−1 is connected) or if Ci < 0.9Ci−1 . We will prove that
1
P(Ci−1 = 1 or Ci < 0.9Ci−1 |G0i−1 ) ≥ . (10.12)
2
To see that (10.12) holds, note first that if G0i−1 is connected then there is
nothing to prove. Otherwise let Hi = (U, F) be the graph obtained from G0i−1
by (i) contracting every connected component of G0i−1 to a single vertex and (ii)
adding to it each edge e0 ∈ Ei0 independently, with probability p0e0 and throwing
away loops. Note that since for every nontrivial S, E ES0 ≥ k, we have that for
every vertex u ∈ U (connected component of G0i−1 ),
pe
∑0 p0e0 = ∑ c k ≥ 1.
u∈e ∈F e∈U:U

Moreover, the probability that a fixed vertex u ∈ U is isolated in Hi is

( )
∏0 (1 − p0e0 ) ≤ exp − ∑0 p0e0 ≤ e−1 .
u∈e ∈F u∈e ∈F

Hence the expected number of isolated vertices of Hi does not exceed |U|e−1 .
Therefore, by the Markov inequality, it is at most 2|U|e−1 with probability at
least 1/2. But in this case the number of connected components of Hi is at most
 
−1 1 −1 1 −1
2|U|e + (|U| − 2|U|e ) = +e |U| < 0.9|U|,
2 2
180 Inhomogeneous Graphs

and so (10.12) follows. Observe that if Ck > 1 then the total number of success-
ful stages is strictly less than log n/ log 0.9 < 10 log n. However, by (10.12), the
probability of this event is at most the probability that a Binomial random vari-
able with parameters k and 1/2 will attain a value at most 10 log n. It follows
from (22.22) that if k = c log n = (20 + t) log n then the probability that Ck > 1
2
(i.e., that G0p0 is disconnected) is at most n−t /4c . This completes the proof of
e
(10.11) and the theorem follows.

10.2 Expected Degree Model

In this section we will consider a special case of Kovalenko’s generalized bi-
nomial model, introduced by Chung and Lu in [183], where edge probabilities
pi j depend on weights assigned to vertices. This was also meant as a model for
“Real World networks”, see Chapter 18.
Let V = {1, 2, . . . , n} and let wi be the weight of vertex i. Now insert edges
between vertices i, j ∈ V independently with probability pi j defined as
n
wi w j
pi j = where W = ∑ wk .
W k=1

We assume that maxi w2i < W so that pi j ≤ 1. The resulting graph is denoted
as Gn,Pw . Note that putting wi = np for i ∈ [n] yields the random graph Gn,p .
Notice that loops are allowed here but we will ignore them in what follows.
Moreover, for vertex i ∈ V its expected degree is
wi w j
∑ = wi .
j W

Denote the average vertex weight by w (average expected vertex degree) i.e.,
W
w= ,
n
while, for any subset U of a vertex set V define the volume of U as

w(U) = ∑ wk .
k∈U

Chung and Lu in [183] and [185] proved the following results summarized
in the next theorem.

Theorem 10.4 The random graph Gn,Pw with a given expected degree se-
quence has a unique giant component w.h.p. if the average expected degree is
 10.2 Expected Degree Model 181

strictly greater than one (i.e., w > 1). Moreover, if w > 1 then w.h.p. the giant
component has volume
√ 
λ0W + O n(log n)3.5 ,

where λ0 is the unique nonzero root of the following equation

n n
∑ wi e−wi λ = (1 − λ ) ∑ wi ,
i=1 i=1

Furthermore w.h.p., the second-largest component has size at most

(1 + o(1)µ(w) log n,
where
(
1/(w − 1 − log w) if 1 < w < 2,
µ(w) =
1/(1 + log w − log 4) if w > 4/e.

Here we will prove a weaker and restricted version of the above theorem. In
the current context, a giant component is one with volume Ω(W ).

Theorem 10.5 If the average expected degree w > 4, then a random graph
Gn,Pw w.h.p. has a unique giant component and its volume is at least
 
2
1− √ W
ew
while the second-largest component w.h.p. has the size at most
log n
(1 + o(1)) .
1 + log w − log 4

The proof is based on a key lemma given below, proved under stronger con-
ditions on w than in fact Theorem 10.5 requires.

4
Lemma 10.6 For any positive ε < 1 and w > e(1−ε) 2 w.h.p. every connected
component in the random graph Gn,Pw either has volume at least εW or has at
log n
most 1+log w−log 4+2 log(1−ε) vertices.

Proof We first estimate the probability of the existence of a connected com-

ponent with k vertices (component of size k) in the random graph Gn,Pw . Let
S ⊆ V and suppose that vertices from S = {vi1 , vi2 , . . . , vik } have respective
weights wi1 , wi2 , . . . , wik . If the set S induces a connected subgraph of Gn,Pw
182 Inhomogeneous Graphs

than it contains at least one spanning tree T . The probability of such event
equals
P(T ) = ∏ wi j wil ρ,
{vi j ,vil }∈E(T )

where
1 1
ρ := = .
W nw
So, the probability that S induces a connected subgraph of our random graph
can be bounded from above by

∑ P(T ) = ∑ ∏ wi j wil ρ,
T T {vi ,vi }∈E(T )
j l

where T ranges over all spanning trees on S.

By the matrix-tree theorem (see West [745]) the above sum equals the determi-
nant of any k − 1 by k − 1 principal sub-matrix of (D − A)ρ, where A is defined
as
wi1 wi2 · · · wi1 wik
 
0
 wi wi
 2 1 0 · · · wi2 wik 
A= ,

.. .. .. ..
 . . . . 
wik wi1 wik wi2 · · · 0
while D is the diagonal matrix


D = diag wi1 (W − wi1 ), . . . , wik (W − wik ) .

(To evaluate the determinant of the first principal co-factor of D − A, delete

row and column k of D − A; Take out a factor wi1 wi2 · · · wik−1 ; Add the last k − 2
rows to row 1; Row 1 is now (wik , wik , . . . , wik ), so we can take out a factor wik ;
Now subtract column 1 from the remaining columns to get a (k − 1) × (k − 1)
upper triangular matrix with diagonal equal to diag(1, w(S), w(S), . . . , w(S))).
It follows that

∑ P(T ) = wi1 wi2 · · · wik w(S)k−2 ρ k−1 . (10.13)

T

To show that this subgraph is in fact a component one has to multiply by the
probability that there is no edge leaving S in Gn,Pw . Obviously, this probability
equals ∏vi ∈S,v j 6∈S (1 − wi w j ρ) and can be bounded from above

∏ (1 − wi w j ρ) ≤ e−ρw(S)(W −w(S)) . (10.14)

vi ∈S,v j ∈V \S
 10.2 Expected Degree Model 183

Let Xk be the number of components of size k in Gn,Pw . Then, using bounds

from (10.13) and (10.14) we get

E Xk ≤ ∑ w(S)k−2 ρ k−1 e−ρw(S)(W −w(S)) ∏ wi ,

S i∈S

where the sum ranges over all S ⊆ V, |S| = k. Now, we focus our attention on
k-vertex components whose volume is at most εW . We call such components
small or ε-small. So, if Yk is the number of small components of size k in Gn,Pw
then
EYk ≤ ∑ w(S)k−2 ρ k−1 e−w(S)(1−ε) ∏ wi = f (k). (10.15)
small S i∈S

Now, using the arithmetic-geometric mean inequality, we have

 k
w(S)
f (k) ≤ ∑ w(S)k−2 ρ k−1 e−w(S)(1−ε) .
small S k

The function x2k−2 e−x(1−ε) achieves its maximum at x = (2k − 2)/(1 − ε).
Therefore

2k − 2 2k−2 −(2k−2)
  k−1  
n ρ
f (k) ≤ e
k kk 1−ε
 ne k ρ k−1  2k − 2 2k−2
≤ e−(2k−2)
k kk 1−ε
2k
(nρ)k

2
≤ e−k
4ρ(k − 1)2 1 − ε
 k
1 4
=
4ρ(k − 1)2 ew(1 − ε)2
e−ak
= ,
4ρ(k − 1)2

where
a = 1 + log w − log 4 + 2 log(1 − ε) > 0

under the assumption of Lemma 10.6.

Let k0 = loga n . When k satisfies k0 < k < 2k0 we have
 
1 1
f (k) ≤ =o ,
4nρ(k − 1)2 log n
184 Inhomogeneous Graphs
2 log n
while, when a ≤ k ≤ n, we have
 
1 1
f (k) ≤ = o .
4n2 ρ(k − 1)2 n log n
So, the probability that there exists an ε-small component of size exceeding k0
is at most
   
log n 1 1
∑ f (k) ≤ × o + n × o = o(1).
k>k a log n n log n
0

This completes the proof of Lemma 10.6.

To prove Theorem 10.5 assume that for some fixed δ > 0 we have
4 2
w = 4+δ = 2
where ε = 1 − (10.16)
e (1 − ε) (ew)1/2

and suppose that w1 ≥ w2 ≥ · · · ≥ wn . We show next that there exists i0 ≥ n1/3

such that
v 
u
u 1+ δ W
t 8
wi0 ≥ . (10.17)
i0

Suppose the contrary, i.e., for all i ≥ n1/3 ,

v 
u
u 1+ δ W
t 8
wi < .
i
Then
v 
u
n
u 1+ δ W
8
W ≤ n1/3W 1/2 +
t
∑ i
i=n1/3
s 
δ
≤ n1/3W 1/2 + 2 1+ W n.
8
Hence
 
1/2 1/3 δ
W ≤n +2 1+ n1/2 .
8
This is a contradiction since for our choice of w

W = nw ≥ 4(1 + δ )n.

We have therefore verified the existence of i0 satisfying (10.17).

 10.2 Expected Degree Model 185

Now consider the subgraph G of Gn,Pw on the first i0 vertices. The proba-
bility that there is an edge between vertices vi and v j , for any i, j ≤ i0 , is at
least
1 + δ8
wi w j ρ ≥ w2i0 ρ ≥ .
i0
So the asymptotic behavior of G can be approximated by a random graph
Gn,p with n = i0 and p > 1/i0 . So, w.h.p. G has a component of size Θ(i0 ) =
Ω(n1/3 ). Applying Lemma 10.6 with ε as in (10.16) we see that any compo-
nent with size  log n has volume at least εW .
Finally, consider the volume of a giant component. Suppose first that there
exists a giant component of volume cW which is ε-small i.e. c ≤ ε. By Lemma
10.6, the size of the giant component is then at most 2log n
log 2 . Hence, there must
be at least one vertex with weight w greater than or equal to the average
2cW log 2
w≥ .
log n
But it implies that w2  W , which contradicts the general assumption that all
pi j < 1.
We now prove uniqueness in the same way that we proved the uniqueness
of the giant component in Gn,p . Choose η > 0 such that w(1 − η) > 4. Then
define w0i = (1 − η)wi and decompose

Gn,Pw = G1 ∪ G2
w0 w0
where the edge probability in G1 is p0i j = (1−η)W
i j
and the edge probability in G2
ww ηw w
is p00i j where 1 − Wi j
= (1 − p0i, j )(1 − p00i j ). Simple algebra gives p00i j ≥ Wi j . It
follows from the previous analysis that G1 contains between one and 1/ε giant
components. Let C1 ,C2 be two such components. The probability that there is
no G2 edge between them is at most
 
 ηwi w j  ηw(C1 )w(C2 )
∏ 1 − ≤ exp − ≤ e−ηW = o(1).
i∈C1 W W
j∈C2

As 1/ε < 4, this completes the proof of Theorem 10.5.

To add to the picture of the asymptotic behavior of the random graph Gn,Pw
we will present one more result from [183]. Denote by w2 the expected second-
order average degree, i.e.,
w2j
w2 = ∑ .
j W
186 Inhomogeneous Graphs

Notice that
∑ j w2j W
w2 = ≥ = w.
W n
Chung and Lu [183] proved the following.

Theorem 10.7 If the average expected square degree w2 < 1 then, with prob-
 2
w w2
ability at least 1 −  , all components of Gn,Pw have volume at most
C2 1−w2
√
C n.

Proof Let

x = P(∃S : w(S) ≥ Cn1/2 and S is a component).

Randomly choose two vertices u and v from V , each with probability propor-
tional to its weight. Then, for each vertex, the probability that it is in a set S
√ √
with w(S) ≥ C n is at least C nρ. Hence the probability that both vertices
are in the same component is at least
√
x(C nρ)2 = C2 xnρ 2 . (10.18)

On the other hand, for any two fixed vertices, say u and v, the probability
Pk (u, v) of u and v being connected via a path of length k + 1 can be bounded
from above as follows

Pk (u, v) ≤ ∑ (wu wi1 ρ)(wi1 wi2 ρ) · · · (wik wv ρ) ≤ wu wv ρ(w2 )k .

i1 ,i2 ,...,ik

So the probability that u and v belong to the same component is at most

n ∞
wu wv ρ
∑ Pk (u, v) ≤ ∑ wu wv ρ(w2 )k = 1 − w2 .
k=0 k=0

Recall that the probabilities of u and v being chosen from V are wu ρ and wv ρ,
respectively. so the probability that a random pair of vertices are in the same
component is at most
 2
wu wv ρ w2 ρ
∑ wu ρ wv ρ 1 − w2 = 1 − w2 .
u,v

Combining this with (10.18) we have

 2
w2 ρ
C2 xnρ 2 ≤ ,
1 − w2
 10.3 Kronecker Graphs 187

which implies
 2
w w2
x≤  ,
C2 1 − w2

and Theorem 10.7 follows.

10.3 Kronecker Graphs

Kronecker random graphs were introduced by Leskovec, Chakrabarti, Klein-
berg and Faloutsos [537] (see also [536]). It is meant as a model of “Real World
networks”, see Chapter 18. Here we consider a special case of this model of an
inhomogeneous random graph. To construct it we begin with a seed matrix
 
α β
P= ,
β γ

where 0 < α, β , γ < 1, and let P[k] be the kth Kronecker power of P. Here P[k]
is obtained from P[k−1] as in the diagram below:
 [k−1]
β P[k−1]

αP
P[k] =
β P[k−1] γP[k−1]
and so for example
α2 β2
 
αβ βα
αβ αγ β2 β γ
P[2] =  .
β α β2 γα γβ 
β2 βγ γβ γ2

Note that P[k] is symmetric and has size 2k × 2k .

We define a Kronecker random graph as a copy of Gn,P[k] for some k ≥ 1
and n = 2k . Thus each vertex is a binary string of length k, and between any
two such vertices (strings) u, v we put an edge independently with probability
pu,v = α uv γ (1−u)(1−v) β k−uv−(1−u)(1−v) ,
or equivalently
puv = α i β j γ k−i− j ,
where i is the number of positions t such that ut = vt = 1, j is the number of t
where ut 6= vt and hence k −i− j is the number of t that ut = vt = 0. We observe
that when α = β = γ then Gn,P[k] becomes Gn,p with n = 2k and p = α k .
188 Inhomogeneous Graphs

Connectivity
We will first examine, following Mahdian and Xu [562], conditions under
which is Gn,P[k] connected w.h.p.

Theorem 10.8 Suppose that α ≥ β ≥ γ. The random graph Gn,P[k] is con-

nected w.h.p. (for k → ∞) if and only if either (i) β + γ > 1 or (ii) α = β =
1, γ = 0.

Proof We show first that β + γ ≥ 1 is a necessary condition. Denote by 0 the

vertex with all 0’s. Then the expected degree of vertex 0 is
k  
k
∑ 0v ∑ j β j γ k− j = (β + γ)k = o(1), when β + γ < 1.
p =
v j=0

Thus in this case vertex 0 is isolated w.h.p.

Moreover, if β +γ = 1 and 0 < β < 1 then Gn,P[k] cannot be connected w.h.p.
since the probability that vertex 0 is isolated is bounded away from 0. Indeed,
0 < β < 1 implies that β j γ k− j ≤ ζ < 1, 0 ≤ j ≤ k for some absolute constant
ζ . Thus, using Lemma 22.1(b),
k  (k)
j
P(0 is isolated) = ∏(1 − p0v ) ≥ ∏ 1 − β j γ k− j
v j=0
k
j k− j
( )
k
j β γ
≥ exp − ∑ = e−1/ζ .
j=0 1−ζ

Now when α = β = 1, γ = 0, the vertex with all 1’s has degree n − 1 with
probability one and so Gn,P[k] will be connected w.h.p. in this case.
It remains to show that the condition β + γ > 1 is also sufficient. To show
that β + γ > 1 implies connectivity we will apply Theorem 10.3. Notice that
the expected degree of vertex 0, excluding its self-loop, given that β and γ are
constants independent of k and β + γ > 1, is
(β + γ)k − γ k ≥ 2c log n,
for some constant c > 0, which can be as large as needed.
Therefore the cut (0,V \ {0}) has weight at least 2c log n. Remove vertex 0
and consider any cut (S,V \ S). Then at least one side of the cut gets at least
half of the weight of vertex 0. Without loss of generality assume that it is S,
i.e.,
∑ p0u ≥ c log n.
u∈S
 10.3 Kronecker Graphs 189

Take any vertices u, v and note that puv ≥ pu0 because we have assumed that
α ≥ β ≥ γ. Therefore

∑ ∑ puv ≥ ∑ pu0 > c log n,

u∈S v∈V \S u∈S

and so the claim follows by Theorem 10.3.

To add to the picture of the structure of Gn,P[k] when β + γ > 1 we state

(without proof) the following result on the diameter of Gn,P[k] .

Theorem 10.9 If β + γ > 1 then w.h.p. Gn,P[k] has constant diameter.

Giant Component
We now consider when Gn,P[k] has a giant component (see Horn and Radcliffe
[421]).

Theorem 10.10 Gn,P[k] has a giant component of order Θ(n) w.h.p., if and
only if (α + β )(β + γ) > 1.

Proof We prove a weaker version of the Theorem 10.10, assuming that for
α ≥ β ≥ γ as in [562]. For the proof of the more general case, see [421].
We will show first that the above condition is necessary. We prove that if

(α + β )(β + γ) ≤ 1,

then w.h.p. Gn,P[k] has n − o(n) isolated vertices. First let

(α + β )(β + γ) = 1 − ε, ε > 0.

First consider those vertices with weight (counted as the number of 1’s in their
label) less than k/2 + k2/3 and let Xu be the degree of a vertex u with weight l
where l = 0, . . . , k. It is easily observed that

E Xu = (α + β )l (β + γ)k−l . (10.19)

Indeed, if for vertex v, i = i(v) is the number of bits that ur = vr = 1, r =

1, . . . , k and j = j(v) is the number of bits where ur = 0 and vr = 1, then the
probability of an edge between u and v equals

puv = α i β j+l−i γ k−l− j .

190 Inhomogeneous Graphs

Hence,
l k−l   
l k−l
E Xu = ∑ puv = ∑ ∑ α i β j+l−i γ k−l− j
v∈V i=0 j=0 i j
l
 k−l
l
=∑ α i β l−i ∑ β j γ k−l−l
i=0 i j=0

and (10.19) follows. So, if l < k/2 + k2/3 , then assuming that α ≥ β ≥ γ,
2/3 2/3
E Xu ≤(α + β )k/2+k (β + γ)k−(k/2+k )
 k2/3
k/2 α + β
= ((α + β )(β + γ))
β +γ
 k2/3
α +β
=(1 − ε)k/2
β +γ
=o(1). (10.20)

Suppose now that l ≥ k/2 + k2/3 and let Y be the number of 1’s in the label of
a randomly chosen vertex of Gn,P[k] . Since EY = k/2, the Chernoff bound (see
(22.26)) implies that
 
k 4/3 1/3
P Y ≥ +k 2/3
≤ e−k /(3k/2) ≤ e−k /2 = o(1).
2

Therefore, there are o(n) vertices with l ≥ k/2 + k2/3 . It then follows from
(10.20) that the expected number of non-isolated vertices in Gn,P[k] is o(n) and
the Markov inequality then implies that this number is o(n) w.h.p.
Next, when α + β = β + γ = 1, which implies that α = β = γ = 1/2, then
random graph Gn,P[k] is equivalent to Gn,p with p = 1/n and so by Theorem
2.21 it does not have a component of order n, w.h.p.
To prove that the condition (α + β )(β + γ) > 1 is sufficient we show that the
subgraph of Gn,P[k] induced by the vertices of H of weight l ≥ k/2 is connected
w.h.p. This will suffice as there are at least n/2 such vertices. Notice that for
any vertex u ∈ H its expected degree, by (10.19), is at least

((α + β )(β + γ))k/2  log n. (10.21)

We first show that for u ∈ V ,

1
∑ puv ≥ 4 ∑ puv . (10.22)
v∈H v∈V

For the given vertex u let l be the weight of u. For a vertex v let i(v) be
 10.3 Kronecker Graphs 191

the number of bits where ur = vr = 1, r = 1, . . . , k, while j(v) stands for the

number of bits where ur = 0 and vr = 1. Consider the partition

V \ H = S1 ∪ S2 ∪ S3 ,

where
S1 = {v : i(v) ≥ l/2, j(v) < (k − l)/2},

S2 = {v : i(v) < l/2, j(v) ≥ (k − l)/2},

S3 = {v : i(v) < l/2, j(v) < (k − l)/2}.

Next, take a vertex v ∈ S1 and turn it into v0 by flipping the bits of v which
correspond to 0’s of u. Surely, i(v0 ) = i(v) and

j(v0 ) ≥ (k − l)/2 > j(v).

Notice that the weight of v0 is at least k/2 and so v0 ∈ H. Notice also that
α ≥ β ≥ γ implies that puv0 ≥ puv . Different vertices v ∈ S1 map to different
v0 . Hence
∑ puv ≥ ∑ puv . (10.23)
v∈H v∈S1

The same bound (10.23) holds for S2 and S3 in place of S1 . To prove the same
relationship for S2 one has to flip the bits of v corresponding to 1’s in u, while
for S3 one has to flip all the bits of v. Adding up these bounds over the partition
of V \ H we get
∑ puv ≤ 3 ∑ puv
v∈V \H v∈H

and so the bound (10.22) follows.

Notice that combining (10.22) with the bound given in (10.21) we get that for
u ∈ H we have
∑ puv > 2c log n, (10.24)
v∈H

where c can be a large as needed.

We finish the proof by showing that a subgraph of Gn,P[k] induced by vertex
set H is connected w.h.p. For that we make use of Theorem 10.3. So, we will
show that for any cut (S, H \ S)

∑ ∑ puv ≥ 10 log n.
u∈S v∈H\S
192 Inhomogeneous Graphs

Without loss of generality assume that vertex 1 ∈ S. Equation (10.24) implies

that for any vertex u ∈ H either

∑ puv ≥ c log n, (10.25)

v∈S
or
∑ puv ≥ c log n. (10.26)
v∈H\S

If there is a vertex u such that (10.26) holds then since u ≤ 1 and α ≥ β ≥ γ,

∑ ∑ puv ≥ ∑ p1v ≥ ∑ puv > c log n.

u∈S v∈H\S v∈H\S v∈H\S

Otherwise, (10.25) is true for every vertex u ∈ H. Since at least one such vertex
is in H \ S, we have
∑ ∑ puv ≥ c log n,
u∈S v∈H\S

and the Theorem follows.

10.4 Exercises
10.4.1 Prove Theorem 10.3 (with c = 10) using the result of Karger and Stein
[474] that in any weighted graph on n vertices the number of r-minimal
cuts is O (2n)2r . (A cut (S,V \ S), S ⊆ V, in a weighted graph G is

called r-minimal if its weight, i.e., the sum of weights of the edges con-
necting S with V \ S, is at most r times the weight of minimal weighted
cut of G).
10.4.2 Suppose that the entries of an n × n symmetric matrix A are all non-
negative. Show that for any positive constants c1 , c2 , . . . , cn , the largest
eigenvalue λ (A) satisfies
!
1 n
λ (A) ≤ max ∑ c j ai, j .
1≤i≤n ci j=1

10.4.3 Let A be the adjacency matrix of Gn,Pw and for a fixed value of x let
(
wi wi > x
ci = .
x wi ≤ x
1
Let m = max {wi : i ∈ [n]}. Let Xi = ci ∑nj=1 c j ai, j . Show that
m 2
E Xi ≤ w2 + x and Var Xi ≤ w + x.
x
 10.5 Notes 193

10.4.4 Apply Theorem 22.11 with a suitable value of x to show that w.h.p.
λ (A) ≤ w2 + (6(m log n)1/2 (w2 + log n))1/2 + 3(m log n)1/2 .
10.4.5 Show that if w2 > m1/2 log n then w.h.p. λ (A) = (1 + o(1))w2 .
10.4.6 Suppose that 1 ≤ wi  W 1/2 for 1 ≤ i ≤ n and that wi w j w2  W log n.
Show that w.h.p. diameter(Gn,Pw ) ≤ 2.
10.4.7 Prove, by the Second Moment Method, that if α + β = β + γ = 1 then
w.h.p. the number Zd of the vertices of degree d in the random graph
Gn,P[k] , is concentrated around its mean, i.e., Zd = (1 + o(1)) E Zd .
10.4.8 Fix d ∈ N and let Zd denote the number of vertices of degree d in the
Kronecker random graph Gn,P[k] . Show that
k  
k (α + β )dw (β + γ)d(k−w)
EZd = (1 + o(1)) ∑ ×
w=0 w d!
 
× exp −(α + β )w (β + γ)k−w + o(1).

10.4.9 Depending on the configuration of the parameters 0 < α, β , γ < 1, show

that we have either
 k 
EZd = Θ (α + β )d + (β + γ)d ,

or
EZd = o(2k ).

10.5 Notes
General model of inhomogeneous random graph
The most general model of inhomogeneous random graph was introduced by
Bollobás, Janson and Riordan in their seminal paper [144]. They concentrate
on the study of the phase transition phenomenon of their random graphs, which
includes as special cases the models presented in this chapter as well as, among
others, Dubins’ model (see Kalikow and Weiss [469] and Durrett [265]), the
mean-field scale-free model (see Riordan [664]), the CHKNS model (see Call-
away, Hopcroft, Kleinberg, Newman and Strogatz [171]) and Turova’s model
(see [730], [731] and [732]).
The model of Bollobás, Janson and Riordan is an extension of one defined
by Söderberg [708]. The formal description of their model goes as follows.
Consider a ground space being a pair (S , µ), where S is a separable metric
space and µ is a Borel probability measure on S . Let V = (S , µ, (xn )n≥1 )
194 Inhomogeneous Graphs

be the vertex space, where (S , µ) is a ground space and (xn )n≥1 ) is a ran-
dom sequence (x1 , x2 , . . . , xn ) of n points of S satisfying the condition that for
every µ-continuity set A, A ⊆ S , |{i : xi ∈ A}|/n converges in probability to
µ(A). Finally, let κ be a kernel on the vertex space V (understood here as a
kernel on a ground space (S , µ)), i.e., a symmetric non-negative (Borel) mea-
surable function on S × S. Given the (random) sequence (x1 , x2 , . . . , xn ) we let
GV (n, κ) be the random graph GV (n, (pi j )) with pi j := min{κ(xi , x j )/n, 1}.
In other words, GV (n, κ) has n vertices and, given x1 , x2 , . . . , xn , an edge {i, j}
(with i 6= j) exists with probability pi j , independently for all other unordered
pairs {i, j}.
Bollobás, Janson and Riordan present in [144] a wide range of results de-
scribing various properties of the random graph GV (n, κ). They give a neces-
sary and sufficient condition for the existence of a giant component, show its
uniqueness and determine the asymptotic number of edges in the giant com-
ponent. They also study the stability of the component, i.e., they show that its
size does not change much if we add or delete a few edges. They also estab-
lish bounds on the size of small components, the asymptotic distribution of the
number of vertices of given degree and study the distances between vertices
(diameter). Finally they turn their attention to the phase transition of GV (n, κ)
where the giant component first emerges.
Janson and Riordan [443] study the susceptibility, i.e., the mean size of the
component containing a random vertex, in a general model of inhomogeneous
random graphs. They relate the susceptibility to a quantity associated to a cor-
responding branching process, and study both quantities in various examples.
Devroye and Fraiman [241] find conditions for the connectivity of inhomo-
geneous random graphs with intermediate density. They draw n independent
points Xi from a general distribution on a separable metric space, and let their
indices form the vertex set of a graph. An edge i j is added with probability
min{1, κ(Xi , X j ) log n/n}, where κ > 0 is a fixed kernel. They show that, un-
der reasonably weak assumptions, the connectivity threshold of the model can
be determined.
Lin and Reinert [540] show via a multivariate normal and a Poisson process
approximation that, for graphs which have independent edges, with a possibly
inhomogeneous distribution, only when the degrees are large can we reason-
ably approximate the joint counts for the number of vertices with given degrees
as independent (note that in a random graph, such counts will typically be de-
pendent). The proofs are based on Stein’s method and the Stein–Chen method
(see Chapter 21.3) with a new size-biased coupling for such inhomogeneous
random graphs.
 10.5 Notes 195

Rank one model

An important special case of the general model of Bollobás, Janson and Rior-
dan is the so called rank one model, where the kernel κ has the form κ(x, y) =
ψ(x)ψ(y), for some function ψ > 0 on S . In particular, this model includes
the Chung-Lu model (expected degree model) discussed earlier in this Chapter.
Recall that in their approach we attach edges (independently) with probabili-
ties
nw w o n
i j
pi j = min , 1 where W = ∑ wk .
W k=1

Similarly, Britton, Deijfen and Martin-Löf [162] define edge probabilities as

wi w j
pi j = ,
W + wi w j
while Norros and Reittu [626] attach edges with probabilities
 ww 
i j
pi j = exp − .
W
For those models several characteristics are studied, such as the size of the gi-
ant component ([184], [185] and [626]) and its volume ([184]) as well as spec-
tral properties ([186] and [187]). It should be also mentioned here that Janson
[436] established conditions under which all those models are asymptotically
equivalent.
Recently, van der Hofstad [417], Bhamidi, van der Hofstad and
Hooghiemstra[88], van der Hofstad, Kliem and van Leeuwaarden [419] and
Bhamidi, Sen and Wang [89] undertake systematic and detailed studies of var-
ious aspects of the rank one model in its general setting.
Finally, consider random dot product graphs (see Young and Scheinerman
[756]) where to each vertex a vector in Rd is assigned and we allow each
edge to be present with probability proportional to the inner product of the
vectors assigned to its endpoints. The paper [756] treats these as models of
social networks.

Kronecker Random Graph

Radcliffe and Young [659] analysed the connectivity and the size of the giant
component in a generalized version of the Kronecker random graph. Their
results imply that the threshold for connectivity in Gn,P[k] is β + γ = 1. Tabor
[724] proved that it is also the threshold for a k-factor. Kang, Karoński, Koch
and Makai [471] studied the asymptotic distribution of small subgraphs (trees
and cycles) in Gn,P[k] .
196 Inhomogeneous Graphs

Leskovec, Chakrabarti, Kleinberg and Faloutsos [538] and [539] have shown
empirically that Kronecker random graphs resemble several real world net-
works. Later, Leskovec, Chakrabarti, Kleinberg, Faloutsos and Ghahramani
[539] fitted the model to several real world networks such as the Internet, cita-
tion graphs and online social networks.
The R-MAT model, introduced by Chakrabarti, Zhan and Faloutsos [175],
is closely related to the Kronecker random graph. The vertex set of this model
is also Zn2 and one also has parameters α, β , γ. However, in this case one needs
the additional condition that α + 2β + γ = 1.
The process of generating a random graph in the R-MAT model creates a
multigraph with m edges and then merges the multiple edges. The advantage of
the R-MAT model over the random Kronecker graph is that it can be generated
significantly faster when m is small. The degree sequence of this model has
been studied by Groër, Sullivan and Poole [393] and by Seshadhri, Pinar and
Kolda [698] when m = Θ(2n ), i.e. the number of edges is linear in the number
of vertices. They have shown, as in Kang, Karoński, Koch and Makai [471]
for Gn,P[k] , that the degree sequence of the model does not follow a power law
distribution. However, no rigorous proof exists for the equivalence of the two
models and in the Kronecker random graph there is no restriction on the sum
of the values of α, β , γ.
Further extensions of Kronecker random graphs can be found [109] and
[110].
 11
Fixed Degree Sequence

The graph Gn,m is chosen uniformly at random from the set of graphs with
vertex set [n] and m edges. It is of great interest to refine this model so that
all the graphs chosen have a fixed degree sequence d = (d1 , d2 , . . . , dn ). Of
particular interest is the case where d1 = d2 = · · · = dn = r, i.e., the graph
chosen is a uniformly random r-regular graph. It is not obvious how to do
this and this is the subject of the current chapter. We discuss the configuration
model in the next section and show its usefulness in (i) estimating the number
of graphs with a given degree sequence and (ii) showing that w.h.p. random
d-regular graphs are connected w.h.p., for 3 ≤ d = O(1).
We finish by showing in Section 11.5 how for large r, Gn,m can be embedded
in a random r-regular graph. This allows one to extend some results for Gn,m
to the regular case.

11.1 Configuration Model

Let d = (d1 , d2 , . . . , dn ) where d1 + d2 + · · · + dn = 2m is even. Let

Gn,d = {simple graphs with vertex set [n] s.t. degree d(i) = di , i ∈ [n]}

and let Gn,d be chosen randomly from Gn,d . We assume that

n
d1 , d2 , . . . , dn ≥ 1 and ∑ di (di − 1) = Ω(n).
i=1

We describe a generative model of Gn,d due to Bollobás [123]. It is referred

to as the configuration model. Let W1 ,W2 , . . . ,Wn be a partition of a set of
points W , where |Wi | = di for 1 ≤ i ≤ n and call the Wi ’s cells. We will assume
some total order < on W and that x < y if x ∈ Wi , y ∈ W j where i < j. For
x ∈ W define ϕ(x) by x ∈ Wϕ(x) . Let F be a partition of W into m pairs (a
configuration). Given F we define the (multi)graph γ(F) as

γ(F) = ([n], {(ϕ(x), ϕ(y)) : (x, y) ∈ F}).

Let us consider the following example of γ(F). Let n = 8 and d1 = 4, d2 =

3, d3 = 4, d4 = 2, d5 = 1, d6 = 4, d7 = 4, d8 = 2. The accompanying diagrams,

197
198 Fixed Degree Sequence

11
00
00
11 11
00
00
11 11
00
00
11 11
00
00
11
00
11 00
11 00
11 00
11

11
00
00
11 11
00
00
11 11
00
00
11 11
00
00
11 11
00
00
11
00
11 00
11 00
11 00
11 00
11

00
11 00
11 00
11 00
11 00
11 00
11 00
11
11
00 11
00 11
00 11
00 11
00 11
00 11
00
00
11 00
11 00
11 00
11 00
11 00
11 00
11

11
00 11
00 11
00 11
00 11
00 11
00 11
00 11
00
00
11 00
11 00
11 00
11 00
11 00
11 00
11 00
11
00
11 00
11 00
11 00
11 00
11 00
11 00
11 00
11

1 2 3 4 5 6 7 8

Figure 11.1 Partition of W into cells W1 , . . . ,W8 .

00
11 00
11 00
11 00
11
11
00
00
11 11
00
00
11 11
00
00
11 11
00
00
11

00
11 00
11 00
11 00
11 00
11
11
00
00
11 11
00
00
11 11
00
00
11 11
00
00
11 11
00
00
11

00
11 00
11 00
11 00
11 00
11 00
11 00
11
11
00
00
11 11
00
00
11 11
00
00
11 11
00
00
11 11
00
00
11 11
00
00
11 11
00
00
11

00
11 00
11 00
11 00
11 00
11 00
11 00
11 11
00
11
00 11
00 11
00 11
00 11
00 11
00 11
00 00
11
00
11
00
11 00
11 00
11 00
11 00
11 00
11 00
11

1 2 3 4 5 6 7 8

Figure 11.2 A partition F of W into m = 12 pairs

Figures 11.1, 11.2, 11.3 show a partition of W into W1 , . . . ,W8 , a configuration

and its corresponding multi-graph.
Denote by Ω the set of all configurations defined above for d1 + · · · + dn =
2m and notice that
(2m)!
|Ω| = = (2m)!! . (11.1)
m!2m

To see this, take di “distinct” copies of i for i = 1, 2, . . . , n and take a permuta-

 11.1 Configuration Model 199

2 11
00
11
00
00
11
3
11
00
00
11
00
11

111
00 4
00
11
00
11 11
00
00
11
00
11

00
11
11
00 11
00
00
11 00
11
8 00
11 5

00
11 11
00
00
11
11
00 00
11
00
11
7 6

Figure 11.3 Graph γ(F)

tion σ1 , σ2 , . . . , σ2m of these 2m symbols. Read off F, pair by pair {σ2i−1 , σ2i }
for i = 1, 2, . . . , m. Each distinct F arises in m!2m ways.
We can also give an algorithmic, construction of a random element F of the
family Ω.

Algorithm F-GENERATOR
begin
U ←− W, F ←− 0/
for t = 1, 2, . . . , m do
begin
Choose x arbitrarily from U;
Choose y randomly from U \ {x};
F ←− F ∪ {(x, y)};
U ←− U \ {(x, y)}
end
end

Note that F arises with probability 1/[(2m − 1)(2m − 3) · · · 1] = |Ω|−1 .

Observe that the following relationship between a simple graph G ∈ Gn,d

and the number of configurations F for which γ(F) = G.
200 Fixed Degree Sequence

Lemma 11.1 If G ∈ Gn,d , then

n
|γ −1 (G)| = ∏ di ! .
i=1

Proof Arrange the edges of G in lexicographic order. Now go through the

sequence of 2m symbols, replacing each i by a new member of Wi . We obtain
all F for which γ(F) = G.
The above lemma implies that we can use random configurations to “ap-
proximate” random graphs with a given degree sequence.

Corollary 11.2 If F is chosen uniformly at random from the set of all config-
urations Ω and G1 , G2 ∈ Gn,d then

P(γ(F) = G1 ) = P(γ(F) = G2 ).

So instead of sampling from the family Gn,d and counting graphs with a
given property, we can choose a random F and accept γ(F) iff there are no
loops or multiple edges, i.e. iff γ(F) is a simple graph.
This is only a useful exercise if γ(F) is simple with sufficiently high proba-
bility. We will assume for the remainder of this section that

∆ = max{d1 , d2 , . . . , dn } ≤ nα , α < 1/7.

We will prove later (see Lemma 11.7 and Corollary 11.8) that if F is chosen
uniformly (at random) from Ω,

P(γ(F) is simple) = (1 + o(1))e−λ (λ +1) , (11.2)

where
∑ di (di − 1)
λ= .
2 ∑ di
Hence, (11.1) and (11.2) will tell us not only how large is Gn,d , (Theorem
11.5) but also lead to the following conclusion.

Theorem 11.3 Suppose that ∆ ≤ nα , α < 1/7. For any (multi)graph property
P
P(Gn,d ∈ P) ≤ (1 + o(1))eλ (λ +1) P(γ(F) ∈ P),

The above statement is particularly useful if λ = O(1), e.g., for random r-

regular graphs, where r is a constant, since then λ = r−1
2 . In the next section
 11.1 Configuration Model 201

we will apply the above result to establish the connectedness of random regular
graphs.
Before proving (11.2) for ∆ ≤ nα , we feel it useful to give a simpler proof
for the case of ∆ = O(1).

Lemma 11.4 If ∆ = O(1) then (11.2) holds.

Proof Let L denote the number of loops and let D denote the number of non-
adjacent double edges in γ(F). Lemma 11.6 below shows that w.h.p. there are
no adjacent double edges. We first estimate that for k = O(1),
 
L di (di − 1)
E( = ∑ ∏ (11.3)
k S⊆[n] i∈S
4m − O(1)
|S|=k
!k
n
∆4
 
1 di (di − 1)
= ∑ 4m +O
k! i=1 m
λk
≈ .
k!
Explanation for (11.3): We assume that F-Generator begins with pairing up
points in S. Therefore the random choice here is always from a set of size
2m − O(1).
It follows from Theorem 21.11 that L is asymptotically Poisson and hence
that

Pr(L = 0) ≈ e−λ . (11.4)

We now show that D is also asymptotically Poisson and asymptotically in-

dependent of L. So, let k = O(1). If Dk denotes the set of collections of 2k
configuration points making up k double edges, then
  
D 
E L = 0 = ∑ Pr(Dk ⊆ F | L = 0)
k  D k

Pr(L = 0 | Dk ⊆ F) Pr(Dk ⊆ F)
=∑ .
Dk Pr(L = 0)

Now because k = O(1), we see that the calculations that give us (11.4) will
give us Pr(L = 0 | Dk ⊆ F) ≈ Pr(L = 0). So,
  
D 
E L = 0 ≈ ∑ Pr(Dk ⊆ F)
k  D k
202 Fixed Degree Sequence
di
dϕ(i)

1 2 2 2
= ∑ ∑ ∏ (2m − O(1))2
2 S,T ⊆[n] ϕ:S→T i∈S
|S|=|T |=k
S∩T =0/
!2k
n
∆8
 
1 di (di − 1)
= ∑ 4m +O
k! i=1 m
λ 2k
≈ .
k!
It follows from Theorem 21.11 that
2
Pr(D = 0 | L = 0) ≈ e−λ (11.5)
and the lemma follows from (11.4) and (11.5).
Bender and Canfield [71] gave an asymptotic formula for |Gn,d | when ∆ =
O(1). The paper [123] by Bollobás gives the same asymptotic formula when
∆ < (2 log n)1/2 . The following theorem allows for some more growth in ∆.
Its proof uses the notion of switching. Switchings were introduced by McKay
[581] and McKay and Wormald [582] and independently by Frieze [330], The
bound α < 1/7 is not optimal. For example, α < 1/2 in [582].

Theorem 11.5 Suppose that ∆ ≤ nα , α < 1/7.

(2m)!!
|Gn,d | ≈ e−λ (λ +1) .
∏ni=1 di !

In preparation we first prove

Lemma 11.6 Suppose that ∆ ≤ nα where α < 1/7. Let F be chosen uniformly
(at random) from Ω. Then w.h.p. γ(F) has
(a) No double loops.
(b) At most ∆ log n loops.
(c) No adjacent double edges.
(d) No triple edges.
(e) At most ∆2 log n double edges.
(f) No vertex incident to a loop and a double edge.
(g) There are at most ∆3 log n triangles.
(h) No vertex is adjacent to two distinct vertices that have loops.
Proof. We will use the following inequality repeatedly.
Let fi = {xi , yi }, i = 1, 2, . . . , k be k pairwise disjoint pairs of points. Then,
1
P( fi ∈ F, i = 1, 2, . . . , k) ≤ . (11.6)
(2m − 2k)k
 11.1 Configuration Model 203

This follows immediately from

1
P( fi ∈ F | f1 , f2 , . . . , fi−1 ∈ F) = .
2m − 2i + 1
This follows from considering Algorithm F-GENERATOR with x = xi and
y = yi in the main loop.
(a) Using (11.6) we obtain

P(F contains a pair of double loops) ≤

n  2  2
di 1 ∆4 n
∑ 2 ≤ = o(1).
i=1 2m − 8 (2m − 8)2
(b) Let k1 = ∆ log n.
P(F has at least k1 loops)
n   xi
di 1
≤ o(1) + ∑ ∏ 2 2m − 2k1 · (11.7)
x +···+xn =k , i=1
1 1
xi =0,1
 k1 n
∆
≤ o(1) +
2m ∑ ∏ dixi
x1 +···+xn =k1 , i=1
xi =0,1

∆ k1 (d1 + · · · + dn )k1
 
≤ o(1) +
2m k1 !
 k1
∆e
≤ o(1) +
k1
= o(1).
The o(1) term in (11.7) accounts for the probability of having a double loop.
(c)
n  2  2
di ∆
P(F contains two adjacent double edges) ≤ ∑ ≤
i=1 2 2m − 8
∆5 m
= o(1).
(2m − 8)2
(d)
    3
di dj 1
P(F contains a triple edge) ≤ ∑ 6 ≤
1≤i< j≤n 3 3 2m − 6
∆5 m
= o(1).
(2m − 6)3
204 Fixed Degree Sequence

(e) Let k2 = ∆2 log n.

P(F has at least k2 double edges)

n   xi
di ∆
≤ o(1) + ∑ ∏ 2 2m − 4k2 · (11.8)
x +···+xn =k , i=1
1 2
xi =0,1
k2 n
∆2

≤ o(1) +
m ∑ ∏ dixi
x1 +···+xn =k2 , i=1
xi =0,1
k2
∆2 (d1 + · · · + dn )k2

≤ o(1) +
m k2 !
k2
2∆2 e

≤ o(1) +
k2
= o(1).

The o(1) term in (11.8) accounts for adjacent multiple edges and triple edges.
The ∆/(2m − 4k2
) term can be justified as follows: We have chosen two points
x1 , x2 in Wa in d2i ways and this term bounds the probability that x2 chooses a
partner in the same cell as x1 .
(f)

P(∃ vertex v incident to a loop and a multiple edge

n  2
di 1 ∆
≤∑
i=1 2 2m − 1 2m −5
∆4 m
≤
(2m − 1)(2m − 5)
= o(1).

(g) Let X denote the number of triangles in F. Then

n di
2
∆ ∆3 m
E(F) ≤ ∑ 2 ≤ ≤ ∆3 .
i=1 2m − 4 2m − 4
Now use the Markov inequality.
(h) The probability that there is a vertex adjacent to two loops is at most
!2  4
n
1 n ∆ M(∆M)2 ∆4
∑ di 2 ∑ di (di − 1) M1 − O(1)
≤
(M − O(1))4
= o(1).
i=1 i=1

Let now Ωi, j be the set of all F ∈ Ω such that F has i loops; j double edges,
 11.1 Configuration Model 205

at most ∆3 log n triangles and no double loops or triple edges and no vertex
incident with two double edges or with a loop and a multiple edge.
Lemma 11.7 (Switching Lemma) Suppose that ∆ ≤ nα , α < 1/7. Let M1 =
2m and M2 = ∑i di (di − 1). For i ≤ k1 + 2k2 and j ≤ k2 , where k1 = ∆ log n and
k2 = ∆2 log n,
|Ωi+2, j−1 | j
= ,
|Ωi, j | (i + 2)(i + 1)
and
  5 
|Ωi−1,0 | 2iM1 ∆ log n
= 1+O .
|Ωi,0 | M2 M1
The corollary that follows is an immediate consequence of the Switching
Lemma. It immediately implies Theorem 11.5.
Corollary 11.8 Suppose that ∆ ≤ nα where α < 1/7. Then,
|Ω0,0 |
= (1 + o(1))e−λ (λ +1) ,
|Ω|
where
M2
λ= .
2M1

Proof It follows from the Switching Lemma that i ≤ k1 and j ≤ k2 implies

 3  i+2 j
λ i+2 j

|Ωi, j | ∆ k2 λ
= 1 + Õ = (1 + o(1)) .
|Ω0,0 | n i! j! i! j!
Lemma 11.6 implies that
k1 k2
λ i+2 j
(1 − o(1))|Ω| = (1 + o(1))|Ω0,0 | ∑ ∑
i=0 j=0 i! j!

= (1 + o(1))|Ω0,0 |eλ (λ +1) .

To prove the Switching Lemma we need to introduce two specific operations

on configurations, called a “d-switch” and an “`-switch”.
Figure 11.4 illustrates the “double edge removal switch” (“d-switch”) opera-
tion. Here we have four points x1 , x2 , x3 , x4 and a double edge associated with
the pairs {x1 , x3 }, {x2 , x4 } ∈ F where x1 , x2 are in cell Wa and x3 , x4 are in cell
Wb . The d-switch operation replaces these pairs by a new set of pairs: {x1 , x2 },
206 Fixed Degree Sequence

Wa Wb Wa Wb
x1 x3 x1 x3

x2 x4 x2 x4

Figure 11.4 d-switch

{x3 , x4 }. This replaces a multiple edge by two loops and no other multiple
edges are created.
In general, a forward d-switch operation takes F, a member of Ωi, j , to F 0 ,
a member Ωi+2, j−1 , see Figure 11.4. A reverse d-switch operation takes F 0 , a
member of Ωi+2, j−1 , to F 0 , a member Ωi, j . The number of choices η f for a
forward d-switch is j and the number of choices ηr for a reverse d-switch is
(i + 2)(i + 1).
Now for F ∈ Ωi, j let dL (F) = j denote the number of F 0 ∈ Ωi+2, j−1 that can
be obtained from F by an d-switch. Similarly, for F 0 ∈ Ωi+2, j−1 let dR (F 0 ) =
(i + 1)(i + 2) denote the number of F ∈ Ωi, j that can be transformed into F 0 by
an d-switch. Then,

∑ dL (F) = ∑ dR (F 0 ).
F∈Ωi, j F 0 ∈Ωi+2, j−1

So,
|Ωi+2, j−1 | j
= ,
|Ωi, j | (i + 1)(i + 2)
which shows that the first statement of the Switching Lemma holds.

Now consider the second operation on configurations, described as a “loop re-

moval switch”(“`-switch”), Figure 11.5. Here we have four points x1 , x2 , x3 , x4
from three different cells, where x1 and x2 are in cell Wa , x3 is in cell Wb and
x4 is in cell Wc . {x1 , x2 } ∈ F forms a loop and {x3 , x4 } ∈ F. The `-switch oper-
ation replaces these pairs by new pairs: {x1 , x3 }, {x2 , x4 } or {x1 , x4 }, {x2 , x3 }
if in these operations no double edge is created.
We estimate the number of choices η during an `-switch of F ∈ Ωi,0 . For a
forward switching operation

i M1 − 2∆2 − 2i ≤ η ≤ iM1 ,


(11.9)
 11.1 Configuration Model 207
Wb

x3 x3
Wa
x1 x1

x2 x2
x4 x4

Wc

Figure 11.5 `-switch

while, for the reverse procedure,

M2 M2
− 3∆3 log n − i∆2 − i∆3 ≤ η ≤ . (11.10)
2 2
Proof of (11.9):
To see why the above bounds hold, note that in the case of the forward loop
removal switch, we have i choices for {x1 , x2 } and at most M1 choices for
{x3 , x4 } and there are two choices given these points. This explains the up-
per bound in (11.9). To get the lower bound we subtract the number of “bad”
choices. We can enumerate these bad choices as follows: We consider a fixed
loop {x1 , x2 } contained in cell Wa and we choose a pair x3 ∈ Wb and x4 ∈ Wc .
The transformation is bad only if there is x ∈ Wa \ {x1 , x2 } (≤ ∆ choices) that
is paired in F with y ∈ (Wb \ {x3 }) ∪ (Wc \ {x4 }) (≤ ∆ choices and two ways of
pairing x, y). We also subtract 2i to account for avoiding the other i − 1 loops
in the choice of x3 , x4 .
Proof of (11.10):
In the reverse procedure, we choose a pair {x1 , x2 } ⊆ Wa in M2 /2 ways to
arrive at the upper bound. The points x3 ∈ Wb , x4 ∈ Wc are those paired with
x1 , x2 in F 0 . For the lower bound, a choice is bad only if (a, b, c) is a triangle.
In this case, we create a double edge. There are at most ∆3 log n choices for the
triangle and then three choices for a. We subtract a further i∆2 to avoid creating
another loop. Finally, we subtract i∆3 in order to avoid increasing the number
of triangles by choosing an edge that is within distance two of the loop. We
also note here that forward d-switches do not increase the number of triangles.
208 Fixed Degree Sequence

Now for F ∈ Ω0, j let dL (F) denote the number of F 0 ∈ Ωi−1,0 that can be
obtained from F by an `-switch. Similarly, for F 0 ∈ Ωi−1,0 let dR (F 0 ) denote
the number of F ∈ Ωi,0 that can be transformed into F 0 by an `-switch. Then,

∑ dL (F) = ∑ dR (F 0 ).
F∈Ωi,0 F 0 ∈Ω i−1,0

But, Lemma 11.6 implies that i ≤ 2∆2 log n and so

2∆2 + 2∆2 log n
 
iM1 |Ωi,0 | 1 − ≤ ∑ dL (F) ≤ iM1 |Ωi,0 |,
M1 F∈Ω i,0

while
 
M2
− 3∆3 log n − 2∆2 log n(∆2 + ∆3 ) ≤ |Ωi−1,0 | ≤
2
M2
∑ dR (F 0 ) ≤ |Ωi−1,0 |.
0
F ∈Ωi−1,0
2

So
  5 
|Ωi−1,0 | 2iM1 ∆ log n
= 1+O .
|Ωi,0 | M2 M1

11.2 Connectivity of Regular Graphs

For an excellent survey of results on random regular graphs, see Wormald
[751].
Bollobás [123] used the configuration model to prove the following: Let Gn,r
denote a random r-regular graph with vertex set [n] and r ≥ 3 constant.
Theorem 11.9 Gn,r is r-connected, w.h.p.

Since an r-regular, r-connected graph, with n even, has a perfect matching,

the above theorem immediately implies the following Corollary.
Corollary 11.10 Let Gn,r be a random r-regular graph, r ≥ 3 constant, with
vertex set [n] even. Then w.h.p. Gn,r has a perfect matching.
Proof (of Theorem 11.9)
Partition the vertex set V = [n] of Gn,r into three parts, K, L and V \ (K ∪ L),
such that L = N(K), i.e., such that L separates K from V \ (K ∪ L) and |L| =
l ≤ r − 1. We will show that w.h.p. there are no such K, L for k ranging from
 11.2 Connectivity of Regular Graphs 209

2 to n/2. We will use the configuration model and the relationship stated in
Theorem 11.3. We will divide the whole range of k into three parts.

(i) 2 ≤ k ≤ 3.

Put S := K ∪ L, s = |S| = k + l ≤ r + 2. The set S contains at least 2r − 1

edges (k = 2) or at least 3r − 3 edges (k = 3). In both cases this is at least s + 1
edges.

P(∃S, s = |S| ≤ r + 2 : S contains s + 1 edges)

r+2   
n rs  rs s+1
≤∑ (11.11)
s=4 s s+1 rn
r+2
≤ ∑ ns 2rs ss+1 n−s−1
s=4
= o(1).

Explanation for (11.11): Having chosen a set of s vertices, spanning rs points

rs
R, we choose s + 1 of these points T . rn bounds the probability that one of
these points in T is paired with something in a cell associated with S. This
bound holds conditional on other points of R being so paired.

(ii) 4 ≤ k ≤ ne−10 .
The number of edges incident with the set K, |K| = k, is at least (rk + l)/2.
Indeed let a be the number of edges contained in K and b be the number of
K : L edges. Then 2a + b = rk and b ≥ l. This gives a + b ≥ (rk + l)/2. So,

ne−10 r−1 
r(k + l) (rk+l)/2
   
n n rk
P(∃K, L) ≤ ∑ ∑ rk+l
k=4 l=0 k l 2 rn
ne−10 r−1 r l ek+l rk
≤ ∑ ∑ n−( 2 −1)k+ 2 kk l l
2 (k + l)(rk+l)/2
k=4 l=0

Now
 l/2  k/2
k+l k/2 k+l
≤e and ≤ el/2 ,
l k

and so

(k + l)(rk+l)/2 ≤ l l/2 krk/2 e(lr+k)/2 .

210 Fixed Degree Sequence

Therefore, with Cr a constant,

ne−10 r−1 r l
P(∃K, L) ≤ Cr ∑ ∑ n−( 2 −1)k+ 2 e3k/2 2rk k(r−2)k/2
k=4 l=0
ne−10 r−1  r l r
k
= Cr ∑ ∑ n−( 2 −1)+ 2k e3/2 2r k 2 −1
k=4 l=0
= o(1).

(iii) ne−10 < k ≤ n/2

Assume that there are a edges between sets L and V \ (K ∪ L). Denote also
 m
(2m)! 1/2 2m
ϕ(2m) = ≈2 .
m! 2m e

Then, remembering that r, l, a = O(1) we can estimate that

P(∃K, L)
   
n n rl ϕ(rk + rl − a)ϕ(r(n − k − l) + a)
≤∑ (11.12)
k,l,a k l a ϕ(rn)
 ne k  ne l
≤ Cr ∑ ×
k,l,a k l
(rk + rl − a)rk+rl−a (r(n − k − l) + a)r(n−k−l)+a
(rn)rn
 ne k  ne l (rk)rk+rl−a (r(n − k))r(n−k−l)+a
≤ Cr0 ∑
k,l,a k l (rn)rn
 ne k  ne l  k rk  k r(n−k)

00
≤ Cr ∑ 1−
k,l,a k l n n
 r−1 !k
00 k 1−r/2 r/k
≤ Cr ∑ e n
k,l,a n
= o(1).

Explanation of (11.12): Having chosen K, L we choose a points in WK∪L =

i∈K∪L Wi that will be paired outside WK∪L . This leaves rk + rl − a points in
S

WK∪L to be paired up in ϕ(rk + rl − a) ways and then the remaining points can
be paired up in ϕ(r(n − k − l) + a) ways. We then multiply by the probability
1/ϕ(rn) of the final pairing.
 11.3 Existence of a giant component 211

11.3 Existence of a giant component

Molloy and Reed [595] provide an elegant and very useful criterion for when
Gn,d has a giant component. Suppose that there are λi n + o(n3/4 ) vertices of
degree i = 1, 2, . . . , L. We will assume that L = O(1) and that the λi , i ∈ [L] are
constants independent of n. The paper [595] allows for L = O(n1/4−ε ). We will
assume that λ1 + λ2 + · · · + λL = 1.
Theorem 11.11 Let Λ = ∑Li=1 λi i(i − 2). Let ε > 0 be arbitrary.
(a) If Λ < −ε then w.h.p. the size of the largest component in Gn,d is O(log n).
(b) If Λ > ε then w.h.p. there is a unique giant component of linear size ≈ Θn
where Θ is defined as follows: let K = ∑Li=1 iλi and
L
2α i/2
 
f (α) = K − 2α − ∑ iλi 1 − . (11.13)
i=1 K
Let ψ be the smallest positive solution to f (α) = 0. Then
L
2ψ i/2
 
Θ = 1 − ∑ λi 1 − .
i=1 K
If λ1 = 0 then Θ = 1, otherwise 0 < Θ < 1.
(c) In Case (b), the degree sequence of the graph obtained by deleting the
giant component satisfies the conditions of (a).
Proof We consider the execution of F-GENERATOR. We keep a sequence
of partitions Ut , At , Et ,t = 1, 2, . . . , m of W . Initially U0 = W and A0 = E0 = 0.
/
The (t + 1)th iteration of F-GENERATOR is now executed as follows: it is
designed so that we construct γ(F) component by component. At is the set of
points associated with the partially exposed vertices of the current component.
These are vertices in the current component, not all of whose points have been
paired. Ut is the set of unpaired points associated with the entirely unexposed
vertices that have not been added to any component so far. Et is the set of
paired points. Whenever possible, we choose to make a pairing that involves
the current component.
(i) If At = 0/ then choose x from Ut . Go to (iii).
We begin the exploration of a new component of γ(F).
(ii) if At 6= 0/ choose x from At . Go to (iii).
Choose a point associated with a partially exposed vertex of the current
component.
(iii) Choose y randomly from (At ∪Ut ) \ {x}.
(iv) F ← F ∪ {(x, y)}; Et+1 ← Et ∪ {x, y}; At+1 ← At \ {x}.
212 Fixed Degree Sequence

(v) If y ∈ At then At+1 ← At+1 \ {y}; Ut+1 ← Ut .

y is associated with a vertex in the current component.
(vi) If y ∈ Ut then At+1 ← At ∪ (Wϕ(y) \ y); Ut+1 ← Ut \Wϕ(y) .
y is associated with a vertex v = ϕ(y) not in the current component. Add
all the points in Wv \ {y} to the active set.
(vii) Goto (i).

(a) We fix a vertex v and estimate the size of the component containing v. We
keep track of the size of At for t = O(log n) steps. Observe that

∑Li=1 iλi n(i − 2) Λn ε

E(|At+1 | − |At | | |At | > 0) . = ≤ − . (11.14)
M1 − 2t − 1 M1 − 2t − 1 L
Here M1 = ∑Li=1 iλi n as before. The explanation for (11.14) is that |A| increases
only in Step (vi) and there it increases by i − 2 with probability . Miλ1 −2t
in
. The
two points x, y are missing from At+1 and this explains the -2.
Let ε1 = ε/L and let
(
|At | + ε1t |A1 |, |A2 |, . . . , |At | > 0.
Yt =
0 Otherwise.

It follows from (11.14) that if t = O(log n) and Y1 ,Y2 , . . . ,Yt > 0 then

E(Yt+1 | Y1 ,Y2 , . . . ,Yt ) = E(|At+1 | + ε1 (t + 1) | Y1 ,Y2 , . . . ,Yt ) ≤ |At | + ε1t = Yt .

Otherwise, E(Yt+1 | ·) = 0 = Yt . It follows that the sequence (Yt ) is a super-

martingale. Next let Z1 = 0 and Zt = Yt − Yt−1 for t ≥ 1. Then we have (i)
−2 ≤ Zi ≤ L and (ii) E(Zi ) ≤ −ε1 for i = 1, 2, . . . ,t. Now,

P(Aτ 6= 0,
/ 1 ≤ τ ≤ t) ≤ P(Yt = Z1 + Z2 + · · · + Zt > 0),

It follows from Lemma 22.16 that if Z = Z1 + Z2 + · · · + Zt then

 2 2
ε t
P(Z > 0) ≤ P(Z − E(Z) ≥ tε1 ) ≤ exp − 1 .
8t
It follows that with probability 1 − O(n−2 ) that At will become empty af-
ter at most 16ε1−2 log n rounds. Thus for any fixed vertex v, with probability
1 − O(n−2 ) the component contain v has size at most 4ε1−2 log n. (We can ex-
pose the component containing v through our choice of x in Step (i).) Thus
the probability there is a component of size greater than 16ε1−2 log n is O(n−1 ).
This completes the proof of (a).
(b)
If t ≤ δ n for a small positive constant δ then
 11.3 Existence of a giant component 213
−2|At | + (1 + o(1)) ∑Li=1 i(λi n − 2t)(i − 2)
E(|At+1 | − |At |) ≥
M1 − 2δ n
(Λ − 4δ )n ε
≥ ≥ . (11.15)
M1 − 2δ n 2L
Let ε2 = ε/2L and let
(
|At | − ε2t |A1 |, |A2 |, . . . , |At | > 0.
Yt =
0 Otherwise.

It follows from (11.14) that if t ≤ δ n and Y1 ,Y2 , . . . ,Yt > 0 then

E(Yt+1 | Y1 ,Y2 , . . . ,Yt ) = E(|At+1 | − ε2 (t + 1) | Y1 ,Y2 , . . . ,Yt ) ≥ |At | − ε2t = Yt .

Otherwise, E(Yt+1 | ·) = 0 = Yt . It follows that the sequence (Yt ) is a sub-

martingale. Next let Z1 = 0 and Zt = Yt − Yt−1 for t ≥ 1. Then we have (i)
−2 ≤ Zi ≤ L and (ii) E(Zi ) ≥ ε2 for i = 1, 2, . . . ,t. Now,

P(At 6= 0)
/ ≥ P(Yt = Z1 + Z2 + · · · + Zt > 0),

It follows from Lemma 22.16 that if Z = Z1 + Z2 + · · · + Zt then

 2 2
ε t
P(Z ≤ 0) ≤ P(Z − E(Z) ≥ tε2 ) ≤ exp − 2 .
2t
It follows that if L0 = 100ε2−2 then
 ε2t 
P ∃L0 log n ≤ t ≤ δ n : Z ≤ ≤
2
 ε2 t 
P ∃L0 log n ≤ t ≤ δ n : Z − E(Z) ≥
2
 2 
ε2 L0 log n
≤ n exp − = O(n−2 ).
8
It follows that if t0 = δ n then w.h.p. |At0 | = Ω(n) and there is a giant component
and that the edges exposed between time L0 log n and time t0 are part of exactly
one giant.
We now deal with the special case where λ1 = 0. There are two cases. If in
addition we have λ2 = 1 then w.h.p. Gd is the union of O(log n) vertex disjoint
cycles, see Exercise 10.5.1. If λ1 = 0 and λ2 < 1 then the only solutions to
f (α) = 0 are α = 0, K/2. For then 0 < α < K/2 implies
L
2α i/2 L
   
2α
∑ i iλ 1 − < ∑ i iλ 1 − = K − 2α.
i=2 K i=2 K
This gives Θ = 1. Exercise 10.5.2 asks for a proof that w.h.p. in this case,
214 Fixed Degree Sequence

Gn,d consists a giant component plus a collection of small components that are
cycles of size O(log n).
Assume now then that λ1 > 0. We show that w.h.p. there are Ω(n) isolated
edges. This together with the rest of the proof implies that Ψ < K/2 and hence
that Θ < 1. Indeed, if Z denotes the number components that are isolated edges,
then
   
λ1 n 1 λ1 n 6
E(Z) = and E(Z(Z − 1)) =
2 2M1 − 1 4 (2M1 − 1)(2M1 − 3)
and so the Chebyshev inequality (21.3) implies that Z = Ω(n) w.h.p.
Now for i such that λi > 0, we let Xi,t denote the number of entirely unex-
posed vertices of degree i. We focus on the number of unexposed vertices of a
give degree. Then,
iXi,t
E(Xi,t+1 − Xi,t ) = − . (11.16)
M1 − 2t − 1
This suggests that we employ the differential equation approach of Section 23
in order to keep track of the Xi,t . We would expect the trajectory of (t/n, Xi,t /n)
to follow the solution to the differential equation
dx ix
=− (11.17)
dτ K − 2τ
x(0) = λi . Note that K = M/n.
The solution to (11.17) is
2τ i/2
 
x = λi 1 − . (11.18)
K
In what follows, we use the notation of Section 23, except that we replace
λ0 by ξ0 = n−1/4 to avoid confusion with λi .

(P0) D = (τ, x) : 0 < τ < Θ−ε2 , 2ξ0 < x < 1 where ε is small and positive.
(P1) C0 = 1.
(P2) β = L.
ix
(P3) f (τ, x) = − K−2τ and γ = 0.
(P4) The Lipschitz constant L1 = 2K/(K − 2Θ)2 . This needs justification and
follows from
x x0 K(x − x0 ) + 2τ(x − x0 ) + 2x(τ − τ 0 )
− 0
= .
K − 2τ K − 2τ (K − 2τ)(K − 2τ 0 )
1/4 )
Theorem 23.1 then implies that with probability 1 − O(n1/4 e−Ω(n ),
  
2t i/2 
 
Xi,t − niλi 1 −  = O(n3/4 ), (11.19)

 K 
 11.3 Existence of a giant component 215

up a point where Xi,t = O(ξ0 n). (The o(n3/4 ) term for the number of vertices
of degree i is absorbed into the RHS of (11.19).)
Now because
L L
|At | = M1 − 2t − ∑ iXi,t = Kn − 2t − ∑ iXi,t ,
i=1 i=1

we see that w.h.p.

 !
L
2t i/2

2t
|At | = n K − − ∑ iλi 1 − + O(n3/4 )
n i=1 Kn
t 
= nf + O(n3/4 ), (11.20)
n
so that w.h.p. the first time after time t0 = δ n that |At | = O(n3/4 ) is as at time
t1 = Ψn + O(n3/4 ). This shows that w.h.p. there is a component of size at least
Θn + O(n3/4 ). Indeed, we simply subtract the number of entirely unexposed
vertices from n to obtain this.
To finish, we must show that this component is unique and no larger than
Θn + O(n3/4 ). We can do this by proving (c), i.e. showing that the degree se-
quence of the graph GU induced by the unexposed vertices satisfies the con-
dition of Case (a). For then by Case (a), the giant component can only add
O(n3/4 × log n) = o(n) vertices from t1 onwards.
We observe first that the above analysis shows that w.h.p. the degree se-
quence of GU is asymptotically equal to nλi0 , i = 1, 2, . . . , L, where

2Ψ i/2
 
λi0 = λi 1 − .
K
(The important thing here is that the number of vertices of degree i is asymp-
totically proportional to λi0 .) Next choose ε1 > 0 sufficiently small and let
tε1 = max {t : |At | ≥ ε1 n}. There must exist ε2 < ε1 such that tε1 ≤ (Ψ − ε2 )n
and f 0 (Ψ − ε2 ) ≤ −ε1 , else f cannot reach zero. Recall that Ψ < K/2 here and
then,

2Ψ − 2ε2 i/2
 
1
−ε1 ≥ f 0 (Ψ − ε2 ) = −2 + ∑ i2
λi 1 −
K − 2(Ψ − ε2 ) i≥1 K
 i/2
1 + O(ε2 ) 2Ψ
= −2 +
K − 2Ψ i≥1 ∑ i2 λ i 1 − K
 !
2Ψ i/2 2Ψ i/2
  
1 + O(ε2 ) 2
= −2 ∑ iλi 1 − + ∑ i λi 1 −
K − 2Ψ i≥1 K i≥1 K
216 Fixed Degree Sequence

2Ψ i/2
 
1 + O(ε2 )
= ∑ i(i − 2)λi 1 −
K − 2Ψ i≥1 K
1 + O(ε2 )
=
K − 2Ψ i≥1∑ i(i − 2)λi0 . (11.21)

This completes the proofs of (b), (c).

11.4 Gn,r is asymmetric

In this section, we prove that w.h.p. Gn,r , r ≥ 3 ony has one isomorphism, viz.
the identity isomorphism. This was proved by Bollobás [129]. For a vertex v
we let dk (v) denote the number of vertices at graph distance k from v in Gn,r .
We show that if k0 = 35 logr−1 n then w.h.p. no two vertices have the same
sequence (dk (v), k = 1, 2, . . . , k0 ). In the following G = Gn,r .
Lemma 11.12
 
Let `0 = 100 logr−1 log n . Then w.h.p., eG (S) ≤ |S| for all S ⊆ [n], |S| ≤ 2`0 .
Proof Arguing as for (11.11), we have that
2`0    s+1
n sr sr
P(∃S : |S| ≤ 2`0 , eG (S) ≥ |S| + 1) ≤ ∑
s=4 s s+1 2m − 4`0
2`0   s+1
ne s s
≤∑ (er)s+1
s=4 s n − o(n)
1 2`0 2s+1+o(1)
≤ ∑ se = o(1).
n s=4

Let E denote the high probability event in Lemma 11.12. We will condition
on the occurence of E .
Now for v ∈ [n] let Sk (v) denote the set of vertices at distance k from v and
S
let S≤k (v) = j≤k S j (v). We note that

|Sk (v)| ≤ r(r − 1)k−1 for all v ∈ [n], k ≥ 1. (11.22)

Furthermore, Lemma 11.12 implies that w.h.p. we have that for all v, w ∈
[n], 1 ≤ k ≤ `0 ,
|Sk (v)| ≥ (r − 2)(r + 1)(r − 1)k−2 . (11.23)
k−1
|Sk (v) \ Sk (w)| ≥ (r − 2)(r − 1) . (11.24)
 11.4 Gn,r is asymmetric 217

This is because there can be at most one cycle in S≤`0 (v) and the sizes of the
relevant sets are reduced by having the cycle as close to v, w as possible.
Now consider k > `0 . Consider doing breadth first search from v or v, w
exposing the configuration pairing as we go. Let an edge be dispensable if ex-
posing it joins two vertices already known to be in S≤k . Lemma 11.12 implies
that w.h.p. there is at most one dispensable edge in S≤`0 .
Lemma 11.13 With probability 1 − o(n−2 ), (i) at most 20 of the first n2/5
exposed edges are dispensable and (ii) at most n1/4 of the first n3/5 exposed
edges are dispensable.
Proof The probability that the kth edge is dispensable is at most (k−1)r
rn−2k , inde-
pendent of the history of the process. Hence,
 2/5  !20
n rn2/5
2/5
P(∃ 20 dispensable edges in first n ) ≤ = o(n−2 ).
20 rn − o(n)

 3/5  !n1/4
n rn3/5
P(∃ n 1/4 3/5
dispensable edges in first n ) ≤ 1/4 = o(n−2 ).
n rn − o(n)

Now let `1 = logr−1 n2/5 and `2 = logr−1 n3/5 . Then we have that, con-
   

ditional on E , with probability 1 − o(n−2 ),

|Sk (v)| ≥ ((r − 2)(r + 1)(r − 1)`0 −2 − 40))(r − 1)k−`0 : `0 < k ≤ `1 .
|Sk (v)| ≥ ((r − 2)(r + 1)(r − 1)`1 −1 − 40(r − 1)`1 −`0 − 2n1/4 )(r − 1)k−`1 : `1 < k ≤ `2 .
|Sk (w) \ Sk (v)| ≥ ((r − 2)(r − 1)`0 −1 − 40)(r − 1)k−`0 : `0 < k ≤ `1 .
|Sk (w) \ Sk (v)| ≥ ((r − 2)(r − 1)`1 −1 − 40(r − 1)`1 −`0 − 2n1/4 )(r − 1)k−`1 : `1 < k ≤ `2 .

We deduce from this that if `3 = logr−1 n4/7 and k = `3 + a, a = O(1) then

 

with probability 1 − o(n−2 ),

|Sk (w)| ≥ ((r − 2)(r + 1) − o(1))(r − 1)k−2 ≈ (r − 2)(r + 1)(r − 1)a−2 n4/7 .
|Sk (w) \ Sk (v)| ≥ (r − 2 − o(1))(r − 1)k−1 ≈ (r − 2)(r − 1)a−1 n4/7 .
Suppose now that we consider the execution of breadth first search up until we
have exposed Sk (v). Then in order to have dk (v) = dk (w), conditional on the
history of the search, there has to be an exact outcome for |Sk (w) \ Sk (v)|.
Now consider the pairings of the Wx , x ∈ Sk (w) \ Sk (v). Now at most n1/4
of these pairings are with vertices in S≤k (v) ∪ S≤k (w). Condition on these.
There must now be s = Θ(n4/7 ) pairings between Wx , x ∈ Sk (w) \ Sk (v) and
218 Fixed Degree Sequence

Wy , y ∈
/ Sk (v)∪Sk (w). Furthermore, to have dk (v) = dk (w) these s pairings must
involve exactly t of the sets Wy , y ∈
/ Sk (v) ∪ Sk (w), where t is determined before
the choice of these s pairings. The following lemma will easily show that G is
asymmetric w.h.p.
Lemma 11.14 Let R = m
S
i=1 Ri be a partitioning of an rm set R into m subsets
of size r. Suppose that S is a random s-subset of R, where m5/9 < s < m3/5 . Let
XS denote the number of sets Ri intersected by S. Then
c0 m1/2
max P(XS = j) ≤ ,
j s
for some constant c0 .
Proof We may assume that s ≥ m1/2 . The probability that S has at least 3
elements in some set Ri is at most
m 3r rm−3



s−3 r 3 s3 r3 m1/2
rm
≤ ≤ .
s
6m2 6s
But
   
P(XS = j) ≤ P max |S ∩ Ri | ≥ 3 + P XS = j and max |S ∩ Ri | ≤ 2 .
i i

So the lemma will follow if we prove that for every j,

c1 m1/2
 
Pj = P XS = j and max |S ∩ Ri | ≤ 2 ≤ , (11.25)
i s
for some constant c1 .
Clearly, Pj = 0 if j < s/2 and otherwise
m
j
2 j−s r
s− j
j s− j r 2
Pj = rm
. (11.26)
s

Now for s/2 ≤ j < s we have

Pj+1 (m − j)(s − j) 2r
= . (11.27)
Pj (2 j + 2 − s)(2 j + 1 − s) r − 1
10s2 Pj+1 10(r−1)
We note that if s − j ≥ m then Pj ≥ 3r ≥ 2 and so the j maximising Pj
αs2 2
is of the form s − m where α ≤ 10. If we substitute j = s − αs
m into (11.27)
then we see that
Pj+1 2αr h si
∈ 1 ± c2
Pj r−1 m
for some absolute constant c2 > 0.
 11.5 Gn,r versus Gn,p 219

It follows that if j0 is the index maximising Pj then

2 
  
 j0 − s − (r − 1)s  ≤ 1.

 2rm 
s
Furthermore, if j1 = j0 − m1/2 then

Pj+1 m1/2
≤ 1 + c3 for j1 ≤ j ≤ j0 ,
Pj s
for some absolute constant c3 > 0.
This implies that
!−( j0 − j1 ) ( !)
m1/2 m1/2 m
Pj ≥ Pj0 1 + c3 = Pj0 exp −( j0 − j1 ) c3 +O 2
s s s
≥ Pj0 e−2c3 .
It follows from this that
e2c3 m1/2
Pj0 ≤ .
s

We apply Lemma 11.14 with m = n, s = ρ = Θ(n4/7 to show that

!15
c0 n1/2
P(dk (v) = dk (w), k ∈ [`3 , `3 + 14]) ≤ = o(n−2 ).
n4/7
This proves
Theorem 11.15 W.h.p. Gn,r has a unique trivial automorphism.

11.5 Gn,r versus Gn,p

The configuration model is most useful when the maximum degree is bounded.
When r is large, one can learn a lot about random r-regular graphs from the
following theorem of Kim and Vu [493]. They proved that if log n  r 
n1/3 /(log n)2 then there is a joint distribution G0 , G = Gn,r , G1 such that w.h.p.
(1+o(1)) log n
(i) G0 ⊆ G, (ii) the maximum degree ∆(G1 \ G) ≤ log(ϕ(r)/ log n) where ϕ(r) is
any function satisfying (r log n)1/2 ≤ ϕ(r)  r. Here Gi = Gn,pi , i = 0, 1 where
p0 = (1 − o(1)) nr and p1 = (1 + o(1)) nr . In this way we can deduce properties
of Gn,r from Gn,r/n . For example, G0 is Hamiltonian w.h.p. implies that Gn,r is
Hamiltonian w.h.p. Recently, Dudek, Frieze, Ruciński and Šileikis [260] have
220 Fixed Degree Sequence

increased the range of r for which (i) holds. The cited paper deals with random
hypergraphs and here we describe the simpler case of random graphs.

Theorem 11.16 There is a positive constant C such that if

r log n 1/3
 
C + ≤ γ = γ(n) < 1,
n r
and m = b(1 − γ)nr/2c, then there is a joint distribution of G(n, m) and Gn,r
such that
P(Gn,m ⊂ Gn,r ) → 1.

Corollary 11.17 Let Q be an increasing property of graphs such that Gn,m

satisfies Q w.h.p. for some m = m(n), n log n  m  n2 . Then Gn,r satisfies Q
w.h.p. for r = r(n) ≈ 2mn .

Our approach to proving Theorem 11.16 is to represent Gn,m and Gn,r as the
outcomes of two graph processes which behave similarly enough to permit a
good coupling. For this let M = nr/2 and define

GM = (ε1 , . . . , εM )

to be an ordered random uniform graph on the vertex set [n], that is, Gn,M with
a random uniform ordering of edges. Similarly, let

Gr = (η1 , . . . , ηM )

be an ordered random r-regular graph on [n], that is, Gn,r with a random
uniform ordering of edges. Further, write GM (t) = (ε1 , . . . , εt ) and Gr (t) =
(η1 , . . . , ηt ), t = 0, . . . , M.
For every ordered graph G of size t and every edge e ∈ Kn \ G we have
1
Pr (εt+1 = e | GM (t) = G) = n
.
2 −t

This is not true if we replace GM by Gr , except for the very first step t = 0.
However, it turns out that for most of time the conditional distribution of the
next edge in the process Gr (t) is approximately uniform, which is made precise
in the lemma below. For 0 < ε < 1, and t = 0, . . . , M consider the inequalities
1−ε
Pr (ηt+1 = e | Gr (t)) ≥ n
for every e ∈ Kn \ Gr (t), (11.28)
2 −t

and define a stopping time

Tε = max {u : ∀t ≤ u condition (11.28) holds } .

 11.5 Gn,r versus Gn,p 221

Lemma 11.18 There is a positive constant C0 such that if

log n 1/3
 
0 r
C + ≤ ε = ε(n) < 1, (11.29)
n r
then
Tε ≥ (1 − ε)M w.h.p.

From Lemma 11.18, which is proved in Section 11.5, we deduce Theorem

11.16 using a coupling.

Proof of Theorem 11.16 Let C = 3C0 , where C0 is the constant from Lemma
11.18. Let ε = γ/3. The distribution of Gr is uniquely determined by the con-
ditional probabilities

pt+1 (e|G) := Pr (ηt+1 = e | Gr (t) = G) , t = 0, . . . , M − 1. (11.30)

Our aim is to couple GM and Gr up to the time Tε . For this we will define a
graph process G0r := (ηt0 ),t = 1, . . . , M such that the conditional distribution of
(ηt0 ) coincides with that of (ηt ) and w.h.p. (ηt0 ) shares many edges with GM .
Suppose that Gr = G0r (t) and GM = GM (t) have been exposed and for every
e∈/ Gr the inequality
1−ε
pt+1 (e|Gr ) ≥ n
(11.31)
2 −t

holds (we have such a situation, in particular, if t ≤ Tε ). Generate a Bernoulli

(1 − ε) random variable ξt+1 independently of everything that has been re-
vealed so far; expose the edge εt+1 . Moreover, generate a random edge ζt+1 ∈
Kn \ Gr according to the distribution
1−ε
pt+1 (e|Gr ) −
(n2)−t
P(ζt+1 = e|G0r (t) = Gr , GM (t) = GM ) := ≥ 0,
ε
where the inequality holds because of the assumption (11.31). Observe also
that
∑ P(ζt+1 = e|G0r (t) = Gr , GM (t) = GM ) = 1,
e6∈Gr

so ζt+1 has a well-defined distribution. Finally, fix a bijection fGr ,GM : Gr \

GM → GM \ Gr between the sets of edges and define

εt+1 ,

 if ξt+1 = 1, εt+1 ∈
/ Gr ,
0
ηt+1 = fGr ,GM (εt+1 ), if ξt+1 = 1, εt+1 ∈ Gr ,


ζ , if ξ = 0.
t+1 t+1
222 Fixed Degree Sequence

Note that
ξt+1 = 1 ⇒ εt+1 ∈ G0r (t + 1). (11.32)
We keep generating ξt ’s even after the stopping time has passed, that is, for t >
0
Tε , whereas ηt+1 is then sampled according to probabilities (11.30), without
coupling. Note that ξt ’s are i.i.d. and independent of GM . We check that
0
P(ηt+1 = e | G0r (t) = Gr , GM (t) = GM )
= P(εt+1 = e) P(ξt+1 = 1) + P(ζt+1 = e) P(ξt+1 = 0)
pt+1 (e|Gr ) − 1−ε
 
n
1−ε (2 )−t
= n
+ ε
2 −t
ε

= pt+1 (e|Gr )
for all admissible Gr , GM , i.e., such that P (Gr (t) = Gr , GM (t) = GM ) > 0, and
for all e 6∈ Gr .
Further, define a set of edges which are potentially shared by GM and Gr :
S := {εi : ξi = 1 , 1 ≤ i ≤ (1 − ε)M} .
Note that
b(1−ε)Mc
|S| = ∑ ξi
i=1

is distributed as Bin(b(1 − ε)Mc, 1 − ε).

Since (ξi ) and (εi ) are independent, conditioning on |S| ≥ m, the first m
edges in the set S comprise a graph which is distributed as Gn,m . Moreover, if
Tε ≥ (1 − ε)M, then by (11.32) we have S ⊂ Gr , therefore
P (Gn,m ⊂ Gn,r ) ≥ P (|S| ≥ m, Tε ≥ (1 − ε)M) .
We have E |S| ≥ (1 − 2ε)M. Recall that ε = γ/3 and therefore m = b(1 −
γ)Mc = b(1 − 3ε)Mc. Applying the Chernoff bounds and our assumption on
ε, we get
2 m)
P (|S| < m) ≤ e−Ω(γ = o(1).
Finally, by Lemma 11.18 we have Tε ≥ (1 − ε)M w.h.p., which completes the
proof of the theorem.

Proof of Lemma 11.18

In all proofs of this section we will assume the condition (11.29). To prove
Lemma 11.18 we will start with a fact which allows one to control the degrees
of the evolving graph Gr (t).
 11.5 Gn,r versus Gn,p 223

For a vertex v ∈ [n] and t = 0, . . . , M, let

degt (v) = |{i ≤ t : v ∈ ηi }| .

Lemma 11.19 Let τ = 1 − t/M. We have that w.h.p.

p
∀t ≤ (1 − ε)M, ∀v ∈ [n], | degt (v) − tr/M| ≤ 6 τr log n. (11.33)

In particular w.h.p.

∀t ≤ (1 − ε)M, ∀v ∈ [n], degt (v) ≤ (1 − ε/2)r. (11.34)

Proof Observe that if we fix an r-regular graph H and condition Gr to be a

permutation of the edges of H, then X := degt (v) is a hypergeometric random
variable with expected value tr/M = (1 − τ)r. Using the result of Section 22.5
and Theorem 22.11, and checking that the variance of X is at most τr, we get
x2
 
P (|X − tr/M| ≥ x) ≤ 2 exp − .
2 (τr + x/3)
√
Let x = 6 τr log n. From (11.29), assuming C0 ≥ 1, we get
r r
x log n log n
=6 ≤6 ≤ 6ε,
τr τr εr
and so x ≤ 6τr. Using this, we obtain
 
1 36τr log n
P (|X − tr/M| ≥ x) ≤ exp − = n−6 .
2 2(τr + 2τr)
Inequality (11.33) now follows by taking a union bound over nM ≤ n3 choices
of t and v.
To get (11.34), it is enough to prove the inequality for t = (1 −ε)M. Inequal-
ity (11.33) implies
p
deg(1−ε)M (v) ≤ (1 − ε)r + 6 εr log n.

Thus it suffices to show that

p
6 εr log n ≤ εr/2,

or, equivalently, ε ≥ 144 log n/r, which is implied by (11.29) with C0 ≥ 144.

Given an ordered graph G = (e1 , . . . , et ), we say that an ordered r-regular

graph H is an extension of G if the first t edges of H are equal to G. Let
GG (n, r) be the family of extensions of G and GG = GG (n, r) be a graph chosen
uniformly at random from GG (n, r).
224 Fixed Degree Sequence

Further, for a graph H ∈ GG (n, r) and u, v ∈ [n] let

degH|G (u, v) = | {w ∈ [n] : {u, w} ∈ H \ G, {v, w} ∈ H} |.
Note that degH|G (u, v) is not in general symmetric in u and v, but for G = 0/
coincides with the usual co-degree in a graph H.
The next fact is used in the proof of Lemma 11.21 only.
Lemma 11.20 Let graph G with t ≤ (1 − ε)M edges be such that GG (n, r) is
nonempty. For each e ∈
/ G we have
4r
P (e ∈ GG ) ≤ . (11.35)
εn
Moreover, if l ≥ l0 := 4r2 /(εn), then for every u, v ∈ [n] we have
 
P degGG |G (u, v) > l ≤ 2−(l−l0 ) . (11.36)

Proof To prove (11.35) define the families

Ge∈ = {H ∈ GG (n, r) : e ∈ H} Ge∈/ = H 0 ∈ GG (n, r) : e ∈
/ H0 .

and
Let us define an auxiliary bipartite graph B between Ge∈ and Ge∈/ in which
H ∈ Ge∈ is connected to H 0 ∈ Ge∈/ whenever H 0 can be obtained from H by the
following switching operation. Fix an ordered edge {w, x} in H \ G which is
disjoint from e = {u, v} and such that there are no edges between {u, v} and
{w, x} and replace the edges {u, v} and {w, x} by {u, w} and {v, x} to obtain
H 0 . Writing f (H) for the number of graphs H 0 ∈ Ge∈/ which can be obtained
from H by a switching, and b(H 0 ) for the number of graphs H ∈ Ge∈ such that
H 0 can be obtained H by a switching, we get that
|Ge∈ | min f (H) ≤ |E(B)| ≤ |Ge∈/ | max b(H 0 ). (11.37)
H H0

We have b(H 0 ) ≤ degH 0 (u) degH 0 (v) ≤ r2 . On the other hand, recalling that
t ≤ (1 − ε)M, for every H ∈ Ge∈ we get
2r2
 
εM
f (H) ≥ M − t − 2r2 ≥ εM 1 − ≥ ,
εM 2
because, assuming C0 ≥ 8, we have
2r2 4r  n 1/3 4 1
≤ 0 ≤ 0≤ .
εM C n r C 2
Therefore (11.37) implies that
|Ge∈ | 2r2 4r
P (e ∈ GG ) ≤ ≤ = ,
|Ge∈/ | εM εn
 11.5 Gn,r versus Gn,p 225
u0 w0 u0 w0

u u
w w
v v

Figure 11.6 Switching between G (l) and G (l − 1): Before and after.

which concludes the proof of (11.35).

To prove (11.36), fix u, v ∈ [n] and define the families
n o
G (l) = H ∈ GG (n, r) : degH|G (u, v) = l , l = 0, 1, . . . .

We compare sizes of G (l) and G (l − 1) in a similar way as above. For this we

define the following switching which maps a graph H ∈ G (l) to a graph H 0 ∈
G (l − 1). Select a vertex w contributing to degH|G , that is, such that {u, w} ∈
H \ G and {v, w} ∈ H; pick an ordered pair u0 , w0 ∈ [n] \ {u, v, w} such that
{u0 , w0 } ∈ H \ G and there are no edges of H between {u, v, w} and {u0 , w0 };
replace edges {u, w} and {u0 , w0 } by {u, u0 } and {w, w0 } (see Figure 11.6).
The number of ways to apply a forward switching to H is
3r2
 
f (H) ≥ 2l(M − t − 3r2 ) ≥ 2lεM 1 − ≥ lεM,
εM
since, assuming C0 ≥ 12 we have
3r2 6r 6  r 2/3 1
= ≤ 0 ≤ ,
εM εn C n 2
and the number of ways to apply a backward switching is b(H) ≤ r3 . So,
|G (l)| maxH∈G (l−1) b(H) 2r2 1
≤ ≤ ≤ ,
|G (l − 1)| minH∈G (l) f (H) εln 2

by the assumption l ≥ l0 := 4r2 /(εn). Then

  |G (i)| |G (i)|
P degGG |G (u, v) > l ≤ ∑ ≤∑
i>l |G G (n, r)| i>l |G (l0 )|
i
|G ( j)|
=∑ ∏ ≤ ∑ 2−(i−l0 ) = 2−(l−l0 ) ,
i>l j=l0 +1 |G ( j − 1)| i>l

which completes the proof of (11.36).

For the last lemma, which will be directly used in Lemma 11.18, we need to
provide a few more definitions regarding random r-regular multigraphs.
226 Fixed Degree Sequence

Let G be an ordered graph with t edges. Let MG (n, r) be a random multi-

graph extension of G to an ordered r-regular multigraph. Namely, MG (n, r)
is a sequence of M edges (some of which may be loops), the first t of which
comprise G, while the remaining ones are generated by taking a uniform ran-
dom permutation Π of the multiset {1, . . . , 1, . . . , n, . . . , n} with multiplicities
r − degG (v), v ∈ [n], and splitting it into consecutive pairs.
Recall that the number of such permutations is
(2(M − t))!
NG := ,
∏v∈[n] (r − degG (v))!
and note that if a multigraph extension H of G has l loops, then

P (MG (n, r) = H) = 2M−t−l /NG . (11.38)

Thus, MG (n, r) is not uniformly distributed over all multigraph extensions of

G, but it is uniform over GG (n, r). Thus, MG (n, r), conditioned on being simple,
has the same distribution as GG (n, r). Further, for every edge e ∈
/ G, let us write

Me = MG∪e (n, r) and Ge = GG∪e (n, r). (11.39)

The next claim shows that the probabilities of simplicity P (Me ∈ Ge ) are
asymptotically the same for all e 6∈ G.

Lemma 11.21 Let G be graph with t ≤ (1 − ε)M edges such that GG (n, r) is
nonempty. If ∆G ≤ (1 − ε/2)r, then for every e0 , e00 ∈
/ G we have
P (Me00 ∈ Ge00 ) ε
≥ 1− .
P (Me0 ∈ Ge0 ) 2
Proof Set

M0 = Me0 M00 = Me00 , G 0 = Ge0 , and G 00 = Ge00 , (11.40)

for convenience. We start by constructing a coupling of M0 and M00 in which

they differ in at most three positions (counting in the replacement of e0 by e00
at the (t + 1)st position).
Let e0 = {u0 , v0 } and e00 = {u00 , v00 }. Suppose first that e0 and e00 are disjoint.
Let Π0 be the permutation underlying the multigraph M0 . Let Π∗ be obtained
from Π0 by replacing a uniform random copy of u00 by u0 and a uniform random
copy of v00 by v0 . If e0 and e00 share a vertex, then assume, without loss of
generality, that v0 = v00 , and define Π∗ by replacing only a random u00 in Π0 by
u0 . Then define M∗ by splitting Π∗ into consecutive pairs and appending them
to G ∪ e00 .
It is easy to see that Π∗ is uniform over permutations of the multiset {1, . . . ,
 11.5 Gn,r versus Gn,p 227

1, . . . , n, . . . , n} with multiplicities d −degG∪e00 (v), v ∈ [n], and therefore M∗ has

the same distribution as M00 . Thus, we will further identify M∗ and M00 .
Observe that if we condition M0 on being a simple graph H, then M∗ = M00
can be equivalently obtained by choosing an edge incident to u00 in H \ (G ∪
e0 ) uniformly at random, say, {u00 , w}, and replacing it by {u0 , w}, and then
repeating this operation for v00 and v0 . The crucial idea is that such a switching
of edges is unlikely to create loops or multiple edges.
It is, however, possible, that for certain H this is not true. For example, if
e ∈ H \ (G ∪ e0 ), then the random choice of two edges described above is
00

unlikely to destroy this e00 , but e0 in the non-random part will be replaced by
e00 , thus creating a double edge e00 . Moreover, if almost every neighbor of u00 in
H \ (G ∪ e0 ) is also a neighbor of u0 , then most likely the replacement of u00 by
u0 will create a double edge. To avoid such instances, we want to assume that
(i) e00 ∈
/ H 
(ii) max degH|G∪e0 (u0 , u00 ), degH|G∪e0 (v0 , v00 ) ≤ l0 + log2 n,

where l0 = 4r2 /εn is as in Lemma 11.20. Define the following subfamily of

simple extensions of G ∪ e0 :
0
Gnice = H ∈ G 0 : H satisfies (i) and (ii) .


Since M0 , conditioned on M0 ∈ G 0 , is distributed as GG∪e0 (n, r), by Lemma

11.20 and the assumption (11.29) with C0 ≥ 20,

Pr M0 ∈ 0
/ Gnice | M0 ∈ G 0 = P GG∪e0 (n, r) 6∈ Gnice
0

4r ε
≤ + 2 × 2− log2 n ≤ . (11.41)
εn 4
We have

Pr M00 ∈ G 00 | M0 ∈ Gnice
0
Pr M0 ∈ Gnice
0
| M0 ∈ G 0 =

P(M00 ∈ G 00 , M0 ∈ Gnice
0 ) P(M0 ∈ G 0 , M0 ∈ G 0 )
nice
· ≤
P(M0 ∈ Gnice0 ) P(M0 ∈ G 0 )
P (M00 ∈ G 00 )
. (11.42)
P (M0 ∈ G 0 )
To complete the proof of the claim, it suffices to show that
ε
Pr M00 ∈ G 00 | M0 ∈ Gnice
0


≥ 1− , (11.43)
4
since plugging (11.41) and (11.43) into (11.42) will complete the proof of the
statement.
To prove (11.43), fix H ∈ Gnice
0 and condition on M0 = H. A loop can only be
228 Fixed Degree Sequence

created in M00 when u00 is incident to u0 in H \ (G ∪ e0 ) and the randomly chosen

edge is {u0 , u00 }, or, provided v0 6= v00 , when v00 is incident to v0 in H \ (G ∪ e0 )
and we randomly choose {v0 , v00 }. Therefore, recalling that ∆G ≤ (1 − ε/2)r,
we get

1 1
Pr M00 has a loop | M0 = H ≤


+
degH\(G∪e0 ) (u00 ) degH\(G∪e0 ) (v00 )
4 ε
≤ ≤ , (11.44)
εr 8
where the second term is present only if e0 ∩ e00 = 0, / and the last inequality is
implied by (11.29).
A multiple edge can be created in three ways: (i) by choosing, among the
edges incident to u00 , an edge {u00 , w} ∈ H \ (G ∪ e0 ) such that {u0 , w} ∈ H;
(ii) similarly for v00 (if v0 6= v00 ); (iii) choosing both edges {u00 , v0 } and {v00 , u0 }
(provided they exist in H \ (G ∪ e0 )). Therefore, by (ii) and assumption ∆G ≤
(1 − ε/2)r,

Pr M00 has a multiple edge | M0 = H

degH|G∪e0 (u00 , u0 ) degH|G∪e0 (v00 , v0 )

≤ +
degH\(G∪e0 ) (u00 ) degH\(G∪e0 ) (v00 )
1
+
degH\(G∪e0 ) (u00 ) degH\(G∪e0 ) (v00 )
 
8r 2 log2 n 4 ε
≤2 2 + + 2 2 ≤ , (11.45)
ε n εr ε r 8

because (11.29) implies ε > C0 (r/n)1/3 and

ε > C0 (log n/r)1/3 > C0 (log n/r)1/2

and we can choose arbitrarily large C0 . (Again, in case when |e0 ∩ e00 | = 1, the
R-H-S of (11.45) reduces to only the first summand.)
Combining (11.44) and (11.45), we have shown (11.43).

Proof of Lemma 11.18 In view of Lemma 11.19 it suffices to show that

1−ε
Pr (ηt+1 = e | Gr (t) = G) ≥ n
, e∈
/ G.
2 −t

for every t ≤ (1 − ε)M and G such that

εr
r(τ + δ ) ≥ r − degG (v) ≥ r(τ − δ ) ≥ , v ∈ [n], (11.46)
2
 11.5 Gn,r versus Gn,p 229

where
p
τ = 1 − t/M, δ = 6 τ log n/r.

For every e0 , e00 ∈

/ G we have (recall the definitions (11.39) and (11.40))
Pr (ηt+1 = e00 | Gr (t) = G) |GG∪e00 (n, r)| |G 00 |
= = . (11.47)
Pr (ηt+1 = e0 | Gr (t) = G) |GG∪e0 (n, r)| |G 0 |
By (11.38) we have

|G 0 |2M−t |G 0 |2M−t ∏v∈[n] (r − degG∪e0 (v))!
P M0 ∈ G 0 = = ,
NG (2(M − t))!
and similarly for the family G 00 . This yields, after a few cancellations, that

|G 00 | ∏v∈e00 \e0 (r − degG (v)) P (M00 ∈ G 00 )

= · (11.48)
|G 0 | ∏v∈e0 \e00 (r − degG (v)) P (M0 ∈ G 0 )
By (11.46), the ratio of products in (11.48) is at least

τ −δ 2 2δ 2
    r r
log n log n ε
≥ 1− ≥ 1 − 24 ≥ 1 − 24 ≥ 1− ,
τ +δ τ τr εr 2
where the last inequality holds by the assumption (11.29). Since by
Lemma 11.21 the ratio of probabilities in (11.48) is
P (M00 ∈ G 00 ) ε
≥ 1− ,
P (M ∈ G )
0 0 2
we have obtained that
Pr (ηt+1 = e00 | Gr (t) = G)
≥ 1 − ε.
Pr (ηt+1 = e0 | Gr (t) = G)
Finally, noting that

max Pr ηt+1 = e0 | Gr (t) = G

e0 ∈G
/

is at least as large as the average over all e0 ∈

/ G, which is 1
, we conclude
(n2)−t
that for every e ∈
/G

Pr (ηt+1 = e | Gr (t) = G) ≥ (1 − ε) max Pr ηt+1 = e0 | Gr (t) = G

e0 ∈G
/
1−ε
≥ n
,
2 −t

which finishes the proof.

230 Fixed Degree Sequence

11.6 Exercises
11.6.1 Show that w.h.p. a random 2-regular graph on n vertices consists of
O(log n) vertex disjoint cycles.
11.6.2 Suppose that in the notation of Theorem 11.11, λ1 = 0, λ2 < 1. Show
that w.h.p. Gn,d consists of a giant component plus a collection of small
components of size O(log n).
11.6.3 Let H be a subgraph of Gn,r , r ≥ 3 obtained by independently including
each vertex with probability 1+ε r−1 , where ε > 0 is small and positive.
Show that w.h.p. H contains a component of size Ω(n).
11.6.4 Let x = (x1 , x2 , . . . , x2m ) be chosen uniformly at random from [n]2m .
Let Gx be the multigraph with vertex set [n] and edges (x2i−1 , x2i ), i =
1, 2, . . . , m. Let dx (i) be the number of times that i appears in x.
Show that conditional on dx (i) = di , i ∈ [n], Gx has the same distri-
bution as the multigraph γ(F) of Section 11.1.
11.6.5 Suppose that we condition on dx (i) ≥ k for some non-negative integer
k. For r ≥ 0, let
xk−1
fr (x) = ex − 1 − x − · · · − .
(k − 1)!
Let Z be a random variable taking values in {k, k + 1, . . . , } such that
λ i e−λ
P(Z = i) = for i ≥ k,
i! fk (λ )
where λ is arbitrary and positive.
Show that the degree sequence of Gx is distributed as independent
copies Z1 , Z2 , . . . , Zn of Z, subject to Z1 + Z2 + · · · + Zn = 2m.
11.6.6 Show that
λ fk−1 (λ )
E(Z) = .
fk (λ )
Show using the Local Central Limit Theorem (see e.g. Durrett [267])
that if E(Z) = 2m
n then
!
v
1
1 + O((k2 + 1)v−1 σ −2 )


P ∑ Z j = 2m − k = √
j=1 σ 2πn

where σ 2 = E(Z 2 ) − E(Z)2 is the variance of Z.

11.6.7 Use the model of (i)–(iii) to show that if c = 1 + ε and ε is sufficiently
small and ω → ∞ then w.h.p. the 2-core of Gn,p , p = c/n does not con-
tain a cycle C, |C| = ω in which more than 10% of the vertices are of
degree three or more.
 11.7 Notes 231

11.6.8 Let G = Gn,r , r ≥ 3 be the random r-regular configuration multigraph

of Section 11.2. Let X denote the number of Hamilton cycles in G.
Show that
!n
r − 2 (r−2)/2
r  
π
E(X) ≈ (r − 1) .
2n r
11.6.9 Show that if graph G = G1 ∪ G2 then its rainbow connection satisfies
rc(G) ≤ rc(G1 ) + rc(G2 ) + |E(G1 ) ∩ E(G2 )|. Using the contiguity of
Gn,r to the union of r independent matchings, (see Chapter 20), show
that rc(Gn,r ) = O(logr n) for r ≥ 6.
11.6.10 Show that w.h.p. Gn,3 is not planar.

11.7 Notes
Giant Components and Cores
Hatami and Molloy [406] discuss the size of the largest component in the scal-
ing window for a random graph with a fixed degree sequence.
Cooper [204] and Janson and Luczak [438] discuss the sizes of the cores of
random graphs with a given degree sequence.

Hamilton cycles
Robinson and Wormald [668], [671] showed that random r-regular graphs are
Hamiltonian for 3 ≤ r = O(1). In doing this, they introduced the important
new method of small subgraph conditioning. It is a refinement on the Cheby-
shev inequality. Somewhat later Cooper, Frieze and Reed [229] and Krivele-
vich, Sudakov, Vu Wormald [524] removed the restriction r = O(1). Frieze,
Jerrum, Molloy, Robinson and Wormald [338] gave a polynomial time algo-
rithm that w.h.p. finds a Hamilton cycle in a random regular graph. Cooper,
Frieze and Krivelevich [223] considered the existence of Hamilton cycles in
Gn,d for certain classes of degree sequence.

Chromatic number
r
Frieze and Łuczak [347] proved that w.h.p. χ(Gn,r ) = (1 + or (1)) 2 log r for
r = O(1). Here or (1) → 0 as r → ∞. Achlioptas and Moore [4] determined the
chromatic number of a random r-regular graph to within three values, w.h.p.
Kemkes, Pérez-Giménez and Wormald [490] reduced the range to two val-
ues. Shi and Wormald [704], [705] consider the chromatic number of Gn,r for
232 Fixed Degree Sequence

small r. In particular they show that w.h.p. χ(Gn,4 ) = 3. Frieze, Krivelevich

and Smyth [342] gave estimates for the chromatic number of Gn,d for certain
classes of degree sequence.

Eigenvalues
The largest eigenvalue of the adjacency matrix of Gn,r is always r. Kahn and
Szemerédi [467] showed that w.h.p. the second eigenvalue is of order O(r1/2 ).
Friedman [323] proved that w.h.p. the second eigenvalue is at most 2(r −
1)1/2 + o(1). Broder, Frieze, Suen and Upfal [165] considered Gn,d where
C−1 d ≤ di ≤ Cd for some constant C > 0 and d ≤ n1/10 . They show that w.h.p.
the second eigenvalue of the adjacency matrix is O(d 1/2 ).

First Order Logic

Haber and Krivelevich [398] studied the first order language on random d-
regular graphs. They show that if r = Ω(n) or r = nα where α is irrational,
then Gn,r obeys a 0-1 law.

Rainbow Connection
Dudek, Frieze and Tsourakakis [261] studied the rainbow connection of ran-
dom regular graphs. They showed that if 4 ≤ r = O(1) then rc(Gn,r ) = O(log n).
This is best possible up to constants, since rc(Gn,r ) ≥ diam(Gn,r ) = Ω(log n).
Kamčev, Krivelevich and Sudakov [470] gave a simpler proof when r ≥ 5, with
a better hidden constant.
 12
Intersection Graphs

Let G be a (finite, simple) graph. We say that G is an intersection graph if we

can assign to each vertex v ∈ V (G) a set Sv , so that {v, w} ∈ E(G) exactly when
Sv ∩ Sw 6= 0.
/ In this case, we say G is the intersection graph of the family of
sets S = {Sv : v ∈ V (G)}.
Although all graphs are intersection graphs (see Marczewski [567]) some
classes of intersection graphs are of special interest. Depending on the choice
of family S , often reflecting some geometric configuration, one can consider,
for example, interval graphs defined as the intersection graphs of intervals on
the real line, unit disc graphs defined as the intersection graphs of unit discs on
the plane etc. In this chapter we will discuss properties of random intersection
graphs, where the family S is generated in a random manner.

12.1 Binomial Random Intersection Graphs

Binomial random intersection graphs were introduced by Karoński,
Scheinerman and Singer-Cohen in [482] as a generalisation of the classical
model of the binomial random graph Gn,p .
Let n, m be positive integers and let 0 ≤ p ≤ 1. Let V = {1, 2, . . . , n} be the
set of vertices and for every 1 ≤ k ≤ n, let Sk be a random subset of the set
M = {1, 2, . . . , m} formed by selecting each element of M independently with
probability p. We define a binomial random intersection graph G(n, m, p) as
the intersection graph of sets Sk , k = 1, 2, . . . n. Here S1 , S2 , . . . , Sn are generated
independently. Hence two vertices i and j are adjacent in G(n, m, p) if and only
if Si ∩ S j 6= 0.
/
There are other ways to generate binomial random intersection graphs. For
example, we may start with a classical bipartite random graph Gn,m,p , with
vertex set bipartition

(V, M),V = {1, 2, . . . , n}, M = {1, 2, . . . , m},

where each edge between V and M is drawn independently with probability p.

Next, one can generate a graph G(n, m, p) with vertex set V and vertices i and
j of G(n, m, p) connected if and only if they share a common neighbor (in M)

233
234 Intersection Graphs

in the random graph Gn,m,p . Here the graph Gn,m,p is treated as a generator of
G(n, m, p).
One observes that the probability that there is an edge {i, j} in G(n, m, p)
equals 1 − (1 − p2 )m , since the probability that sets Si and S j are disjoint is
(1− p2 )m , however, in contrast with Gn,p , the edges do not occur independently
of each other.
Another simple observation leads to some natural restrictions on the choice
of probability p. Note that the expected number of edges of G(n, m, p) is,
 
n
(1 − (1 − p2 )m ) ≈ n2 mp2 ,
2
√
provided mp2 → 0 as n → ∞. Therefore, if we take p = o((n m)−1 ) then the
expected number of edges of G(n, m, p) tends to 0 as n → ∞ and therefore
w.h.p. G(n, m, p) is empty.
On the other hand the expected number of non-edges in G(n, m, p) is
 
n 2
(1 − p2 )m ≤ n2 e−mp .
2

Thus if we take p = (2 log n + ω(n))/m)1/2 , where ω(n) → ∞ as n → ∞, then

the random graph G(n, m, p) is complete w.h.p. One can also easily show that
when ω(n) → −∞ then G(n, m, p) is w.h.p. not complete. So, when studying
the evolution of G(n, m, p) we may restrict ourselves to values of p in the range
√
between ω(n)/(n m) and ((2 log n − ω(n))/m)1/2 , where ω(n) → ∞.

Equivalence
One of the first interesting problems to be considered is the question as to
when the random graphs G(n, m, p) and Gn,p have asymptotically the same
properties. Intuitively, it should be the case when the edges of G(n, m, p) oc-
cur “almost independently”, i.e., when there are no vertices of degree greater
than two in M in the generator Gn,m,p of G(n, m, p). Then each of its edges
is induced by a vertex of degree two in M, “almost” independently
of other
edges. One can show that this happens w.h.p. when p = o 1/(nm1/3 ) , which
in turn implies that both random graphs are asymptotically equivalent for all
graph properties P. Recall that a graph property P is defined as a subset of
n
the family of all labeled graphs on vertex set [n], i.e., P ⊆ 2(2) . The follow-
ing equivalence result is due to Rybarczyk [683] and Fill, Scheinerman and
Singer-Cohen [304].
 12.1 Binomial Random Intersection Graphs 235

Let 0 ≤ a ≤ 1, P be any graph property, p = o 1/(nm1/3 )

Theorem 12.1
and
p̂ = 1 − exp −mp2 (1 − p)n−2 .


(12.1)
Then
P(Gn, p̂ ∈ P) → a
if and only if
P(G(n, m, p) ∈ P) → a
as n → ∞.
Proof Let X and Y be random variables taking values in a common finite
(or countable) set S. Consider the probability measures L (X) and L (Y ) on S
whose values at A ⊆ S are P(X ∈ A) and P(Y ∈ A). Define the total variation
distance between L (X) and L (Y ) as
dTV (L (X), L (Y )) = sup | P(X ∈ A) − P(Y ∈ A)|,
A⊆S

which is equivalent to
1
dTV (L (X), L (Y )) = ∑ | P(X = s) − P(Y = s)|.
2 s∈S
Notice (see Fact 4 of [304]) that if there exists a probability space on which ran-
dom variables X 0 and Y 0 are both defined, with L (X) = L (X 0 ) and L (Y ) =
L (Y 0 ), then
dTV (L (X), L (Y )) ≤ P(X 0 6= Y 0 ). (12.2)
Furthermore (see Fact 3 in [304]) if there exist random variables Z and Z 0 such
that L (X|Z = z) = L (Y |Z 0 = z), for all z, then
dTV (L (X), L (Y )) ≤ 2dTV (L (Z), L (Z 0 )). (12.3)
We will need one more observation. Suppose that a random variable X has dis-
tribution the Bin(n, p), while a random variable Y has the Poisson distribution,
and E X = EY . Then
dTV (X,Y ) = O(p). (12.4)
We leave the proofs of (12.2), (12.3) and (12.4) as exercises.
To prove Theorem 12.1 we also need some auxiliary results on a special coupon
collector scheme.
Let Z be a non-negative integer valued random variable, r a non-negative
integer and γ a real, such that rγ ≤ 1. Assume we have r coupons Q1 , Q2 , . . . , Qr
236 Intersection Graphs

and one blank coupon B. We make Z independent draws (with replacement),

such that in each draw,
P(Qi is chosen) = γ, for i = 1, 2, . . . , r,
and
P(B is chosen) = 1 − rγ.
Let Ni (Z), i = 1, 2, . . . , r be a random variable counting the number of times
that coupon Qi was chosen. Furthermore, let

1 if Ni (Z) ≥ 1,
Xi (Z) =
0 otherwise.
The number of different coupons selected is given by
r
X(Z) = ∑ Xi (Z). (12.5)
i=1

With the above definitions we observe that the following holds.

Lemma 12.2 If a random variable Z has the Poisson distribution with ex-
pectation λ then Ni (Z), i = 1, 2, . . . , r, are independent and identically Poisson
distributed random variables, with expectation λ γ. Moreover the random vari-
able X(Z) has the distribution Bin(r, 1 − e−λ γ ).

Let us consider the following special case of the scheme defined above,
n
n

assuming that r = 2 and γ = 1/ 2 . Here each coupon represents a distinct
edge of Kn .
Lemma 12.3 Suppose
p = o(1/n) and let a random variable Z be the
Bin m, n2 p2 (1 − p)n−2 distributed,


 while a random variable Y be the
n
−mp 2 (1−p)n−2
Bin 2 , 1 − e distributed. Then

dTV (L (X(Z)), L (Y )) = o(1).

Proof Let Z 0 be a Poisson random variable with the same expectation as Z,
i.e.,
 
n 2
E Z0 = m p (1 − p)n−2 .
2
By Lemma 12.2, X(Z 0 ) has the binomial distribution
  
n −mp2 (1−p)n−2
Bin ,1−e ,
2
 12.1 Binomial Random Intersection Graphs 237

and so, by (12.3) and (12.4), we have

dTV (L (Y ), L (X(Z)))
= dTV (L (X(Z 0 )), L (X(Z))) ≤ 2dTV (L (Z 0 ), L (Z))
  
n 2
p (1 − p)n−2 = O n2 p2 = o(1).


≤O
2

Now define a random intersection graph G2 (n, m, p) as follows. Its vertex set
is V = {1, 2, . . . , n}, while e = {i, j} is an edge in G2 (n, m, p) iff in a (generator)
bipartite random graph Gn,m,p , there is a vertex w ∈ M of degree two such that
both i and j are connected by an edge with w.
To complete the proof of our theorem, notice that,

dTV (L (G(n, m, p)), L (Gn, p̂ )) ≤

dTV (L (G(n, m, p)), L (G2 (n, m, p))) + dTV (L (G2 (n, m, p)), L (Gn, p̂ ))

where p̂ is defined in (12.1). Now, by (12.2)

dTV (L (G(n, m, p)), L (G2 (n, m, p)))

≤ P(L (G(n, m, p)) 6= L (G2 (n, m, p)))
 
n 3
≤ P(∃w ∈ M of Gn,m,p s.t. deg(w) > 2) ≤ m p = o(1),
3

for p = o(1/(nm1/3 ).
Hence it remains to show that

dTV (L (G2 (n, m, p)), L (Gn, p̂ )) = o(1). (12.6)

Let Z be distributed as Bin m, n2 p2 (1 − p)n−2 , X(Z)is defined as in (12.5)

2 n−2
and let Y be distributed as Bin n2 , 1 − e−mp (1−p)


. Then the number of
edges |E(G2 (n, m, p))| = X(Z) and |E(Gn, p̂ ))| = Y . Moreover for any two
graphs G and G0 with the same number of edges

P(G2 (n, m, p) = G) = P(G2 (n, m, p) = G0 )

and
P(Gn, p̂ = G) = P(Gn, p̂ = G0 ).

Equation (12.6) now follows from Lemma 12.3. The theorem follows immedi-
ately.
238 Intersection Graphs

For monotone properties (see Chapter 1) the relationship between the clas-
sical binomial random graph and the respective intersection graph is more pre-
cise and was established by Rybarczyk [683].

Theorem 12.4 Let 0 ≤ a ≤ 1, m = nα , α ≥ 3, Let P be any monotone graph

property. For α > 3, assume
p
Ω(1/(nm1/3 )) = p = O( log n/m)

while for α = 3 assume 1/(nm1/3 ) = o(p). Let

p̂ = 1 − exp −mp2 (1 − p)n−2 .

If for all ε = ε(n) → 0

P(Gn,(1+ε) p̂ ∈ P) → a,

then
P(G(n, m, p) ∈ P) → a

as n → ∞.

Small subgraphs
Let H be any fixed graph. A clique cover C is a collection of subsets of vertex
set V (H) such that, each induces a complete subgraph (clique) of H, and for
every edge {u, v} ∈ E(H), there exists C ∈ C , such that u, v ∈ C. Hence, the
cliques induced by sets from C exactly cover the edges of H. A clique cover is
allowed to have more than one copy of a given set. We say that C is reducible
if for some C ∈ C , the edges of H induced by C are contained in the union of
the edges induced by C \C, otherwise C is irreducible. Note that if C ∈ C and
C is irreducible, then |C| ≥ 2.
In this section, |C | stands for the number of cliques in C , while ∑ C denotes
the sum of clique sizes in C , and we put ∑ C = 0 if C = 0. /
Let C = {C1 ,C2 , . . . ,Ck } be a clique cover of H. For S ⊆ V (H) define the
following two restricted clique covers

Ct [S] := {Ci ∩ S : |Ci ∩ S| ≥ t, i = 1, 2, . . . , k},

where t = 1, 2. For a given S and t = 1, 2, let

 −1
τt = τt (H, C , S) = n|S|/ ∑ Ct [S] m|Ct [S]|/ ∑ Ct [S] .
 12.1 Binomial Random Intersection Graphs 239

Finally, let
τ(H) = min max {τ1 , τ2 },
C S⊆V (H)

where the minimum is taken over all clique covers C of H. We can in this cal-
culation restrict our attention to irreducible covers.
Karoński, Scheinerman and Singer-Cohen [482] proved the following theo-
rem.

Theorem 12.5 Let H be a fixed graph and mp2 → 0. Then

(
0 if p/τ(H) → 0
lim P(H ⊆ G(n, m, p)) =
n→∞ 1 if p/τ(H) → ∞.

As an illustration, we will use this theorem to show the threshold for com-
plete graphs in G(n, m, p), when m = nα , for different ranges of α > 0.

Corollary 12.6 For a complete graph Kh with h ≥ 3 vertices and m = nα , we

have
(
n−1 m−1/h for α ≤ 2h/(h − 1)
τ(Kh ) = −1/(h−1) −1/2
n m for α ≥ 2h/(h − 1).
Proof There are many possibilities for clique covers to generate a copy of a
complete graph Kh in G(n, m, p). However in the case of Kh only two play a
dominating role. Indeed, we will show that for α ≤ α0 , α0 = 2h/(h − 1) the
clique cover C = {V (Kh )} composed of one set containing
all h vertices of K
h
only matters, while for α ≥ α0 the clique cover C = K2h , consisting of h2
pairs of endpoints of the edges of Kh , takes the leading role.
Let V = V (Kh ) and denote those two clique covers by {V } and {E}, respec-
tively. Observe that for the cover {V } the following equality holds.

max{τ1 (Kh , {V }, S), τ2 (Kh , {V }, S)} = τ1 (Kh , {V },V ). (12.7)

S⊆V

To see this, check first that for |S| = h,

τ1 (Kh , {V },V ) = τ2 (Kh , {V },V ) = n−1 m−1/h .

For S of size |S| = s, 2 ≤ s ≤ h − 1 restricting the clique cover {V } to S, gives

a single s-clique, so for t = 1, 2

τt (Kh , {V }, S) = n−1 m−1/s < n−1 m−1/h .

240 Intersection Graphs

Finally, when |S| = 1, then τ1 = (nm)−1 , while τ2 = 0, both smaller than

n−1 m−1/h , and so equation (12.7) follows.
For the edge-clique cover {E} we have a similar expression, viz.
max{τ1 (Kh , {E}, S), τ2 (Kh , {E}, S)} = τ1 (Kh , {E},V ). (12.8)
S⊆V

To see this, check first that for |S| = h,

τ1 (Kh , {E},V ) = n−1/(h−1) m−1/2 .
Let S ⊂ V , with s = |S| ≤ h − 1, and consider restricted clique covers with
cliques of size at most two, and exactly two.
For τ1 , the clique cover restricted to S is the edge-clique cover of Ks , plus a
1-clique for each of the h − s external edges for each vertex of Ks , so
τ1 (Kh , {E}, S)
 −1
= ns/[s(s−1)+s(h−s)] m[s(s−1)/2+s(h−s)]/[s(s−1)+s(h−s)]
 −1
= n1/(h−1) m[h−(s+1)/2]/(h−1)
 −1
≤ n1/(h−1) mh/(2(h−1))
 −1
< n1/(h−1) m1/2 ,

while for τ2 we have

 −1  −1
τ2 (Kh , {E}, S) = n1/(s−1) m1/2 < n1/(h−1) m1/2 ,

thus verifying equation (12.8).

Let C be any irreducible clique cover of Kh (hence each clique has size at
least two). We will show that for any fixed α
(
τ1 (Kh , {V },V ) for α ≤ 2h/(h − 1)
τ1 (Kh , C ,V ) ≥
τ1 (Kh , {E},V ) for α ≥ 2h/(h − 1).
Thus,
τ1 (Kh , C ,V ) ≥ min {τ1 (Kh , {V },V ), τ1 (Kh , {E},V )} . (12.9)
Because m = nα we see that
τ1 (Kh , C ,V ) = n−xC (α) ,
where
h |C | α 1 α
xC (α) = + α, x{V } (α) = 1 + , x{E} (α) = + .
∑C ∑C h h−1 2
 12.1 Binomial Random Intersection Graphs 241

(To simplify notation, below we have replaced x{V } , x{E} by xV , xE , respec-

tively). Notice, that for α0 = 2h/(h − 1) exponents
2
xV (α0 ) = xE (α0 ) = 1 + .
h−1
Moreover, for all values of α < α0 the function xV (α) > xE (α), while for
α > α0 the function xV (α) < xE (α).
Now, observe that xC (0) = ∑hC ≤ 1 since each vertex is in at least one clique
of C . Hence xC (0) ≤ xV (0) = 1. We will show also that xC (α) ≤ xV (α) for
α > 0. To see this we need to bound |C |/ ∑ C .
Suppose that u ∈ V (Kh ) appears in the fewest number of cliques of C , and
let r be the number of cliques Ci ∈ C to which u belongs. Then

∑C = ∑ |Ci | + ∑ |Ci | ≥ ((h − 1) + r) + 2(|C | − r),

i:Ci 3u i:Ci 63u

where h − 1 counts all other vertices aside from u since they must appear in
some clique with u.
For any v ∈ V (Kh ) we have

∑ C + |{i : Ci 3 v}| − (h − 1) ≥ ∑ C + r − (h − 1)
≥ (h − 1) + r + 2(|C | − r) + r − (h − 1)
= 2|C |.

Summing the above inequality over all v ∈ V (Kh ),

h ∑ C + ∑ C − h(h − 1) ≥ 2h|C |,

and dividing both sides by 2h ∑ C , we finally get

|C | h+1 h−1
≤ − .
∑ C 2h 2∑C
Now, using the above bound,
 
h |C | 2h
xC (α0 ) = +
∑C ∑C h−1
  
h h+1 h−1 2h
≤ + −
∑C 2h 2∑C h−1
2
= 1+
h−1
= xV (α0 ).

Now, since xC (α) ≤ xV (α) at both α = 0 and α = α0 , and both functions are
linear, xC (α) ≤ xV (α) throughout the interval (0, α0 ).
242 Intersection Graphs

Since xE (α0 ) = xV (α0 ) we also have xC (α0 ) ≤ xE (α0 ). The slope of xC (α)
is ∑|CC| , and by the assumption that C consists of cliques of size at least 2,
this is at most 1/2. But the slope of xE (α) is exactly 1/2. Thus for all α ≥
α0 , xC (α) ≤ xE (α). Hence the bounds given by formula (12.9) hold.
One can show (see [681]) that for any irreducible clique-cover C that is not
{V } nor {E},

max{τ1 (Kh , C , S), τ2 (Kh , C , S)} ≥ τ1 (Kh , C ,V ).

S

Hence, by (12.9),

max{τ1 (Kh , C , S), τ2 (Kh , C , S)} ≥ min{τ1 (Kh , {V },V ), τ1 (Kh , {E},V )}.
S

This implies that

(
n−1 m−1/h for α ≤ α0
τ(Kh ) =
n−1/(h−1) m−1/2 for α ≥ α0 ,
which completes the proof of Corollary 12.6.
To add to the picture of asymptotic behavior of small cliques in
G(n, m, p) we will quote the result of Rybarczyk and Stark [681], who with
use of Stein’s method (see Chapter 21.3) obtained an upper bound on the total
variation distance between the distribution of the number of h-cliques and a
respective Poisson distribution for any fixed h.

Theorem 12.7 Let G(n, m, p) be a random intersection graph, where m = nα .

Let c > 0 be a constant and h ≥ 3 a fixed integer, and Xn be the random variable
counting the number of copies of a complete graph Kh in G(n, m, p).

(i) If α < 2h
h−1 , p ≈ cn−1 m−1/h then

λn = EXn ≈ ch /h!

and
 
dTV (L (Xn ), Po(λn )) = O n−α/h ;

(ii) If α = 2h
h−1 , p ≈ cn−(h+1)/(h−1) then
 
λn = EXn ≈ ch + ch(h−1) /h!

and
 
dTV (L (Xn ), Po(λn )) = O n−2/(h−1) ;
 12.2 Random Geometric Graphs 243

(iii) If α > 2h
h−1 , p ≈ cn−1/(h−1) m−1/2 then

λn = EXn ≈ ch(h−1) /h!

and
  α(h−1)  
2
h− 2 − h−1 −1
dTV (L (Xn ), Po(λn )) = O n +n .

12.2 Random Geometric Graphs

The graphs we consider in this section are the intersection graphs that we ob-
tain from the intersections of balls in the d-dimensional unit cube, D = [0, 1]d
where d ≥ 2. For simplicity we will only consider d = 2 in the text.
We let X = {X1 , X2 , . . . , Xn } be independently and uniformly chosen from
D = [0, 1]2 . For r = r(n) let GX ,r be the graph with vertex set X . We join
Xi , X j by an edge iff X j lies in the disk

B(Xi , r) = X ∈ [0, 1]2 : |X − Xi | ≤ r .



Here | | denotes Euclidean distance.

For a given set X we see that increasing r can only add edges and so thresh-
olds are usually expressed in terms of upper/lower bounds on the size of r.
The book by Penrose [637] gives a detailed exposition of this model. Our
aim here is to prove some simple results that are not intended to be best possi-
ble.

Connectivity
The threshold (in terms of r) for connectivity was shown to be identical with
that for minimum degree one, by Gupta and Kumar [395]. This was extended
to k-connectivity by Penrose [636]. We do not aim for tremendous accuracy.
The simple proof of connectivity was provided to us by Tobias Müller [609].
q
log n
Theorem 12.8 Let ε > 0 be arbitrarily small and let r0 = r0 (n) = πn .
Then w.h.p.

GX ,r contains isolated vertices if r ≤ (1 − ε)r0 (12.10)

GX ,r is connected if r ≥ (1 + ε)r0 (12.11)
244 Intersection Graphs

Proof First consider (12.10) and the degree of X1 . Then

P(X1 is isolated) ≥ (1 − πr2 )n−1 .

The factor (1 − πr2 )n−1 bounds the probability that none of X2 , X3 , . . . , Xn lie
in B(X1 , r), given that B(X1 , r) ⊆ D. It is exact for points far enough from the
boundary of D.
Now
(1 − ε) log n n
 
(1 − πr2 )n−1 ≥ 1 − = nε−1+o(1) .
n
So if I is the set of isolated vertices then E(|I|) ≥ nε−1+o(1) → ∞. Now
n−2
πr2

P(X1 ∈ I | X2 ∈ I) ≤ 1 − ≤ (1 + o(1)) P(X1 ∈ I).
1 − πr2
 
πr2
The expression 1 − 1−πr 2 is the probability that a random point does not
lie in B(X1 , r), given that it does not lie in B(X2 , r), and that |X2 − X1 | ≥ 2r.
Equation (12.10) now follows from the Chebyshev inequality (21.3).
Now consider (12.11). Let η  ε be a sufficiently small constant and divide
D into `20 sub-squares of side length ηr, where `0 = 1/ηr. We refer to these
sub-squares as cells. We can assume that η is chosen so that `0 is an integer.
We say that a cell is good if contains at least i0 = η 3 log n members of X and
bad otherwise. We next let K = 100/η 2 and consider the number of bad cells
in a K × K square block of cells.

Lemma 12.9 Let B be a K × K square block of cells. The following hold

w.h.p.:

(a) If B is further than 100r from the closest boundary edge of D then B con-
tains at most k0 = (1 − ε/10)π/η 2 bad cells.
(b) If B is within distance 100r of exactly one boundary edge of D then B
contains at most k0 /2 bad cells.
(c) If B is within distance 100r of two boundary edges of D then B contains
no bad cells.

Proof (a) There are less than `20 < n such blocks. Furthermore, the probability
that a fixed block contains k0 or more bad cells is at most
!k0
 2 i0 
K n 2 2 i 2 2 n−i
k0 ∑ (η r ) (1 − η r )
i=0 i
 12.2 Random Geometric Graphs 245
k0 i0 !k0
K2e
 
ne 2 2 i0 −η 2 r2 (n−i0 )
≤ 2 (η r ) e . (12.12)
k0 i0

Here we have used Corollary 22.4 to obtain the LHS of (12.12).

Now
 i0
ne 2 2
(η 2 r2 )i0 e−η r (n−i0 )
i0
3 log(1/η)−η 2 (1+ε−o(1))/π 2 (1+ε/2)/π
≤ nO(η ≤ n−η , (12.13)

for η sufficiently small. So we can bound the RHS of (12.12) by

2
!(1−ε/10)π/η 2
2K 2 en−η (1+ε/2)/π
≤ n−1−ε/3 . (12.14)
(1 − ε/10)π/η 2

Part (a) follows after inflating the RHS of (12.14) by n to account for the num-
ber of choices of block.
(b) Replacing k0 by k0 /2 replaces the LHS of (12.14) by
2
!(1−ε/10)π/2η 2
4K 2 en−η (1+ε/2)/π
≤ n−1/2−ε/6 . (12.15)
(1 − ε/10)π/2η 2

Observe now that the number of choices of block is O(`0 ) = o(n1/2 ) and then
Part (b) follows after inflating the RHS of (12.15) by o(n1/2 ) to account for the
number of choices of block.
(c) Equation (12.13) bounds the probability that a single cell is bad. The num-
ber of cells in question in this case is O(1) and (c) follows.
We now do a simple geometric computation in order to place a lower bound
on the number of cells within a ball B(X, r).
√
Lemma 12.10 A half-disk of radius r1 = r(1 − η 2) with diameter part of
the grid of cells contains at least (1 − 2η 1/2 )π/2η 2 cells.

Proof We place the half-disk in a 2r1 × r1 rectangle. Then we partition the

rectangle into ζ1 = r1 /rη rows of 2ζ1 cells. The circumference of the circle
will cut the ith row at a point which is r1 (1 −i2 η 2 )1/2 from the centre of the
row. Thus the ith row will contain at least 2 r1 (1 − i2 η 2 )1/2 /rη complete


cells. So the half-disk contains at least

2r1 1/η
Z 1/η−1
2r1
∑ ((1 − i2 η 2 )1/2 − η) ≥ ((1 − x2 η 2 )1/2 − η)dx
rη i=1 rη x=1
246 Intersection Graphs
Z arcsin(1−η)
2r1
= (cos2 (θ ) − η cos(θ ))dθ
rη 2 θ =arcsin(η)
 arcsin(1−η)
2r1 θ sin(2θ )
≥ 2 − −η .
rη 2 4 θ =arcsin(η)

Now
π
arcsin(1 − η) ≥ − 2η 1/2 and arcsin(η) ≤ 2η.
2
So the number of cells is at least
2r1  π 1/2

− η − η .
rη 2 4
This completes the proof of Lemma 12.10.
We deduce from Lemmas 12.9 and 12.10 that

X ∈ X implies that B(X, r1 ) ∩ D contains at least one good cell. (12.16)

Now let Γ be the graph whose vertex set consists of the good cells and where
cells c1 , c2 are adjacent iff their centres are within distance r1 . Note that if c1 , c2
are adjacent in Γ then any point in X ∩ c1 is adjacent in GX ,r to any point in
X ∩ c2 . It follows from (12.16) that all we need to do now is show that Γ is
connected.
It follows from Lemma 12.9 that at most π/η 2 rows of a K × K block contain
a bad cell. Thus more than 95% of the rows and of the columns of such a
block are free of bad cells. Call such a row or column good. The cells in a
good row or column of some K × K block form part of the same component
of Γ. Two neighboring blocks must have two touching good rows or columns
so the cells in a good row or column of some block form part of a single
component of Γ. Any other component C must be in a block bounded by good
rows and columns. But the existence of such a component means that it is
surrounded by bad cells and then by Lemma 12.10 that there is a block B
with at least (1 − 3η 1/2 )π/η 2 bad cells if it is far from the boundary and at
least half of this if it is close to the boundary. But this contradicts Lemma
12.9. To see this, consider a cell in C whose center c has the largest second
component i.e. is highest in C. Now consider the half disk H of radius r1 that
is centered at c. We can assume √ (i) H is contained entirely in B and (ii) at least
(1 − 2η 1/2 )π/2η 2 − (1 − η 2)/η ≥ (1 − 3η 1/2 )π/2η 2 cells in H are bad.
Property (i) arises because cells above c whose centers are at distance at most
r1 are all bad and for (ii) we have discounted any bad cells on the diameter
through c that might be in C. This provides half the claimed bad cells. We
obtain the rest by considering a lowest cell of C. Near the boundary, we only
 12.2 Random Geometric Graphs 247

need to consider one half disk with diameter parallel to the closest boundary.
Finally observe that there are no bad cells close to a corner.

Hamiltonicity
The first inroads on the Hamilton cycle problem were made by Diaz, Mitsche
and Pérez-Giménez [245]. Best possible results were later given by Balogh,
Bollobás, Krivelevich, Müller and Walters [50] and by Müller, Pérez-Giménez
and
Wormald [610]. As one might expect Hamiltonicity has a threshold at r close
to r0 . We now have enough to prove the result from [245].
We start with a simple lemma, taken from [50].

Lemma 12.11 The subgraph Γ contains a spanning tree of maximum degree

at most six.
Proof Consider a spanning tree T of γ that minimises the sum of the lengths
of the edges joining the centres of the cells. Then T does not have any vertex
of degree greater than 6. This is because, if centre v were to have degree at
least 7, then there are two neighboring centres u, w of v such that the angle
between the line segments [v, u] and [v, w] is strictly less than 60 degrees. We
can assume without loss of generality that [v, u] is shorter than [v, w]. Note that
if we remove the edge {v, w} and add the edge {u, w} then we obtain another
spanning tree but with strictly smaller total edge-length, a contradiction. Hence
T has maximum degree at most 6.

Theorem 12.12 Suppose that r ≥ (1 + ε)r0 . Then w.h.p. GX ,r is Hamilto-

nian.

Proof We begin with the tree T promised by Lemma 12.11. Let c be a good
cell. We partition the points of X ∩ c into 2d roughly equal size sets P1 , P2 , . . . ,
P2d where d ≤ 6 is the degree of c in T . Since, the points of X ∩c form a clique
in G = GX ,r we can form 2d paths in G from this partition.
We next do a walk W through T e.g. by Breadth First Search that goes through
each edge of T twice and passes through each node of Γ a number of times
equal to twice its degree in Γ. Each time we pass through a node we traverse
the vertices of a new path described in the previous paragraph. In this way we
create a cycle H that goes through all the points in X that lie in good cells.
248 Intersection Graphs

Now consider the points P in a bad cell c with centre x. We create a path in
G through P with endpoints x, y, say. Now choose a good cell c0 contained in
the ball B(x, r1 ) and then choose an edge {u, v} of H in the cell c0 . We merge
the points in P into H by deleting {u, v} and adding {x, u} , {y, v}. To make
this work, we must be careful to ensure that we only use an edge of H at most
once. But there are Ω(log n) edges of H in each good cell and there are O(1)
bad cells within distance 2r say of any good cell and so this is easily done.

Chromatic number
We look at the chromatic number of GX ,r in a limited range. Suppose that
nπr2 = log n
ωr where ωr → ∞, ωr = O(log n). We are below the threshold for
connectivity here. We will show that w.h.p.
χ(GX ,r ) ≈ ∆(GX ,r ) ≈ cl(GX ,r )
where will use cl to denote the size of the largest clique. This is a special case
of a result of McDiarmid [577].
We first bound the maximum degree.

Lemma 12.13
log n
∆(GX ,r ) ≈ w.h.p.
log ωr
Proof Let Zk denote the number of vertices of degree k and let Z≥k denote
the number of vertices of degree at least k. Let k0 = log n
ωd where ωd → ∞ and
ωd = o(ωr ). Then
    log n   log n
n 2 k0 neωd log n ωd eωd ωd
E(Z≥k0 ) ≤ n (πr ) ≤ n =n .
k0 nωr log n ωr
So,
log n
log(E(Z≥k0 )) ≤ (ωd + 1 + log ωd − log ωr ) . (12.17)
ωd
−1/2
Now let ε0 = ωr . Then if
ωd + log ωd + 1 ≤ (1 − ε0 ) log ωr
then (12.17) implies that E(Zk ) → 0. This verifies the upper bound on ∆ claimed
in the lemma.
Now let k1 = log n
b where ωd is the solution to
ω
b
d

bd + log ω
ω bd + 1 = (1 + ε0 ) log ωr .
 12.2 Random Geometric Graphs 249

Next let M denote the set of vertices that are at distance greater than r from
any edge of D. Let Mk be the set of vertices of degree k in M. If Zbk = |Mk | then
 
n−1
E(Zbk1 ) ≥ n P(X1 ∈ M) × (πr2 )k1 (1 − πr2 )n−1−k1 .
k1
P(X1 ∈ M) ≥ 1 − 4r. Using Lemma 22.1 we get

E(Zbk1 ) ≥
 k1
n (n − 1)e 2 /(1−πr2 )
(1 − 4r) 1/2
(πr2 )k1 e−nπr
3k1 k1
 log n
n1−1/ωr

eω
bd ω
bd
≥ (1 − o(1)) 1/2
.
3k1 ωr

So,

log(E(Zbk1 )) ≥
 
log n ω
bd − log ωr − d
b
− o(1) − O(log log n) + ωbd + 1 + log ω
ωbd ωr
  !
ε0 log n log ωr log n
=Ω =Ω 1/2
→ ∞.
ω
bd ωr
An application of the Chebyshev inequality finishes the proof of the lemma.
Indeed,

P(X1 , X2 ∈ Mk ) ≤ P(X1 ∈ M) P(X2 ∈ M)×

  2 !
n−1 2 k1 2 n−2k1 −2
P(X2 ∈ B(X1 , r)) + (πr ) (1 − πr )
k1
≤ (1 + o(1)) P(X1 ∈ Mk ) P(X2 ∈ Mk ).

Now cl(GX ,r ) ≤ ∆(GX ,r ) + 1 and so we now lower bound cl(GX ,r ) w.h.p.

But this is easy. It follows from Lemma 12.13 that w.h.p. there is a vertex X j
log n
with at least (1 − o(1)) log(4ω r)
vertices in its r/2 ball B(X j , r/2). But such a
log n
ball provides a clique of size (1 − o(1)) log(4ω r)
. We have therefore proved

log n
Theorem 12.14 Suppose that nπr2 = ωr where ωr → ∞, ωr = O(log n).
Then w.h.p.
log n
χ(GX ,r ) ≈ ∆(GX ,r ) ≈ cl(GX ,r ) ≈ .
log ωr
250 Intersection Graphs

We now consider larger r.

Theorem 12.15 Suppose that nπr2 = ωr log n where ωr → ∞, ωr = o(n/ log n).
Then w.h.p.
√
ωr 3 log n
χ(GX ,r ) ≈ .
2π
Proof First consider the triangular lattice in the plane. This√
is the set of points
T = {m1 a + m2 b : m1 , m2 ∈ Z} where a = (0, 1), b = (1/2, 3/2), see Figure
12.1.

Figure 12.1 The small hexagon is an example of a Cv .

As in the diagram, each v ∈ T can be placed at the centre of a √ hexagon Cv .

The Cv ’s intersect on
√a set of measure zero and each Cv has area 3/2 and is
contained in B(v, 1/ 3). Let Γ(T, d) be the graph with vertex set T where two
vertices x, y ∈ T are joined by an edge if their Euclidean distance |x − y| < d.

Lemma 12.16 [McDiarmid and Reed [579]]

χ(Γ(T, d)) ≤ (d + 1)2 .
Proof Let δ = dde. Let R denote a δ × δ rhombus made up of triangles of
T with one vertex at the origin. This rhombus has δ 2 vertices, if we exclude
those at the top and right hand end. We give each of these vertices a distinct
color and then tile the plane with copies of R. This is a proper coloring, by
construction.
 12.2 Random Geometric Graphs 251

Armed with this lemma we can easily get an upper bound on χ(GX ,r ). Let δ =
1/3
1/ωr and let s = δ r. Let sT be the contraction of the lattice T by a factor s
i.e. sT = {sx : x ∈ T }. Then if v ∈ sT let sCv be the hexagon with centre v, sides
√ sides of Cv but reduced by a factor s. |X ∩ sCv | is distributed
parallel to the
as Bin(n, s2 3/2). So the Chernoff bounds imply that with probability 1 −
o(n−1 ),
l
−1/8 √ m
sCv contains ≤ θ = (1 + ωr )ns2 3/2 members of X . (12.18)
√
Let ρ = r + 2s/ 3. We note that if x ∈ Cv and y ∈ Cw and |x − y| ≤ r then
|v − w| ≤ ρ. Thus, given a proper coloring ϕ of Γ(sT, ρ) with colors [q] we can
w.h.p. extend it to a coloring ψ of GX ,r with color’s [q] × [θ ]. If x ∈ sCv and
ϕ(x) = a then we let ψ(x) = (a, b) where b ranges over [θ ] as x ranges over
sCv ∩ X . So, w.h.p.
ρ 2
χ(GX ,r ) ≤ θ χ(Γ(sT, ρ)) = θ χ(Γ(T, ρ/s)) ≤ θ +1 ≈
√ s √
2
ns 3 r 2 ωr 3 log n
× 2 = . (12.19)
2 s 2π
For the lower bound we use a classic result on packing disks in the plane.

Lemma 12.17 Let An = [0, n]2 and C be a collection

√ of disjoint disks of unit
area that touch An . Then |C | ≤ (1 + o(1))πn2 / 12.
Proof Thue’s theorem states that the densest packing of disjoint same √ size
disks in the plane is the hexagonal packing which has density λ = π/ 12. Let
C 0 denote the disks that are contained entirely in An . Then we have
πn2
|C 0 | ≥ |C | − O(n) and |C 0 | ≤ √ .
12
The first inequality comes from the fact that if C ∈ C \ C 0 then it is contained
in a perimeter of width O(1) surrounding An .
Now consider the subgraph H of GX ,r induced by the points of X that belong
to the square with centre (1/2, 1/2) and side 1 − 2r. It follows from Lemma
√ is the size of the largest independent set in H then α(H) ≤
12.17 that if α(H)
(1+o(1))2/r2 3. This is because if S is an independent set of H then the disks
B(x, r/2) for x ∈ S are necessarily disjoint. Now using the Chernoff bounds, we
see that w.h.p. H contains at least (1 − o(1))n vertices. Thus
√ √
|V (H)| r2 3n ωr 3 log n
χ(GX ,r ) ≥ χ(H) ≥ ≥ (1 − o(1)) = (1 − o(1)) .
α(H) 2 2π
 252 Intersection Graphs

This completes the proof of Theorem 12.15.

12.3 Exercises
√
12.3.1 Show that if p = ω(n)/ (n m), and ω(n) → ∞, then G(n, m, p) has
w.h.p. at least one edge.
12.3.2 Show that if p = (2 log n + ω(n))/m)1/2 and ω(n) → −∞ then w.h.p.
G(n, m, p) is not complete.
12.3.3 Prove that the bound (12.2) holds.
12.3.4 Prove that the bound (12.3) holds.
12.3.5 Prove that the bound (12.4) holds.
12.3.6 Prove the claims in Lemma 12.2.
12.3.7 Let X denotes the number of isolated vertices in the binomial random
intersection graph G(n, m, p), where m = nα , α > 0. Show that if
(
(log n + ϕ(n))/m when α ≤ 1
p= p
(log n + ϕ(n))/(nm) when α > 1,

then E X → e−c if limn→∞ ϕ(n) → c, for any real c.

12.3.8 Find the variance of the random variable X counting isolated vertices
in G(n, m, p).
12.3.9 Let Y be a random variable which counts vertices of degree greater than
one in G(n, m, p), with m = nα and α > 1. Show that for p2 m2 n  log n

lim P Y > 2p2 m2 n = 0.

n→∞

12.3.10 Suppose that r ≥ (1 + ε)r0 , as in Theorem 12.8. Show that if 1 ≤ k =

O(1) then GX ,r is k-connected w.h.p.
k
− 2(k−1)
12.3.11 Show that if 2 ≤ k = O(1) and r  n then w.h.p. GX ,r contains
k
− 2(k−1)
a k-clique. On the other hand, show that if r = o(n ) then GX ,r
contains no k-clique.
q
12.3.12 Suppose that r  logn n . Show that w.h.p. the diameter of GX ,r =
Θ 1r .

12.3.13 Suppose that r ≥ (1 + ε)r0 , as in Theorem 12.8. Show that if 2 ≤ k =

O(1) then GX ,r has k edge disjoint Hamilton cycles w.h.p.
12.3.14 Given X and an integer k we define the k-nearest neighbor graph
Gk−NN,X as follows: We add an edge between x and y of X iff y is one
of x’s k nearest neighbors, in Euclidean distance or vice-versa. Show
 12.4 Notes 253

that if k ≥ C log n for a sufficiently large C then Gk−NN,X is connected

w.h.p.
12.3.15 Suppose that we independently deposit n random black points Xb and
n random white points Xw into D. Let BXb ,Xw ,r be the bipartite graph
where
q we connect x ∈ Xb with Xw iff |x − y| ≤ r. Show that if r 
log n
n then w.h.p. BXb ,Xw ,r contains a perfect matching.

12.4 Notes
Binomial Random Intersection Graphs
For G(n, m, p) with m = nα , α constant, Rybarczyk and Stark [682] provided
a condition, called strictly α-balanced for the Poisson convergence for the
number of induced copies of a fixed subgraph, thus complementing the re-
sults of Theorem 12.5 and generalising Theorem 12.7. (Thresholds for small
subgraphs in a related model of random intersection digraph are studied by
Kurauskas [530]).
Rybarczyk [684] introduced a coupling method to find thresholds for many
properties of the binomial random intersection graph. The method is used to
establish sharp threshold functions for k-connectivity, the existence of a perfect
matching and the existence of a Hamilton cycle.
Stark [713] determined the distribution of the degree of a typical vertex of
G(n, m, p), m = nα and showed that it changes sharply between α < 1, α = 1
and α > 1.
Behrisch [67] studied the evolution of the order of the largest component in
G(n, m, p), m = nα when α 6= 1. He showed that when α > 1 the random graph
G(n, m, p) behaves like Gn,p in that a giant component of size order n appears
w.h.p. when the expected vertex degree exceeds one. This is not the case when
α < 1. There is a jump in the order of size of the largest component, but not to
one of linear size. Further study of the component structure of G(n, m, p) for
α = 1 is due to Lageras and Lindholm in [532].
Behrisch, Taraz and Ueckerdt [68] study the evolution of the chromatic number
of a random intersection graph and showed that, in a certain range of parame-
ters, these random graphs can be colored optimally with high probability using
various greedy algorithms.

Uniform Random Intersection Graphs

Uniform random intersection graphs differ from the binomial random inter-
254 Intersection Graphs

section graph in the way a subset of the set M is defined for each vertex of V .
Now for every k = 1, 2, . . . , n, each Sk has fixed size r and is randomly chosen
from the set M. We use the notation G(n, m, r) for an r-uniform random inter-
section graph. This version of a random intersection graph was introduced by
Eschenauer and Gligor [288] and, independently, by Godehardt and Jaworski
[380].
Bloznelis, Jaworski and Rybarczyk [102] determined the emergence of the gi-
ant component in G(n, m, r) when n(log n)2 = o(m). A precise study of the
phase transition of G(n, m, r) is due to Rybarczyk [685]. She proved that if
c > 0 is a constant, r = r(n) ≥ 2 and r(r − 1)n/m ≈ c, then if c < 1 then w.h.p.
the largest component of G(n, m, r) is of size O(log n), while if c > 1 w.h.p.
there is a single giant component containing a constant fraction of all vertices,
while the second largest component is of size O(log n).
The connectivity of G(n, m, r) was studied by various authors, among them by
Eschenauer and Gligor [288] followed by DiPietro, Mancini, Mei, Panconesi
and Radhakrishnan [251],
Blackbourn and Gerke [91] and Yagan and Makowski [753]. Finally, Rybar-
czyk [685] determined the sharp threshold for this property. She proved that
if c > 0 is a constant, ω(n) → ∞ as n → ∞ and r2 n/m = log n + ω(n), then
similarly as in Gn,p , the uniform random intersection graph G(n, m, r) is dis-
connected w.h.p. if ω(n) → ∞, is connected w.h.p. if ω(n) → ∞, while the
−c
probability that G(n, m, r) is connected tends to e−e if ω(n) → c. The Hamil-
tonicity of G(n, m, r) was studied in [105] and by Nicoletseas, Raptopoulos
and Spirakis [625].
If in the uniform model we require |Si ∩S j | ≥ s to connect vertices i and j by an
edge, then we denote this random intersection graph by Gs (n, m, r). Bloznelis,
Jaworski and Rybarczyk [102] studied phase transition in Gs (n, m, r). Bloznelis
and Łuczak [104] proved that w.h.p. for even n the threshold for the property
that Gs (n, m, r) contains a perfect matching is the same as that for Gs (n, m, r)
being connected. Bloznelis and Rybarczyk [106] show that w.h.p. the edge
density threshold for the property that each vertex of Gs (n, m, r) has degree at
least k is the same as that for Gs (n, m, r) being k-connected (for related results
see [758]).

Generalized Random Intersection Graphs

Godehardt and Jaworski [380] introduced a model which generalizes both the
binomial and uniform models of random intersection graphs. Let P be a prob-
ability measure on the set {0, 1, 2, . . . , m}. Let V = {1, 2, . . . , n} be the vertex
set. Let M = {1, 2, . . . , m} be the set of attributes. Let S1 , S2 , . . . , Sn be inde-
 12.4 Notes 255

pendent random subsets
of M such that for any v ∈ V and S ⊆ M we have

m
P(Sv = S) = P(|S|)/ |S| . If we put an edge between any pair of vertices i
and j when Si ∩ S j 6= 0, / then we denote such a random intersection graph
as G(n, m, P), while if the edge is inserted if |Si ∩ Si | ≥ s, s ≥ 1, the respec-
tive graph is denoted as Gs (n, m, P). Bloznelis [95] extends these definitions to
random intersection digraphs.
The study of the degree distribution of a typical vertex of G(n, m, P) is given
in [455], [234] and [93], see also [456]. Bloznelis ( see [94] and [96]) shows
that the order of the largest component L1 of G(n, m, P) is asymptotically equal
to nρ, where ρ denotes the non-extinction probability of a related multi-type
Poisson branching process. Kurauskas and Bloznelis [531] study the asymp-
totic order of the clique number of the sparse random intersection graph
Gs (n, m, P).
Finally, a dynamic approach to random intersection graphs is studied by Bar-
bour and Reinert [62], Bloznelis and Karoński [103], Bloznelis and Goetze
[100] and Britton, Deijfen, Lageras and Lindholm [161].
One should also notice that some of the results on the connectivity of random
intersection graphs can be derived from the corresponding results for random
hyperghraphs, see for example [509], [695] and [381].

Inhomogeneous Random Intersection Graphs

Nicoletseas, Raptopoulos and Spirakis [624] have introduced a generalisation
of the binomial random intersection graph G(n, m, p) in the following way.
As before let n, m be positive integers and let 0 ≤ pi ≤ 1, i = 1, 2, . . . , m. Let
V = {1, 2, . . . , n} be the set of vertices of our graph and for every 1 ≤ k ≤ n,
let Sk be a random subset of the set M = {1, 2, . . . , m} formed by selecting ith
element of M independently with probability pi . Let p = (pi )m i=1 . We define the
inhomogeneous random intersection graph G(n, m, p) as the intersection graph
of sets Sk , k = 1, 2, . . . n. Here two vertices i and j are adjacent in G(n, m, p)
if and only if Si ∩ S j 6= 0. / Several asymptotic properties of the random graph
G(n, m, p) were studied, such as: large independent sets (in [625]), vertex de-
gree distribution (by Bloznelis and Damarackas in [97]), sharp threshold func-
tions for connectivity, matchings and Hamiltonian cycles (by Rybarczyk in
[684]) as well as the size of the largest component (by Bradonjić, Elsässer,
Friedrich, Sauerwald and Stauffer in [158]).
To learn more about different models of random intersection graphs and about
other results we refer the reader to recent review papers [98] and [99].
256 Intersection Graphs

Random Geometric Graphs

McDiarmid and Müller [578] gives the leading constant for the chromatic num-
ber when the average degree is Θ(log n). The paper also shows a “surprising”
phase change for the relation between χ and ω. Also the paper extends the
setting to arbitrary dimensions. Müller [608] proves a two-point concentration
for the clique number and chromatic number when nr2 = o(log n).
Blackwell, Edmonson-Jones and Jordan [92] studied the spectral properties of
the adjacency matrix of a random geometric graph (RGG). Rai [660] studied
the spectral measure of the transition matrix of a simple random walk. Preci-
ado and Jadbabaie [653] studied the spectrum of RGG’s in the context of the
spreading of viruses.
Sharp thresholds for monotone properties of RGG’s were shown by McColm
[572] in the case d = 1 viz. a graph defined by the intersection of random
sub-intervals. And for all d ≥ 1 by Goel, Rai and Krishnamachari [382].
First order expressible properties of random points
X = {X1 , X2 , . . . , Xn } on a unit circle were studied by McColm [571]. The
graph has vertex set X and vertices are joined by an edge if and only if their
angular distance is less than some parameter d. He showed among other things
that for each fixed d, the set of a.s. FO sentences in this model is a complete
non-categorical theory. McColm’s results were anticipated in a more precise
paper [379] by Godehardt and Jaworski, where the case d = 1, i.e., the evolu-
tion a random interval graph, was studied.
Diaz, Penrose, Petit and Serna [248] study the approximability of several lay-
out problems on a family of RGG’s. The layout problems that they consider
are bandwidth, minimum linear arrangement, minimum cut width, minimum
sum cut, vertex separation, and edge bisection. Diaz, Grandoni and Marchetti-
Spaccemela [247] derive a constant expected approximation algorithm for the
β -balanced cut problem on random geometric graphs: find an edge cut of min-
imum size whose two sides contain at least β n vertices each.
Bradonjić, Elsässer, Friedrich, Sauerwald and Stauffer [157] studied the broad-
cast time of RGG’s. They study a regime where there is likely to be a sin-
gle giant component and show that w.h.p. their broadcast algorithm only re-
quires O(n1/2 /r + log n) rounds to pass information from a single vertex, to
every vertex of the giant. They show on the way that the diameter of the giant
is Θ(n1/2 /r) w.h.p. Friedrich, Sauerwald and Stauffer [325] extended this to
higher dimensions.
A recent interesting development can be described as Random Hyperbolic
Graphs. These are related to the graphs of Section 12.2 and are posed as mod-
els of real world networks. Here points are randomly embedded into hyper-
 12.4 Notes 257

bolic, as opposed to Euclidean space. See for example Bode, Fountoulakis and
Müller [107], [108]; Candellero and Fountoulakis [172]; Chen, Fang, Hu and
Mahoney [181]; Friedrich and Krohmer [324]; Krioukov, Papadopolous, Kit-
sak, Vahdat and Boguñá [512]; Fountoulakis [314]; Gugelmann, Panagiotou
and Peter [394]; Papadopolous, Krioukov, Boguñá and Vahdat [633]. One ver-
sion of this model is described in [314]. The models are a little complicated to
describe and we refer the reader to the above references.
 13
Digraphs

In graph theory, we sometimes orient edges to create a directed graph or di-

graph. It is natural to consider randomly generated digraphs and this chapter
discusses the component size and connectivity of the simplest model Dn,p .
Hamiltonicity is discussed in the final section.

13.1 Strong Connectivity

In this chapter we study the random digraph Dn,p . This has vertex set [n] and
each of the n(n − 1) possible edges occurs independently with probability p.
We will first study the size of the strong components of Dn,p .
Recall the definition of strong components: Given a digraph D = (V, A) we
define the relation ρ on V by xρy if there is a path from x to y in D and there is
a path from y to x in D. It is easy to show that ρ is an equivalence relation and
the equivalence classes are called the strong components of D.

Strong component sizes: sub-critical region.

Theorem 13.1 Let p = c/n, where c is a constant, c < 1. Then w.h.p.

(i) all strong components of Dn,p are either cycles or single vertices,
(ii) the number of vertices on cycles is at most ω, for any ω = ω(n) → ∞.

Proof The expected number of cycles is

n   n
n  c k ck
∑ k (k − 1)! ≤ ∑ = O(1).
k=2 n k=2 k

Part (ii) now follows from the Markov inequality.

To tackle (i) we observe that if there is a component that is not a cycle or a
single vertex then there is a cycle C and vertices a, b ∈ C and a path P from a
to b that is internally disjoint from C.
However, the expected number of such subgraphs is bounded by

258
 13.1 Strong Connectivity 259
n n−k    c k n  c l+1
n
∑∑ k (k − 1)! k2 l!
k=2 l=0 n l n
∞ ∞
k2 ck+l+1
≤ ∑∑ = O(1/n).
k=2 l=0 kn

Here l is the number of vertices on the path P, excluding a and b.

Strong component sizes: super-critical region.

We will prove the following beautiful theorem that is a directed analogue of
the existence of a giant component in Gn,p . It is due to Karp [484].

Theorem 13.2 Let p = c/n, where c is a constant, c > 1, and let x be defined
by x < 1 and xe−x = ce−c . Then w.h.p. Dn,p contains a unique strong com-

2
ponent of size ≈ 1 − xc n. All other strong components are of logarithmic
size.

We will prove the above theorem through a sequence of lemmas.

For a vertex v vet

D+ (v) ={w : ∃ path v to w in Dn,p }

D− (v) ={w : ∃ path w to v in Dn,p }.

We will first prove

Lemma 13.3 There exist constants α, β , dependent only on c, such that w.h.p.
6 ∃ v such that |D± (v)| ∈ [α log n, β n].

Proof If there is a v such that |D+ (v)| = s then Dn,p contains a tree T of size
s, rooted at v such that

(i) all arcs are oriented away from v, and

(ii) there are no arcs oriented from V (T ) to [n] \V (T ).

The expected number of such trees is bounded above by

 
n s−2  c s−1  c s(n−s) n  1−c+s/n s
s s 1− ≤ ce .
s n n cs
260 Digraphs

Now ce1−c < 1 for c 6= 1 and so there exists β such that when s ≤ β n we can
bound ce1−c+s/n by some constant γ < 1 (γ depends only on c). In which case
n s 4
γ ≤ n−3 for log n ≤ s ≤ β n.
cs log 1/γ

Fix a vertex v ∈ [n] and consider a directed breadth first search from v. Let
S0+ = S0+ (v) = {v} and given S0+ , S1+ = S1+ (v), . . . , Sk+ = s+ +
k (v) ⊆ [n] let Tk =
+ Sk +
Tk (v) = i=1 Si and let
+
= w 6∈ Tk+ : ∃x ∈ Tk+ such that (x, w) ∈ E(Dn,p ) .

Sk+1

We similarly define S0− = S0− (v), S1− = S1− (v), . . . , Sk− = Sk− , Tk− (v) ⊆ [n] with
respect to a directed breadth first search into v.
Not surprisingly, we can show that the subgraph Γk induced by Tk+ is close
in distribution to the tree defined by the first k + 1 levels of a Galton-Watson
branching process with Po(c) as the distribution of the number of offspring
from a single parent. See Chapter 24 for some salient facts about such a pro-
cess. Here Po(c) is the Poisson random variable with mean c i.e.
ck e−c
P(Po(c) = k) = for k = 0, 1, 2, . . . , .
k!

Lemma 13.4 If Sˆ0 , Sˆ1 , . . . , Sˆk and T̂k are defined with respect to the Galton-
Watson branching process and if k ≤ k0 = (log n)3 and s0 , s1 , . . . , sk ≤ (log n)4
then
  
1
P |Si+ | = si , 0 ≤ i ≤ k = 1 + O 1−o(1)



P |Ŝi | = si , 0 ≤ i ≤ k .
n

Proof We use the fact that if Po(a), Po(b) are independent then
Po(a) + Po(b) has the same distribution as Po(a + b). It follows that

k
(csi−1 )si e−csi−1
P |Ŝi | = si , 0 ≤ i ≤ k = ∏ .
i=1 si !
+
Furthermore, putting ti−1 = s0 + s1 + . . . + si−1 we have for v ∈
/ Ti−1 ,

(log n)7
  
P(v ∈ Si+ ) = 1 − (1 − p)si−1 = si−1 p 1 + O . (13.1)
n
 13.1 Strong Connectivity 261

P |Si+ | = si , 0 ≤ i ≤ k =


(13.2)
k     7
si
n − ti−1 si−1 c (log n)
=∏ 1+O
i=1 si n n
n−ti−1 −si
(log n)7
  
si−1 c
× 1− 1+O
n n

Here we use the fact that given si−1 ,ti−1 , the distribution of |Si+ | is the bino-
mial with n − ti−1 trials and probability of success given in (13.1). The lemma
follows by simple estimations.

Lemma 13.5 For 1 ≤ i ≤ (log n)3

(a) P |Si+ | ≥ s log n||Si−1

+
| = s ≤ n−10

(b) P |Ŝi | ≥ s log n||Ŝi−1 | = s ≤ n−10 .

Proof

(a) P |Si+ | ≥ s log n||Si−1

+


| = s ≤ P (Bin(sn, c/n) ≥ s log n)
  
sn c s log n
≤
s log n n
 s log n
snec
≤
sn log n
ec log n
 
≤
log n
≤ n−10 .

The proof of (b) is similar.

Keeping v fixed we next let

F = ∃ i : |Ti+ | > (log n)2



= ∃ i ≤ (log n)2 : |T0+ |, |T1+ |, . . . , |Ti−1

+
| < (log n)2 < |Ti+ | .


Lemma 13.6
x
P(F ) = 1 − + o(1).
c
262 Digraphs

Proof Applying Lemma 13.4 we see that

P(F ) = P(Fˆ ) + o(1), (13.3)

where Fˆ is defined with respect to the branching process.

Now let Eˆ be the event that the branching process eventually becomes extinct.
We write
P(Fˆ ) = P(Fˆ |¬Eˆ ) P(¬Eˆ ) + P(Fˆ ∩ Eˆ ). (13.4)

To estimate (13.4) we use Theorem 24.1. Let

∞
ck e−c k
G(z) = ∑ k! z = ecz−c
k=0

be the probability generating function of Po(c). Then Theorem 24.1 implies

that ρ = P(Eˆ ) is the smallest non-negative solution to G(ρ) = ρ. Thus

ρ = ecρ−c .
ξ
Substituting ρ = c we see that

ξ ξ
P(Eˆ ) = where = eξ −c , (13.5)
c c
and so ξ = x.
The lemma will follow from (13.4) and (13.5) and P(Fˆ |¬Eˆ ) = 1 and

P(Fˆ ∩ Eˆ ) = o(1).

This in turn follows from

P(Eˆ | Fˆ ) = o(1), (13.6)

which will be established using the following lemma.

Lemma 13.7 Each member of the branching process has probability at least
ε > 0 of producing (log n)2 descendants at depth log n. Here ε > 0 depends
only on c.

Proof If the current population size of the process is s then the probability
that it reaches size at least c+1
2 s in the next round is

(cs)k e−cs
∑ ≥ 1 − e−αs
k!
k≥ c+1
2 s

for some constant α > 0 provided s ≥ 100, say.

 13.1 Strong Connectivity 263

Now there is a positive probability ε1 say that a single member spawns at least
100 descendants and so there is a probability of at least
!
∞
ε1 1 − ∑ e−αs
s=100

that a single object spawns

 log n
c+1
 (log n)2
2
descendants at depth log n.

Given a population size between (log n)2 and (log n)3 at level i0 , let si denote
the population size at level i0 + i log n. Then Lemma 13.7 and the Chernoff
bounds imply that
   
1 1
P si+1 ≤ εsi (log n)2 ≤ exp − ε 2 si (log n)2 .
2 8
It follows that
 i  !
1
P(Eˆ | Fˆ ) ≤ P ∃i : si ≤ 2 2

ε(log n) s0 s0 ≥ (log n)
2
(  i )
∞
1 2 1 2 2
≤ ∑ exp − ε ε(log n) (log n) = o(1).
i=1 8 2

This completes the proof (13.6) and of Lemma 13.6.

We must now consider the probability that both D+ (v) and D− (v) are large.

Lemma 13.8
x
P |D− (v)| ≥ (log n)2 | |D+ (v)| ≥ (log n)2 = 1 − + o(1).


c
Proof Expose S0+ , S1+ , . . . , Sk+ until either Sk+ = 0/ or we see that |Tk+ | ∈ [(log n)2 ,
(log n)3 ]. Now let S denote the set of edges/vertices defined by
S0+ , S1+ , . . . , Sk+ .
Let C be the event that there are no edges from Tl− to Sk+ where Tl− is the set of
vertices we reach through our BFS into v, up to the point where we first realise
that D− (v) < (log n)2 (because Si− = 0/ and |Ti− | ≤ (log n)2 ) or we realise that
D− (v) ≥ (log n)2 . Then
(log n)4
 
1
P(¬C ) = O = 1−o(1)
n n
264 Digraphs

and, as in (13.2),

P |Si− | = si , 0 ≤ i ≤ k | C =

k  0 si
(log n)7
  
n − ti−1 si−1 c
=∏ 1+O
i=1 si n n
n0 −ti−1 −si
(log n)7
  
si−1 c
× 1− 1+O
n n
where n0 = n − |Tk+ |.
Given this we can prove a conditional version of Lemma 13.4 and continue as
before.

We have now shown that if α is as in Lemma 13.3 and if

S = v : |D+ (v)|, |D− (v)| > α log n



then the expectation

 x 2
E(|S|) = (1 + o(1)) 1 − n.
c
We also claim that for any two vertices v, w

P(v, w ∈ S) = (1 + o(1)) P(v ∈ S) P(w ∈ S) (13.7)

and therefore the Chebyshev inequality implies that w.h.p.

 x 2
|S| = (1 + o(1)) 1 − n.
c
But (13.7) follows in a similar manner to the proof of Lemma 13.8.
All that remains of the proof of Theorem 13.2 is to show that

S is a strong component w.h.p. (13.8)

Recall that any v 6∈ S is in a strong component of size ≤ α log n and so the

second part of the theorem will also be done.
We prove (13.8) by arguing that

P ∃ v, w ∈ S : w 6∈ D+ (v) = o(1).


(13.9)

In which case, we know that w.h.p. there is a path from each v ∈ S to every
other vertex w 6= v in S.
To prove (13.9) we expose S0+ , S1+ , . . . , Sk+ until we find that
|Tk+ (v)| ≥ n1/2 log n. At the same time we expose S0− , S1− , . . . , Sl− until we find
that |Tl− (w)| ≥ n1/2 log n. If w 6∈ D+ (v) then this experiment will have tried at
 13.1 Strong Connectivity 265

2
least n1/2 log n times to find an edge from D+ (v) to D− (w) and failed every
time. The probability of this is at most
 c n(log n)2
1− = o(n−2 ).
n
This completes the proof of Theorem 13.2.

Threshold for strong connectivity

Here we prove

Theorem 13.9 Let ω = ω(n), c > 0 be a constant, and let p = log n+ω
n . Then

0

 if ω → −∞
−c
lim P(Dn,p is strongly connected) = e−2e if ω → c
n→∞ 
if ω → ∞.

1

= lim P(6 ∃ v s.t. d + (v) = 0 or d − (v) = 0)

n→∞

Proof We leave as an exercise to prove that


1

 if ω → −∞
−c
lim P(∃ v s.t. d + (v) = 0 or d − (v) = 0) = 1 − e−2e if ω → c
n→∞ 
if ω → ∞.

0

Given this, one only has to show that if ω 6→ −∞ then w.h.p. there does not
exist a set S such that (i) 2 ≤ |S| ≤ n/2 and (ii) E(S : S̄) = 0/ or E(S̄ : S) = 0/
and (iii) S induces a connected component in the graph obtained by ignoring
orientation. But, here with s = |S|,
n/2 
n s−2
P(∃ S) ≤ 2 ∑ s (2p)s−1 (1 − p)s(n−s)
s=2 s

2n n/2  ne s s−2 2 log n s −s(1−s/n) ωs/n

 
≤ ∑ s s n e
log n s=2 n
2n n/2 −(1−s/n) ω/n
≤ ∑ (2n e log n)s
log n s=2
= o(1).
 266 Digraphs

13.2 Hamilton Cycles

Existence of a Hamilton Cycle
Here we prove the following remarkable inequality: It is due to McDiarmid
[574]

Theorem 13.10

P(Dn,p is Hamiltonian) ≥ P(Gn,p is Hamiltonian)

Proof We consider an ordered sequence of random digraphs

Γ0 , Γ1 , Γ2 , . . . , ΓN , N = n2 defined as follows: Let e1 , e2 , . . . , eN be an enumer-

ation of the edges of the complete graph Kn . Each ei = {vi , wi } gives rise to
two directed edges → −
ei = (vi , wi ) and ←
e−i = (wi , vi ). In Γi we include → −
e j and
←−
e j independently of each other, with probability p, for j ≤ i. While for j > i
we include both or neither with probability p. Thus Γ0 is just Gn,p with each
edge {v, w} replaced by a pair of directed edges (v, w), (w, v) and ΓN = Dn,p .
Theorem 13.10 follows from

P(Γi is Hamiltonian) ≥ P(Γi−1 is Hamiltonian).

To prove this we condition on the existence or otherwise of directed edges

associated with e1 , . . . , ei−1 , ei+1 , . . . , eN . Let C denote this conditioning.
Either

(a) C gives us a Hamilton cycle without arcs associated with ei , or

(b) not (a) and ∃ a Hamilton cycle if at least one of → −
ei , ←
e−i is present, or
(c) 6 ∃ a Hamilton cycle even if both of →
−
ei , ←
e−i are present.

(a) and (c) give the same conditional probability of Hamiltonicity in Γi , Γi−1 .
In Γi−1 (b) happens with probability p. In Γi we consider two cases (i) exactly
one of →
−
ei , ←
e−i yields Hamiltonicity and in this case the conditional probability is
p and (ii) either of →−ei , ←
e−i yields Hamiltonicity and in this case the conditional
probability is 1 − (1 − p)2 > p.
Note that we will never require that both → −
ei , ←
e−i occur.

Theorem 13.10 was subsequently improved by Frieze [329], who proved the
equivalent of Theorem 6.5.
 13.2 Hamilton Cycles 267
log n+cn
Theorem 13.11 Let p = n . Then

0

 if cn → −∞
−c
lim P(Dn,p has a Hamilton cycle) = e−2e if cn → c
n→∞ 
if cn → ∞.

1

Number of Distinct Hamilton Cycles

Here we give an elegant result of Ferber, Kronenberg and Long [298].
 2 
Theorem 13.12 Let p = ω logn n . Then w.h.p. Dn,p contains eo(n) n!pn di-
rected Hamilton cycles.

Proof The upper bound follows from the fisrt moment method. Let XH de-
note the number of Hamilton cycles in D = Dn,p . Now E XH = (n − 1)!pn , and
therefore the Markov inequality implies that w.h.p. we have XH ≤ eo(n) n!pn .

For the lower bound let α := α(n) be a function tending slowly to infinity
with n. Let S ⊆ V (G) be a fixed set of size s, where s ≈ α log n 0
n and let V =
0
V \ S. Moreover, assume that s is chosen so that |V | is divisible by integer
` = 2α log n. From now on the set S will be fixed and we will use it for closing
Hamilton cycles. Our strategy is as follows: we first expose all the edges within
V 0 , and show that one can find the “correct” number of distinct families P
consisting of m := |V 0 |/` vertex-disjoint paths which span V 0 . Then, we expose
all the edges with at least one endpoint in S, and show that w.h.p. one can turn
“most” of these families into Hamilton cycles and that all of these cycles are
distinct.
We take a random partitioning V 0 = V1 ∪ . . . ∪ V` such that all the Vi ’s are of
size m. Let us denote by D j the bipartite graph with parts Vj and  V j+1 . Observe
log n
that D j is distributed as Gm,m,p , and therefore, since p = ω m , by Exercise
13.3.2, with probability 1 − n−ω(1) we conclude that D j contains (1 − o(1))m
edge-disjoint perfect matchings (in particular, a (1 − o(1))m regular subgraph).
The Van der Waerden conjecture proved by Egorychev [277] and by Falikman
[291] implies the following: Let G = (A ∪ B, E) be an r-regular bipartite graph
with part
sizes |A| = |B| = n. Then, the number of perfect matchings in G is at
n
least nr n!.
Applying this and the union bound, it follows that w.h.p. each D j contains at
least (1 − o(1))m m!pm perfect matchings for each j. Taking the union of one
perfect matching from each of the D j ’s we obtain a family P of m vertex
268 Digraphs

disjoint paths which spans V 0 . Therefore, there are

((1 − o(1))m m!pm )` = (1 − o(1))n−s (m!)` pn−s

distinct families P obtained from this partitioning in this manner. Since this
occurs w.h.p. we conclude (applying the Markov inequality to the number
of partitions for which the bound fails) that this bound holds for (1 − o(1))-
fraction of such partitions. Since there are (n−s)!` such partitions, one can find
(m!)
at least
(n − s)!
(1 − o(1)) `
(1 − o(1))n−s (m!)` pn−s
(m!)
= (1 − o(1))n−s (n − s)!pn−s = (1 − o(1))n n!pn

distinct families, each of which consists of exactly m vertex-disjoint paths of

size ` (for the last equality, we used the fact that s = o(n/ log n)).

We show next how to close a given family of paths into a Hamilton cycle. For
each such family P, let A := A(P) denote the collection of all pairs (sP ,tP )
where sP is a starting point and tP is the endpoint of a path P ∈ P, and define an
auxiliary directed graph D(A) as follows. The vertex set of D(A) is V (A) = S ∪
{zP = (sP ,tP ) : zP ∈ A}. Edges of D(A) are determined as follows: if u, v ∈ S and
(u, v) ∈ E(D) then (u, v) is an edge of D(A). The in-neighbors (out-neighbors)
of vertices zP in S are the in-neighbors of sP in D (out-neighbors of tP ). Lastly,
(zP , zQ ) is an edge of D(A) if (tP , sQ ) is an edge D.
Clearly D(A) is distributed as Ds+m,p , and that a Hamilton cycle in D(A) cor-
responds to a Hamilton cycle in D after adding the corresponding paths be-
tween each sP and tP . Now distinct families P 6= P 0 yield distinct Hamilton
cycles (to see this, just delete the vertices of S from the Hamilton cycle, to re-
cover the paths). Using Theorem 13.11 we see that for p = ω (log n/(s + m)) =
ω (log(s + m)/(s + m)), the probability that D(A) does not have a Hamilton cy-
cle is o(1). Therefore, using the Markov inequality we see that for almost all of
the families P, the corresponding auxiliary graph D(A) is indeed Hamiltonian
and we have at least (1 − o(1))n n!pn distinct Hamilton cycles, as desired.

13.3 Exercises
13.3.1 Let p = log n+(k−1)nlog log n+ω for a constant k = 1, 2, . . .. Show that w.h.p.
Dnp is k-strongly connected.
13.3.2 The Gale-Ryser theorem states: Let G = (A ∪ B, E) be a bipartite graph
 13.3 Exercises 269

with parts of sizes |A| = |B| = n. Then, G contains an r-factor if and

only if for every two sets X ⊆ A and Y ⊆ B, we have

eG (X,Y ) ≥ r(|X| + |Y | − n).

Show that if p = ω(log n/n) then with probability 1 − n−ω(1) , Gn,n,p

contains (1 − o(1))np edge disjoint perfect matchings.
13.3.3 Show that if p = ω((log n)2 /n) then w.h.p. Gn,p contains eo(n) n!pn dis-
tinct Hamilton cycles.
13.3.4 A tournament T is an orientation of the complete graph Kn . In a random
tournament, edge {u, v} is oriented from u to v with probability 1/2 and
from v to u with probability 1/2. Show that w.h.p. a random tournament
is strongly connected.
13.3.5 Let T be a random tournament. Show that w.h.p. the size of the largest
acyclic sub-tournament is asymptotic to 2 log2 n. (A tournament is acyclic
if it contains no directed cycles).
13.3.6 Suppose that 0 < p < 1 is constant. Show that w.h.p. the size of the
largest acyclic tournament contained in Dnp is asymptotic to 2 logb n
where b = 1/p.
13.3.7 Let mas(D) denote the number of vertices in the largest acyclic sub-
graph of a digraph D. Suppose that 0 < p < 1 is constant. Show that
w.h.p. mas(Dn,p ) ≤ 4log
log n 1
q where q = 1−p .

13.3.8 Consider the random digraph Dn obtained from Gn,1/2 by orienting

edge (i, j) from i to j when i < j. This can be viewed as a partial or-
der on [n] and is called a Random Graph Order. Show that w.h.p. Dn
contains a path of length at least 0.51n. (In terms of partial orders, this
bounds the height of the order).
10

if np ≥ log
13.3.9 Show that n
then w.h.p. Dn,p is 12 ± o(1) resilient, i.e.
1 1
2 − ε np ≤ ∆H ≤ 2 + ε np. (Hint: just tweak the proof of Theorem
9.3 so that the lemmas refer to digraphs.)
13.3.10 Let O represent an orientation of the edges of a Hamilton cycle. Show
that

P(Dn,p has a Hamilton cycle with orientation O) ≥

P(Gn,p has a Hamilton cycle).
270 Digraphs

13.4 Notes
Packing
The paper of Frieze [329] was in terms of the hitting time for a digraph process
Dt . It proves that the first time that the δ + (Gt ), δ − (Gt ) ≥ k is w.h.p. the time
when Gt has k edge disjoint Hamilton cycles. The paper of Ferber, Kronenberg
and Long [298] shows that if p = ω((log n)4 /n) then w.h.p. Dn,p contains (1 −
o(1))np edge disjoint Hamilton cycles.

Long Cycles
The papers by Hefetz, Steger and Sudakov [414] and by Ferber, Nenadov,
Noever, Peter and Škorić [301] study the local resilience of having a Hamilton
8
cycle. In particular, [301] proves that if p  (lognn) then w.h.p. one can delete
any subgraph H of Dn,p with maximum degree at most ( 12 − ε)np and still
leave a Hamiltonian subgraph.
Krivelevich, Lubetzky and Sudakov [518] proved that w.h.p. the random di-
graph Dn,p , p = c/n contains a directed cycle of length (1 − (1 + εc )e−c )n
where εc → 0 as c → ∞.
Cooper, Frieze and Molloy [225] showed that a random regular digraph with
indegree = outdegree = r is Hamiltonian w.h.p. iff r ≥ 3.

Connectivity
Cooper and Frieze [214] studied the size of the largest strong component in a
random digraph with a given degree sequence. The strong connectivity of an
inhomogeneous random digraph was studied by Bloznelis, Götze and Jaworski
in [101].
 14
Hypergraphs

In this chapter we discuss random k-uniform hypergraphs, where k ≥ 3. We

are concerned with the models Hn,p;k and Hn,m;k . For Hn,p;k we consider the
hypergraph with vertex set [n] in which each possible k-set in [n]


k is included
as an edge with probability p. In Hn,m;k the edge set is a random m-subset of
[n]

k . The parameter k is fixed and independent of n throughout this chapter.
Many of the properties of Gn,p and Gn,m have been generalized without too
much difficulty to these models of hypergraphs. Hamilton cycles have only re-
cently been tackled with any success. Surprisingly enough, in some cases it is
enough to use the Chebyshev inequality and we will describe these cases. We
then describe a remarkable new result concerning thresholds. It easily gives
the threshold for perfect matchings and a number of other structures. We be-
gin however with a more basic question. That is as to when is there a giant
component and when are Hn,m;k , Hn,p;k connected?

14.1 Component Size

c
We remind the reader that k ≥ 3 here. Suppose that p = and c is constant.
(n−1
k−1)
1
We will prove that if c < k−1 then w.h.p. the maximum component size is
1
O(log n) and if c > k−1 then w.h.p. there is a unique giant component of size
Ω(n). This generalises the main result of Chapter 2. We will assume then that
c
p = n−1 for the remainder of this section.
(k−1)
Many of the components of a sparse random graph are small trees. The corre-
sponding objects here are called hypertrees. The size of a hypertree is the num-
ber of edges that it contains. An edge is a hypertree of size one. We obtain a
hypertree of size k +1 by choosing a hypertree C of size k and a vertex v ∈ V (C)
and then adding a new edge {v, v2 , v3 , . . . , vk } where v2 , v3 , . . . , vk ∈
/ V (C).

271
272 Hypergraphs

A hypertree of size 5.

Lemma 14.1 A hypertree of size ` contains `(k − 1) + 1 vertices.

Proof By induction on `. It is clearly true if ` = 1 and if we increase the size
by one, then we add exactly k − 1 new vertices.
A proof of the following generalisation of Cayley’s formula for the number
of spanning hypertrees of the complete graph can be found in Selivanov [701]
and Sivasubramanian [707] .
Lemma 14.2 The number Nk (`) of distinct hypertrees with ` edges contained
in the complete k-uniform hypergraph on [(k − 1)` + 1] is given by
((k − 1)`)!
Nk (`) = ((k − 1)` + 1)`−1 .
`! ((k − 1)!)`
We now look into the structure of small components of Hn,p;k .
c 1
Lemma 14.3 Let p = where c 6= k−1 . Suppose that S ⊆ [n] and |S| =
(n−1
k−1)
s ≤ log4 n. Suppose also that S contains a hypertree C with t = (k − 1)` + 1
vertices and ` edges. Suppose that in addition, S contains a vertices and b
edges that are not part of C. Suppose also that there are no edges joining C to
V \ S. Then, for some A = A(c) > 0, we have that w.h.p.
(a) b ≤ 1 and b(k − 2) + 1 ≥ a + 1 ≥ b(k − 1).
(b) ` < A log n and either (i) a = b = 0 or (ii) b = 1.
(c) | {S : ` ≤ A log n, b = 1} | = O(logk+1 n).
Proof We bound the expected number of sets S that violate this by
   b !`+b !t (n−s)
k−1
n s s ((k − 1)`)! c c
t `−1 1 − (14.1)
`!((k − 1)!)` n−1 n−1



s a k k−1 k−1
` !`+b
 ne s sa+kb  k−2
` (k − 1)k−1 eo(1) c(k − 1)!
≤ t `−1
e−(c+o(1))t
s k!b ek−2+o(1) (k − 1)! nk−1
 14.1 Component Size 273
!`
skb−1 cb na+1−b(k−1) ea+1−c e1−(k−1)c+o(1) c(k − 1)k−1 `k−2t
≤
kb t sk−1
!`
skb−1 cb na+1−b(k−1) ea+1−c eo(1) e1−(k−1)c c(k − 1)k `k−1
≤
kb t ((k − 1)`)k−1
skb−1 cb na+1−b(k−1) ea+1−c  o(1) 1−(k−1)c `
= e ce (k − 1) (14.2)
kb t
Now ce−(k−1)c is maximised at c = 1/(k − 1) and for c 6= 1/(k − 1) we have
eo(1) ce1−(k−1)c (k − 1) ≤ 1 − εc .
We may therefore upper bound the expression in (14.2) by
na+1−b(k−1)+O(b log log n/ log n) e−εc ` ≤ n1−b+O(b log log n/ log n) e−εc ` . (14.3)
For the second expression, we used the fact that a ≤ b(k − 2). The second
expression in (14.3) tends to zero if b > 1 and so we can assume that b ≤ 1.
The first expression in (14.3) then tends to zero if a + 1 < b(k − 1) and this
verifies Part (a). Because a+1−b(k −1) ≤ 1, we see that Part (b) of the lemma
follows with A = 2/εc . Part (c) follows from the Markov inequality.
Lemma 14.4 W.h.p. there are no components in the range [A log n, log4 n] and
all but logk+1 n of the small components (of size at most A log n) are hypertrees.
Proof Now suppose that S is a small component of size s which is not a
hypertree and let C be the set of vertices of a maximal hypertree with t =
(k − 1)` + 1 vertices and ` edges that is a subgraph of S. (Maximal in the sense
that it is not contained in any other hypertree of S.) Lemma 14.3 implies that
w.h.p. there is at most one edge e in S that is not part of C but is incident with at
least two vertices of C. Furthermore, Lemma 14.3(c) implies that the number
of sets with b = 1 is O(logk+1 n) w.h.p.
1
Lemma 14.5 If c < k−1 then w.h.p. the largest component has size O(log n).
Proof Fix an edge e and do a Breadth First Search (BFS) on the edges starting
with e. We start with L1 = e and let Lt denote the number of vertices at depth
t in the search i.e the neighbors of Lt−1 that are not in Lt−1 . Then |Lt+1 | is
n


dominated by (k − 1)Bin |Lt | k−1 , p . So,
 
n
E(|Lt+1 | | Lt ) ≤ (k − 1)|Lt | p ≤ θ |Lt |
k−1
2 log n
where θ = (k − 1)c < 1. It follows that if t0 = log 1/θ then

/ ≤ nkθ t0 = o(1).
Pr(∃e : Lt0 6= 0)
274 Hypergraphs

So, w.h.p. there are at most t0 levels. Furthermore, if |Lt | ever reaches log2 n
then the Chernoff bounds imply that w.h.p. |Lt+1 | ≤ |Lt |. This implies that the
maximum size of a component is O(log3 n) and hence, by Lemma 14.4, the
maximum size is O(log n).
1
Lemma 14.6 If c > k−1 then w.h.p. there is a unique giant component of size
Ω(n) and all other components are of size O(log n).

Proof We consider BFS as in Lemma 14.5. We choose a vertex v and begin

to grow a component from it. Sometimes the component we grow has size
O(log n), in which case, we choose a new vertex, not yet seen in the search
and grow another component. We argue that w.h.p. (i) after O(log n) attempts,
we grow a component of size at least log2 n and (ii) this component is of size
Ω(n).
We say that we explore Lt \Lt−1 when we determine Lt+1 . With Lt as in Lemma
4
14.5, we see that if we have explored at most log
 n vertices then given Lt ,
|Lt+1 | dominates Yt = (k − 1)Bin |Lt | n−o(n)


k−1 , p − Xt where Xt is an over-
count due to vertices outside Lt that are in two edges containing vertices of Lt .
We have
   8 
n log n
E(Xt ) ≤ n|Lt |2 p=O .
k−3 n
Now
 
n − o(n)
E(Yt ) = (k − 1)|Lt | p = (1 − o(1))c|Lt | = θ |Lt | where θ > 1.
k−1
 

The Chernoff bounds applied to Bin |Lt | n−o(n)k−1 , p imply that

(θ − 1)2 |Lt |
   
1+θ
P Yt ≤ |Lt | ≤ exp − .
2 3(k − 1)
So, for t = O(log log n) we have
1+θ t
  
P |Lt | ≥
2
(
` )!
t (θ − 1)2 1+θ t log8 n
 
2
≥ ∏ 1 − exp − −O (14.4)
`=1 3(k − 1) n
≥ γ,

where γ > 0 isa positive constant.

 This lower bound of γ follows from the fact
2 1+θ `
(θ −1) ( 2 )
that ∑∞
`=1 exp − 3(k−1) converges, see Apostol [38], Theorem 8.55.
 14.1 Component Size 275
2 log log n
t
Note that if t = then 1+θ = log2 n. It follows that the probability
log 1+θ
2
2
we fail to grow a component of size log2 n after s attempts is at most (1 − γ)s .
Choosing s = log21/(1−γ)
log n
we see that after exploring O(log3 n) vertices, we find
a component of size at least log2 n, with probability 1 − n−(1−o(1)) .
We show next that with (conditional) probability 1 − O(n−(1−o(1)) ) the compo-
nent of size at least log2 n will in fact have at least n0 = n(k−1)/k log n vertices.
We handle Xt ,Yt exactly as above. Going back to (14.4), if we run BFS for
another O(log n) steps then, starting with |Lt0 | ≈ log2 n, we have

1+θ t 2
   
P |Lt0 +t | ≥ log n
2
(θ − 1)2 log2 n
t     9 
log n
≥ ∏ 1 − exp − −O
`=1 3(k − 1) n
= 1 − n−(1−o(1)) . (14.5)

It follows that w.h.p. Hn,p;k only contains components of size O(log n) and
Ω(no ). For this we use the fact that we only need to apply (14.5) less than n/n0
times.
We now prove that there is a unique giant component. This is a simple sprin-
kling argument. Suppose that we let (1 − p) = (1 − p1 )(1 − p2 ) where p2 =
1
ωnk−1
for some ω → ∞ slowly. Then we know from Lemma 14.4 that there is a
gap in component sizes for H(n, p1 , k). Now add in the second round of edges
with probability p2 . If C1 ,C2 are distinct components of size at least n0 then
the probability there is no C1 : C2 edge added is at most
( n0
n0

)
∑k−1 (n0
)( n0
) ∑k−1
i=1 i k−i
(1 − p2 ) i=1 i k−i ≤ exp − ≤
ωnk−1
( )
2k−1 logk n
 k−1 k 
2 n0
exp − k−1 = exp − .
ωn ω

So, w.h.p. all components of size at least n0 are merged into one component.

We now look at the size of this giant component. The fact that almost all small
1
components are hypertrees when c < k−1 yields the following lemma. The
proof follows that of Lemma 2.13 and is left as Exercise 13.4.5. For c > 0 we
define
(
1
c c ≤ k−1
x = x(c) = .
1
to xe−(k−1)x = ce−(k−1)c c > k−1 1


The solution in 0, k−1
276 Hypergraphs

Lemma 14.7 Suppose that c > 0 is constant and c(k − 1) 6= 1 and x is as

defined above. Then
∞ `
c ((k − 1)` + 1)` e−c((k−1)`+1)  x 1/(k−1)
∑ = .
`=0 `!((k − 1)` + 1) c
It follows from Lemmas 14.6 and 14.7 that we have
1
Lemma 14.8 If c > k−1 then w.h.p. there is a unique giant component of size
 
1/(k−1)
≈ 1 − xc


n.

The connectivity threshold for Hn,p;k coincides with minimum degree at least
one. The proof is straightforward and is left as Exercise 13.4.6.
log n+cn
Lemma 14.9 Let p = . Then
(n−1
k−1)

0

 cn → −∞.
−c
lim P(Hn,p;k is connected) = e−e cn → c.
n→∞ 
cn → ∞.

1

14.2 Hamilton Cycles

Suppose that 1 ≤ r < k. An r-overlapping Hamilton cycle C in a k-uniform
hypergraph H = (V, E ) on n vertices is a collection of mr = n/(k − r) edges of
H such that for some cyclic order of [n] every edge consists of k consecutive
vertices and for every pair of consecutive edges Ei−1 , Ei in C (in the natural
ordering of the edges) we have |Ei−1 ∩ Ei | = r. Thus, in every r-overlapping
Hamilton cycle the sets Ci = Ei \ Ei−1 , i = 1, 2, . . . , mr , are a partition of V
into sets of size k − r. Hence, mr = n/(k − r). We thus always assume, when
discussing r-overlapping Hamilton cycles, that this necessary condition, k − r
divides n, is fulfilled. In the literature, when r = k − 1 we have a tight Hamilton
cycle and when r = 1 we have a loose Hamilton cycle.
In this section we will restrict our attention to the case r = k − 1 i.e. tight
Hamilton cycles. So when we say that a hypergraph is Hamiltonian, we mean
that it contains a tight Hamilton cycle. The proof extends easily to r ≥ 2. The
case r = 1 poses some more problems and is discussed in Frieze [333], Dudek,
Frieze [256] and Dudek, Frieze, Loh and Speiss [258] and in Ferber [296].
Also, see Section 14.3. The following theorem is from Dudek and Frieze [257].
Furthermore, we assume that k ≥ 3.

Theorem 14.10
 14.2 Hamilton Cycles 277

(i) If p ≤ (1 − ε)e/n, then w.h.p. Hn,p;k is not Hamiltonian.

(ii) If k = 3 and np → ∞ then Hn,p;k is Hamiltonian w.h.p.
(iii) For all fixed ε > 0, if k ≥ 4 and p ≥ (1 + ε)e/n, then w.h.p. Hn,p;k is
Hamiltonian.

Proof We will prove parts (i) and (ii) and leave the proof of (iii) as an exer-
cise, with a hint.
Let ([n], E ) be a k-uniform hypergraph. A permutation π of [n] is Hamilton
cycle inducing if

Eπ (i) = {π(i − 1 + j) : j ∈ [k]} ∈ E f or all i ∈ [n].

(We use the convention π(n + r) = π(r) for r > 0.) Let the term hamperm refer
to such a permutation.
Let X be the random variable that counts the number of hamperms π for Hn,p;k .
Every Hamilton cycle induces at least one hamperm and so we can concentrate
on estimating P(X > 0).
Now
E(X) = n!pn .

This is because π induces a Hamilton cycle if and only if a certain n edges are
all in Hn,p;k .
For part (i) we use Stirling’s formula to argue that
√  np n √
E(X) ≤ 3 n ≤ 3 n(1 − ε)n = o(1).
e
This verifies part (i).
We see that
 np n
E(X) ≥ →∞ (14.6)
e
in parts (ii) and (iii).
Fix a hamperm π. Let H(π) = (Eπ (1), Eπ (2), . . . , Eπ (n)) be the Hamilton cycle
induced by π. Then let N(b, a) be the number of permutations π 0 such that
|E(H(π))∩E(H(π 0 ))| = b and E(H(π))∩E(H(π 0 )) consists of a edge disjoint
paths. Here a path is a maximal sub-sequence F1 , F2 , . . . , Fq of the edges of
H(π) such that Fi ∩ Fi+1 6= 0/ for 1 ≤ i < q. The set qj=1 Fj may contain other
S

edges of H(π). Observe that N(b, a) does not depend on π.

Note that
n b
E(X 2 ) n!N(0, 0)p2n n!N(b, a)p2n−b
= + ∑ ∑ E(X)2 .
E(X)2 E(X)2 b=1 a=1
278 Hypergraphs

Since trivially, N(0, 0) ≤ n!, we obtain,

n b
E(X 2 ) n!N(b, a)p2n−b
2
≤ 1+ ∑ ∑ . (14.7)
E(X) b=1 a=1 E(X)2
We show that
n b n b
n!N(b, a)p2n−b N(b, a)pn−b
∑ ∑ E(X)2 = ∑ ∑ E(X) = o(1). (14.8)
b=1 a=1 b=1 a=1

The Chebyshev inequality implies that

E(X 2 )
P(X = 0) ≤ − 1 = o(1),
E(X)2
as required.
It remains to show (14.8). First we find an upper bound on N(b, a). Choose a
vertices vi , 1 ≤ i ≤ a, on π. We have at most
na (14.9)
choices. Let
b1 + b2 + · · · + ba = b,
where bi ≥ 1 is an integer for every 1 ≤ i ≤ a. Note that this equation has
exactly
 
b−1
< 2b (14.10)
a−1
solutions. For every i, we choose a path of length bi in H(π) which starts at vi .
Suppose a path consists of edges F1 , F2 , . . . , Fq , q = bi . Assuming that F1 , . . . , Fj
are chosen, we have at most k possibilities for Fj+1 . Hence, every such a path
can be selected in most kbi ways. Consequently, we have at most
a
∏ kbi = kb
i=1

choices for all a paths.

Thus, by the above considerations we can find a edge disjoint paths in H(π)
with the total of b edges in at most
na (2k)b (14.11)
many ways.
Let P1 , P2 , . . . , Pa be any collection of the above a paths. Now we count the
number of permutations π 0 containing these paths.
First we choose for every Pi a sequence of vertices inducing this path in π 0 . We
 14.2 Hamilton Cycles 279

see the vertices in each edge of Pi in at most k! orderings. Crudely, every such
sequence can be chosen in at most (k!)bi ways. Thus, we have
a
∏(k!)bi = (k!)b (14.12)
i=1

choices for all a sequences.

Now we bound the number of permutations containing these
sequences. First note that

|V (Pi )| ≥ bi + k − 1.

Thus we have at most

a
n − ∑ (bi + k − 1) = n − b − a(k − 1) (14.13)
i=1

vertices not in V (P1 ) ∪ · · · ∪V (Pa ). We choose a permutation σ of V \ (V (P1 ) ∪

· · · ∪V (Pa )). Here we have at most

(n − b − a(k − 1))!

choices. Now we extend σ to a permutation of [n]. We mark a positions on σ

and then insert the sequences. We can do it in
 
n
a! < na
a
ways. Therefore, the number of permutations containing P1 , P2 , . . . , Pa is smaller
than
(k!)b (n − b − a(k − 1))!na . (14.14)

Thus, by (14.11) and (14.14) and the Stirling formula we obtain

N(b, a) < n2a (2k!k)b (n − b − a(k − 1))! <

√  n n−b−a(k−1)
n2a (2k!k)b 2πn .
e
Since
p  n n
E(X) = n!pn = (2 + o(1))πn pn ,
e
we get
N(b, a)pn−b  e b+a(k−1)
< (1 + o(1))n2a (2k!k)b p−b .
E(X) n
280 Hypergraphs

Finally, since a ≤ b we estimate eb+a(k−1) ≤ ekb , and consequently,

b
N(b, a)pn−b 2k!kek

1 + o(1)
< . (14.15)
E(X) np na(k−3)
Proof of (ii):
Here k = 3 and np ≥ ω. Hence, we obtain in (14.15)
b
N(b, a)pn−b 2k!kek

≤ (1 + o(1)) .
E(X) ω
Thus,
n b n b
N(b, a)pn−b 2k!kek

∑ ∑ E(X) < (1 + o(1)) ∑ b ω = o(1). (14.16)
b=1 a=1 b=1

This completes the proof of part (ii).

We prove Part (iii) by estimating N(b, a) more carefully, see Exercise 14.4.2 at
the end of the chapter.
Before leaving Hamilton cycles, we note that Allen, Böttcher,
Kohayakawa, and Person [19] describe a polynomial time algorithm for finding
a tight Hamilton cycle in Hn,p;k w.h.p. when p = n−1+ε for a constant ε > 0.
There is a weaker notion of Hamilton cycle due to Berge [80] viz. an alternating
sequence v1 , e1 , v2 , . . . , vn , en of vertices and edges such that (i) v1 , v2 , . . . , vn
are distinct and (ii) vi ∈ ei−1 ∩ ei for i = 1, 2, . . . , n. The cycle is weak if we do
not insist that the edges are distinct. Poole [648] proves that the threshold for
the existence of a weak Hamilton cycle in Hn,m;k is equal to the threshold for
minimum degree one.

14.3 Thresholds
In this section we describe a breakthrough result of Frankston, Kahn, Narayanan
and Park [319]. It will give us good estimates of the thresholds for various
structures.
A hypergraph H (thought of as a set of edges) is r-bounded if e ∈ H implies
that |e| ≤ r. (As part of the proof, we have to deal with non-uniform hyper-
graphs.). The most important notion comes next. For a set S ⊆ X = V (H ) we
let hSi = {T : S ⊆ T ⊆ X} denote the subsets of X that contain S. We say that
H is κ-spread if
|H |
|H ∩ hSi| ≤ |S| , ∀S ⊆ X. (14.17)
κ
Let Xm denote a random m-subset of X and X p denote a subset of X where
 14.3 Thresholds 281

each x ∈ X is included independently in X p with probability p. The following

theorem is from [319]:

Theorem 14.11 Let H be an r-bounded, κ-spread hypergraph and let X =

V (H ). There is an absolute constant K > 0 such that if
(K log r)|X|
m≥ (14.18)
κ
then w.h.p. Xm contains an edge of H . Here w.h.p. assumes that r → ∞.

Before giving a proof of Theorem 14.11, we will
give some examples of its
applicability. In the first two examples, X = [n] r and H = Hn,m;k (not to be
confused with H ).
Shamir’s Problem: This is the name given to that of finding the threshold for
the existence of a perfect matching in the random hypergraph Hn,m;k . A perfect
matching of an n-vertex, k-uniform hypergraph H, where k|n, is a set of n/k
disjoint edges that cover all vertices of H. If m = cn log n then the expected
number of isolated vertices in Hn,m;k is ≈ n1−ck . If ck < 1 then there will be
isolated vertices, w.h.p., see Exercise 14.4.1. Suppose now that H is the hyper-
graph with vertex set X = [n]


r and an edge of size n/k corresponding to each
perfect matching of the complete k-uniform hypergraph Hn,r on vertex set [n].,
Viz. each partition of [n] into n/k sets of size k. Thus H is n/k-bounded. Now
H has exactly (n/k)!k!
n!
n/k edges. We see more generally, that if we choose a set

S = {e1 , e2 , . . . , es } ⊆ X, i.e. s subsets of [n] of size k then E(H ) ∩ hSi = 0/

unless e1 , e2 , . . . , es are disjoint. If they are disjoint then

|E(H ) ∩ hSi| (n − ks)! (n/k)!k!s

= · =
|E(H )| n! (n/k − s)!
s−1  s
s n/k − i (k − 1)!
(k!) ∏ k−1 ≤ .
i=0 ∏ j=0 (n − ki − j) (n − 2k)k−1
k−1
Thus, H is κ = (n−2k)
(k−1)! spread. Applying Theorem 14.11 with this value of
κ and r = n/k, we see that there exists K > 0 such if m = Kn log n then Hn,m;k
has a perfect matching w.h.p. This was first proved in a breakthrough paper by
Johansson, Kahn and Vu [460].
Loose Hamilton
Cycles: We consider the case r = 1 of Section 14.2. In this
case X = [n]r and the edges of H correspond to the loose Hamilton cycles
of the complete k-uniform hypergraph Hn,r on vertex set [n]. We see from
Exercise 14.4.1 that we need at least Ω(n log n) random edges to have a loose
Hamilton cycle w.h.p. Now there are n/(k − 1) edges in a loose cycle and so
we take r = n/(k − 1). The number of loose Hamilton cycles in Hn,r is given
282 Hypergraphs

by k−1 n!
2n · (k−2)!n/(k−1) , Exercise 14.4.11, and for a set S = {e1 , e2 , . . . , es } ⊆ X we
have
|E(H ) ∩ hSi| (n − (k − 1)s − 1)! 2n(k − 2)!n/(k−1)
≤ ·
|E(H )| (k − 2)!n/(k−1)−s (k − 1)n!
s (k−1)s 
i −1

2(k − 2)!
≤ (k−1)s · ∏ 1 −
n i=1 n
( ) 
(k−1)s (k−1)s 2
2(k − 2)!s O(1) s
  
i 1 i
= (k−1)s · exp ∑ + ∑ n + · · · ≤ . (14.19)
n i=1 n 2 i=1 nk−1
Arguing as for the Shamir problem, we see that there exists K > 0 such if
m = Kn log n then Hn,m;k has a loose hamilton cycle w.h.p. This being the result
of [333], [256] and [258].
Powers of Hamilton cycles: The kth power of a Hamilton cyclein a graph
G = (V, E) is a permutation x1 , x2 , . . . , xn of the vertices V such that xi , xi+ j is
an edge of G for all i ∈ [n], j ∈ [k]. Kühn and Osthus [529] studied the existence
of kth powers in Gn,p . They showed that for k ≥ 3 one could use Riordan’s
Theorem [662] to show that if npk → ∞ then Gn,p contains the kth power of
a Hamilton cycle w.h.p. This is tight as the first moment method shows that
if npk → 0 then w.h.p. there are no kth powers. The problem is more difficult
for k = 2 and then after a series of papers, Nenadov and Škorić [621], Fischer,
Škorić, Steger and Trujić [305], Montgomery [602] we have an upper bound
2n
of p  logn1/2
. Theorem 14.11 reduces the bound to O(n3/2 log n) in Gn,m , as we
will now show.

We take X = [n] 2 and the edges of H correspond to the squares of Hamilton
cycles of Kn . In which case we have for |S| = s, r = n and
 bs/2c+1
|E(H ) ∩ hSi| (n − 2 − bs/2c)! e
≤ ≤ . (14.20)
|E(H )| (n − 1)!/2 n−1
We can therefore take κ = e−1 n1/2 , and then (14.18) yields the claimed upper
bound of O(n3/2 log n) on the threshold for the existence of the square of a
Hamilton cycle in Gn,m .
Bounded degree spanning trees: Let Tn be a sequence
of spanning trees of
Kn all of maximum degree ∆ = O(1). We take X = [n] 2 and the edges of H
correspond to the copies of T in Kn . We prove that (14.17) holds with κ = n/∆.
If S ⊆ X is not isomorphic to a subset of E(T ) then E(H )∩hSi = 0/ and (14.17)
holds. Suppose then that S ⊆ X is isomorphic to a subset of E(T ). Then, where
π is a random premutation of [n] and
π(T ) = ([n], {{π(v), π(w)} : {v, w} ∈ E(T )} ,
 14.3 Thresholds 283

we have
|E(H ) ∩ hSi|
= P(S ⊆ π(T )) ≤ κ |S| . (14.21)
|E(H )|
We leave the verifcation of (14.21) as an exercise – Exercise 14.4.15. Conse-
quently, we can apply Theorem 14.11 with r = n − 1 that O(n log n) random
edges are sufficient to contain a copy of Tn , w.h.p. Montgomery [600] gave the
first proof of this result.

Proof of Theorem 14.11

In the following, some passages have been more or less taken from [319]. Let γ
be a moderately small constant (e.g. γ = 0.1 suffices), and let C0 be a constant
large enough to support the estimates that follow. Let H be an r-bounded,
κ-spread hypergraph on a set X of size n, with r, κ ≥ C02
. Set p = C/κ with
n
C0 ≤ C ≤ κ/C0 (so p ≤ 1/C0 ), r0 = (1 − γ)r and N = np .
Let
[
hH i = hSi.
S∈H

Fix ψ : hH i → H satisfying ψ(Z) ⊆ Z for all Z ∈ hH i. I.e. given Z ∈ hH i,

ψ chooses a member of H contained in Z.
For W ⊆ X and S ∈ H , set

χ(S,W ) = ψ(S ∪W ) \W ;

and say that the pair (S,W ) is bad if |χ(S,W )| > r0 and good otherwise.

ψ(S ∪W ) χ(S,W ) is the

shaded rectangle

The idea now is to choose a small random set W and argue that w.h.p. there
exists H ∈ H such that |H \ W | is significantly smaller than |H|. We then
repeat the argument with repect to the hypergraph H \W .
284 Hypergraphs

Lemma 14.12 For H as above, and W chosen uniformly from X

np ,

E(|{S ∈ H : (S,W ) is bad}|) ≤ |H |C−r/3 .

Proof It is enough to show, for s ∈ (r0 , r],

E (|{S ∈ H : (S,W ) is bad and |S| = s}|) ≤ (γr)−1 |H |C−r/3 . (14.22)

We let Hs = {S ∈ H : |S| = s}. Now

E (|{S ∈ H : (S,W ) is bad and |S| = s}|) = ∑ P((S,W ) is bad)

S∈Hs
   
1  X
H

= S ∈ s ,W ∈ : (S,W )is bad .
N  np 

So, we instead concentrate on showing

|{(S,W ) : (S,W ) is bad and |S| = s}| ≤ (γr)−1 N|H |C−r/3 . (14.23)

(Note that γr = r − r0 bounds the number of s for which the set in question can
be nonempty, whence the negligible factors (γr)−1 .)
For Z ⊇ S ∈ Hs , we say that (S, Z) is pathological if there is T ⊆ S with t :=
|T | > r0 and

|{S0 ∈ Hs : S0 ∈ [T, Z]}| > Cr/2 |H |κ −t ps−t . (14.24)

From now on we will always take Z = W ∪ S (with W as in Lemma 14.12).

Note that in proving (14.23) we may assume s ≤ n/2. Indeed, let µ denote the
maximum multiplicity of an edge of H and let S have this multiplicity. Then,
as we may assume that |Hs | is at least the R.H.S. of (14.22), we have

µ ≤ κ −s |H | ≤ κ −s γrCr/3 |Hs | ≤ κ −s γrCr/3 µ2n < µ,

if s > n/2 (since κ > C).

We bound the nonpathological and pathological parts of (14.23) separately.
Nonpathological contributions. We first bound the number of pairs (S,W )
in (14.23) with (S, Z) nonpathological.
Step 1. There are at most
s    
n n+s
∑ ≤ ≤ N p−s (14.25)
i=0 np + i np + s
choices for Z = W ∪ S.
Step 2. Given Z, let S0 = ψ(Z). Choose T := S ∩ S0 , for which there are at most
0
2|S | ≤ 2r possibilities, and set t = |T | > r0 . (If t ≤ r0 then (S,W ) cannot be bad,
as χ(S,W ) = S0 \W ⊆ T .)
 14.3 Thresholds 285

Step 3. Since we are only interested in nonpathological choices, and we choose

S \ S0 by choosing S ∈ [T, Z], the number of possibilities for S is now at most
Cr/2 |H |κ −t ps−t .
Step 4. Complete the specification of (S,W ) by choosing W ∩ S, the number of
possibilities for which is at most 2s .
In sum, since s ≤ r and t > r0 = (1 − γ)r, the number of nonpathological pos-
sibilities is at most
2r+s N|H |Cr/2 (pκ)−t ≤ N|H |(4C1/2 )rC−t < N|H |[4C−(1/2−γ) ]r . (14.26)
Pathological contributions. We next bound the number of (S,W ) as in (14.23)
with (S, Z) pathological. The main point here is Step 4.
Step 1. There are at most |H | possibilities for S.
Step 2. Choose T ⊆ S witnessing the pathology of (S, Z) (i.e. for which (14.24)
holds); there are at most 2s possibilities for T .
Step 3. Choose U ∈ [T, S] for which
|Hs ∩ [U, (Z \ S) ∪U]| > 2−(s−t)Cr/2 |H |κ −t ps−t . (14.27)
(Here the left hand side counts members of Hs in Z whose intersection with S
is precisely U. Of course, existence of U as in (14.27) follows from (14.24).)
The number of possibilities for this choice is at most 2s−t .
Step 4. Choose Z \ S, the number of choices for which is less than N(2/C1/2 )r .
To see this, write Φ for the
R.H.S. of (14.27). Noting that Z \ S must belong to
X\S
X\S
X\S
np ∪ np−1 ∪ · · · ∪ np−s , we consider, for Y drawn uniformly from this set,

P(|Hs ∩ [U,Y ∪U]| > Φ). (14.28)

Set |U| = u. We have
|Hs ∩ hUi| ≤ |H ∩ hUi| ≤ |H |κ −u ,
while, for any S0 ∈ Hs ∩ hUi,
 s−u
0 np
P(Y ⊇ S \U) ≤
n−s
(of course if S0 ∩ S 6= U the probability is zero); so
np s−u
 
θ := E (|Hs ∩ [U,Y ∪U]|) ≤ |H |κ −u ≤ |H |κ −u (2p)s−u
n−s
(since n − s ≥ n/2). Markov’s Inequality then bounds the probability in (14.28)
by θ /Φ, and this bounds the number of possibilities for Z \ S by N(θ /Φ)
(cf. (14.25)), which is easily seen to be less than N(2/C1/2 )r .
286 Hypergraphs

Step 5. Complete the specification of (S,W ) by choosing S ∩W , which can be

done in at most 2s ways.
Combining (and slightly simplifying), we find that the number of pathological
possibilities is at most
|H |N(16/C1/2 )r . (14.29)

Finally, the sum of the bounds in (14.26) and (14.29) is less than the
(γr)−1 N|H |C−r/3 of (14.23).

Small uniformities
Very small set sizes are handled by a simple use of the Janson inequality, The-
orem 22.13.

Lemma 14.13 For an r-bounded, κ-spread G on Y , and α ∈ (0, 1),

 !−1 
r  
 r 
P(Yα 6∈ hG i) ≤ exp − 2 ∑ (ακ)−t . (14.30)
t=1 t
 

Proof Denote the members of G by Si and set ζi = 1{Yα ⊇Si } . Let Gg = {S ∈ G : |S| = g}.
Then
r
µ := ∑ E(ζi ) = ∑ α g |Gg |
g=1

and
r g
  r
g g
∆= ∑ E(ζi ζ j ) = ∑ ∑ α ∑ ∑ ∑ α h−t
i, j:Si ∩S j 6 0/
= g=1 |G|=g t=1 t h=t |H|=h
H∩G=T
r
r −t r r
r −t r
   
≤µ∑ α ∑ ∑ αh ≤ µ ∑ α ∑ ∑ αh
t=1 t h=t |H|=h t=1 t h=t |H|=h
H∩G=T H⊇T
r
r −t r h
 
≤µ∑ α ∑ α |Gh |. (14.31)
t=1 t h=t

We have the following inequality: suppose that θ1 ≥ θ2 ≥ · · · θN and that m1 , m2 , . . . , mN ≥

0. Then, for 1 ≤ t ≤ N,
θt mt + · · · + θN mN mt + · · · + mN
≤ (14.32)
θ1 m1 + · · · + θN mN m1 + · · · + mN
(Easily proved by induction on N. Observe that increasing θN to θN−1 reduces
the ratio.)
 14.3 Thresholds 287

We apply (14.32) with θh = α h and mh = |Gh | and N = r. The RHS of (14.32)

is bounded by κ −t because G is κ-spread. Therefore,
r r
1
∑ α h |Gh | ≤ κt ∑ α h |Gh |.
h=t h=1

Plugging this into (14.31) we get

r   r  r
r 1 h 2 r 1
∆≤µ∑ t ∑ α |Gh | = µ ∑ t (ακ)t .
t=1 t (ακ) h=1 t=1

Janson’s Inequality bounds the probability in (14.30) by exp −µ 2 /2∆ .



Corollary 14.14 Let G be as in Lemma 14.13, let t = α|Y | be an integer with

ακ ≥ 2r, and let W = Yt Then
n ακ o
P(W 6∈ hG i) ≤ 2 exp − .
4r
Proof Note that
r  
1 r
 
r −t 2r
∑ t (ακ) = 1 + −1 ≤ .
t=1 ακ ακ
Thus Lemma 14.13 gives
exp[−ακ/(4r)] ≥ P(Yα 6∈ hG i) ≥ P(|Yα | ≤ t) P(W 6∈ hG i) ≥ P(W 6∈ hG i)/2,
where we use the fact that any binomial ξ with E[ξ ] ∈ Z satisfies P(ξ ≤
E[ξ ]) ≥ 1/2.

Completing the proof

It will be convenient to prove the theorem assuming H is (2κ)-spread. Let γ
and C0 be as in Section 14.3 and H as in the statement of Theorem 14.11,
and recall that asymptotics refer to r. We may of course assume that κ ≥
2γ −1C0 log r (or the result is trivial with a suitably adjusted K). The statement
is also trivial if Kκ −1 log r ≥ 1.
In what follows we will have a sequence Hi , with H0 = H and
Hi ⊆ {χi (S,Wi ) : S ∈ Hi−1 },
where Wi and χi will be defined below (with χi a version of the χ of Sec-
tion 14.3).
√
Set C = C0 and p = C/κ, define ` by (1 − γ)` = log r/r, and set q = log r/κ.
−1
γ log r
Then ` = 1+ γ
+···
≤ γ −1 log r and Theorem 14.11 will follow from the next
2
assertion.
288 Hypergraphs

Claim 9 If W is a uniform ((`p + q)n)-subset of X, then W ∈ hH i w.h.p.

Proof Set δ = 1/(2`). Let r0 = r and ri = (1 − γ)ri−1 = (1 − γ) i r for i ∈ [`].
0
Xi−1

Let X0 = X and, for i = 1 . . . `, let Wi be uniform from np and set Xi =
Xi−1 \Wi . Note the assumption κ ≥ 2γ −1C0 log r ensures that
C0 γ −1 log r C0 γ −1 log r
   
n
|X` | = n − `np ≥ n 1 − ≥ n 1 − −1 = .
κ 2γ C0 log r 2
For S ∈ Hi−1 let χi (S,Wi ) = S0 \Wi , where S0 is a member of Hi−1 contained
in Wi ∪ S. Say that S is good if |χi (S,Wi )| ≤ ri (and bad otherwise), and set
Hi = {χi (S,Wi ) : S ∈ Hi−1 is good}.
Thus Hi is an ri -bounded collection of subsets of Xi . Note that Hi contains
a copy of set T for each S such that T = χi (S,Wi ). Finally, choose W`+1 uni-
X`

formly from nq . Then W = W1 ∪ · · · ∪ W`+1 is uniformly distributed as re-
quired for Claim 9. Note also that W ∈ hH i whenever W`+1 ∈ hH` i. (More
generally, W1 ∪ · · · ∪Wi ∪Y ∈ hH i whenever Y ⊆ Xi lies in hHi i.)
So to prove the claim, we just need to show
P(W`+1 ∈ hH` i) = 1 − o(1) (14.33)
(where the P refers to the entire sequence W1 . . .W`+1 ).
For i ∈ [`] call Wi successful if |Hi | ≥ (1 − δ )|Hi−1 |, call W`+1 successful if it
lies in hH` i, and say a sequence of Wi ’s is successful if each of its entries is.
We show that
h p i
P(W1 . . .W`+1 is successful) = 1 − exp −Ω( log r) . (14.34)

Now W1 . . .Wi−1 successful implies that |Hi−1 | > (1 − δ )` |H | > |H |/2. So

for I ⊆ Xi−1 we have
|H | |Hi−1 |
|Hi−1 ∩ hIi| ≤ |H ∩ hIi| ≤ |I|
≤ .
(2κ) κ |I|
We therefore have the spread condition (14.17) for Hi−1 . For i ∈ [`], according
to Lemma 14.12 (and Markov’s Inequality),
P(Wi is not successful |W1 . . .Wi−1 is successful) < δ −1C−ri−1 /3 ,
Thus
`
P(W1 . . .W` is successful) > 1 − δ −1 ∑ C−ri−1 /3
i=1
n p o
> 1 − exp − log r/4 (14.35)
 14.3 Thresholds 289
√
(using r` = log r).
Finally, if W1 . . .W` is successful, then Corollary 14.14 applied with G = H` ,
Y = X` , α = nq/|Y | ≥ q, r = r` , and W = W`+1 gives
n p o
P(W`+1 6∈ hH` i) ≤ 2 exp − log r/4 , (14.36)

and we have (14.34) and the claim.

We used the following to obtain (14.36):
2ακ nq · 2κ (n log r/κ) · 2κ n log r n log r
= ≥ `
= √ ≥ √ .
2r 2|X` |r` 2|X` |(1 − γ) r0 |X` |( log r/r)r0 n( log r/r)r

Sparse Hypergraphs
The above proof can be adapted to prove the following result related to the
Shamir problem.
C
Theorem 14.15 Suppose that p = n where C is a constant and C > C0 .
(k−1 )
Then w.h.p. Hn,p;k contains a matching of size at least
n 
1 − e−Cγ/C0 .
k
Proof Let now ` = bC/C0 c and let δ = 1/(2`) as before. The first inequality
in (14.35) continues to hold with r1 , r2 , . . . , r`−1 ≥ r` = Ω(n). It follows that
w.h.p. there is a matching of size
n n 
− r` ≥ 1 − (1 − γ)C/C0 .
k k

Square of a Hamilton cycle

Kahn, Narayanan and Park [466] modified the proof of Theorem 14.11 and
proved
Theorem 14.16 If m ≥ Cn3/2 for sufficiently large C then w.h.p. Gn,m contains
a copy of H2 , the square of a Hamilton cycle.
Proof Strictly speaking, we prove

∀ε > 0, ∃Cε such that m ≥ Cε n3/2

⇒ Pr(Gn,m contains a copy of H2 ) ≥ 1 − ε. (14.37)
290 Hypergraphs

Using machinery developed by Friedgut [320] we can get the claimed theorem.
Let X = [n] H


2 and be the

2n-uniform hypergraph whose edges correspond to
copies of H2 . Let N = n2 and m = Cn3/2 and k = 4n1/2 . For W ⊆ X, |W | = m
and S ∈ H we say that (S,W ) is bad if |T \W | > k for all T ∈ H , T ⊆ S ∪W .
Otherwise (S,W ) is good. (The difference here from Theorem 14.11 is that in
one round we grab almost all of an edge of H .)
In the course of the proof, we make some claims that will be verified later, or
left as an exercise.
Let p = 5C/n1/2 and p0 = 2C/n1/2 and define p1 by (1− p) = (1− p0 )(1− p1 )
so that Gn,p = Gn,p0 ∪Gn,p1 . The Chernoff bounds imply that w.h.p. |E(Gn,p0 | ≥
Cn3/2 ≥ m.
Then let W0 be a randomly chosen m-subset of X. Let W0 be a success if
| {S : (S,W0 ) is bad} | ≤ |H |/2.
Claim 10 If t ≥ k then

| {(S,W ) : (S,W ) is bad and |S ∩W | = t} | ≤

  
2n N − 2n
2C−k/3 |H | . (14.38)
t m−t
Now
|H |
| {W0 : W0 is a failure} | × ≤ ∑ | {(S,W ) : (S,W ) is bad} |.
2 W ∈X

Claim 10 gives
| {W0 : W0 is a failure} |
Pr(W0 is a failure) = N

m
2 ∑W ∈X | {(S,W ) : (S,W ) is bad} |
≤ N

≤ 4C−k/3 .
m |H |

(We have used the Vandermonde identity m = ∑t≥0 2n

N

N−2n

t m−t .)
Suppose now that W0 is a success and then let R = {S ∈ H : |S \W0 | ≤ k} and
for each S ∈ R let χ(S) denote some k-subset of S that contains χ(S). Let Z
denote the number of sets S such that ξ (S) is contained in Gn,p1 . Then we have
E(Z) = |R|pk1 ≥ |H |pk1 /2. (14.39)
Let
ft = |H |−1 ∑ |H ∩ hIi|.
I⊆(X2 )
|I|≥t

The following claim is left as Exercise 14.4.16.

 14.3 Thresholds 291

Claim 11 There exists an absolute constant K0 > 0 such that if 1 ≤ t ≤ n/3

then ft ≤ (K0 n−1/2 )t .
Now
−|A∩B|
Var(Z) ≤ p2k
1 ∑ p1 . (14.40)
A,B∈R
A∩B6=0

For A ∈ R and 1 ≤ t ≤ k, Claim 11 implies that

| {B ∈ R : |A ∩ B| = t} | ≤ ∑ |R ∩ hIi| ≤ ∑ |H ∩ hIi| ≤ (K0 n−1/2 )t |H |.
I⊆A I⊆A
|I|=t |I|=t

So the sum in (14.40) is at most

k k
t −t −t/2 t −t −t/2
Var(Z) ≤ |R| |H |p2k
1 ∑ K0 p1 n ≤ 2|R|2 p2k
1 ∑ K0 p1 n =
t=1 t=1
k
t −t 10K0
2|R|2 p2k
1 ∑ K0C ≤ E(Z)2 .
t=1 C
The Chebyshev inequality implies that
Var(Z) 10K0
Pr(Z = 0) ≤ ≤ .
E(Z)2 C
Putting Cε = 10K0 /ε verifies (14.37).

Proof of Claims
Proof of Claim 10 It is enough to show that
  
2n − t N − 2n
| {(S,W ) : (S,W ) is bad, |W ∩ S| = t} | ≤ 2C−k/3 |H | ,
t m−t
for t = 0, 1, . . . , 2n, since summing over t gives (14.38).
Fix t. We bound the number of bad (S,W )’s with |W ∩ S| =
t. In which case,
X
|W \ S| = m − t and |W ∪ S| = m − t + 2n. Call Y ⊆ m−t+2n pathological if
  
N − 2n N
| {S ⊆ Y : (S,Y \ S) is bad} | > C−k/3| |H | .
m−t m − t + 2n
We say that (S,W ) is pathological, if Y = S ∪W is.
Non-pathological contributions: To specify such, we choose Y = S ∪ W , then
S,W . This gives at most
      
N 2n −k/3| N − 2n N
× ×C |H |
m − t + 2n t m−t m − t + 2n
292 Hypergraphs
  
−k/3| 2n N − 2n
=C |H | . (14.41)
t m−t
Pathological contributions:
Claim 12 For a fixed S ∈ H and random W \ S from X\S

m−t and t ≥ k and
large enough C,
N−2n

−2k/3 m−t
E (| {H 3 J ⊆ W ∪ S : |J ∩ S| = t|}) ≤ C |H | N

. (14.42)
m−t+2n

Assume Claim 12 for now.

 
2n
We choose (S,W ∩ S) in at most |H | ways. (14.43)
t
Now (S,W ) bad implies that every J ⊆ S ∪ W satisfies |J ∩ S| ≥ |J \ W | ≥ k.
Because (S,W ) is pathological,
  
N − 2n N
aW = | {H 3 J ⊆ S ∪W : |J ∩ S| = t} | ≥ C−k/3| |H | .
m−t m − t + 2n
Claim 12 implies that
N−2n

1 −2k/3 m−t
N−2n
∑ aW ≤C |H | N

.
m−t W \S∈( X\S ) m−t+2n
m−t

Thus the number of choices for W \ S is at most



2
C−2k/3 |H | N−2n
m−t
N

m−t+2n
 
−k/3 N − 2n
 =C . (14.44)
m−t
C−k/3| |H | N−2n N



m−t m−t+2n

Claim 10 now follows from (14.41), (14.43) and (14.44).

Proof of Claim 12 With ft equal to the fraction of J’s in H with |J ∩ S| = t,

the left hand side of (14.42) is equal to
m−t

2n−t
ft |H | Pr(J ⊆ W ∪ S) = ft |H | N−2n

,
2n−t

for arbitrary J ∈ H , |J ∩ S| = t.
It is therefore enough to show that
m−t
N
 t
2n−t m−t+2n K0
ft N−2n
N−2n
≤ . (14.45)
2n−t m−t
C
 14.4 Exercises 293

Observe that
m−t
 2n−t N
 2n
2n−t m−t m−t+2n N − 2n
N−2n

≤ and N−2n

≤ .
2n−t
N − 2n m−t
m−t

(For the second inequality use the lower bound in Part (g) of Lemma 22.1.)
This implies that
m−t
N

!t
N − 2n t n1/2
 
2n−t m−t+2n
N−2n
N−2n

≤ ≤ . (14.46)
2n−t m−t
m−t 2C

Now for n/3 ≤ t ≤ 2n, the fact that H is κ = e−1 n1/2 spread implies
   2 t
2n e t 6e
ft ≤ ≤ . (14.47)
t n1/2 n1/2
For k ≤ t ≤ n/3 we have from Claim 11 that

K0 t
 
ft ≤ . (14.48)
n1/2
Equation (14.45) (and Claim 12) now follow from (14.46), (14.47) and (14.48),
assuming K0 ≥ 6e2 .

14.4 Exercises
14.4.1 Show that if m = cn log n then (i) ck < 1 implies that Hn,m;k has isolated
vertices w.h.p. and (ii) if ck > 1 then Hn,m;k is connected w.h.p.
14.4.2 Generalise the notion of configuration model to k-uniform, k ≥ 3, hy-
pergraphs. Use it to show that if r = O(1) then the number of r-regular,
k-uniform hypergraphs with vertex set [n], k|n is asymptotically equal
to
(rn)!
e−(k−1)(r−1)/2 .
(k!)rn/k r!n (rn/k)!

14.4.3 Generalise the notion of switchings to k-uniform hypergraphs. Use them

to extend the result of Exercise 14.4.2 to r = nε , for sufficiently small
ε > 0.
14.4.4 Extend the result of Exercise 14.4.2 to k-uniform, k ≥ 3, hypergraphs
with a fixed degree sequence d = (d1 , d2 , . . . , dn ), and maximum de-
gree ∆ = O(1). Let M1 = ∑ni=1 di where k|M1 and M2 = ∑ni=1 di (di − 1).
294 Hypergraphs

Show that the number of k-uniform hypergraphs with degree sequence

d is asymptotically equal to
 
M1 ! (k − 1)M2
exp − .
k!M1 /k (∏ni=1 di !) (M1 /k)! 2M1

14.4.5 Prove Lemma 14.7.

14.4.6 Prove Lemma 14.9.
14.4.7 Let Hn,d be a random k-uniform hypergraph with a fixed degree se-
quence. Suppose that there are λi n vertices of of degree i = 1, 2, . . . , L =
O(1). Let Λ = ∑Li=1 λi i((i−1)(k −1)−1). Prove that for ε > 0, we have
(a) If Λ < −ε then w.h.p. the size of the largest component in Hn,d is
O(log n).
(b) If Λ > ε then w.h.p. there is a unique giant component of linear
size ≈ Θn where Θ is defined as follows: let K = ∑Li=1 iλi and
L
kα i/k
 
f (α) = K − kα − ∑ iλi 1 − . (14.49)
i=1 K
Let ψ be the smallest positive solution to f (α) = 0. Then
L
kψ i/k
 
Θ = 1 − ∑ λi 1 − .
i=1 K
If λ1 = 0 then Θ = 1, otherwise 0 < Θ < 1.
(c) In Case (b), the degree sequence of the graph obtained by deleting
the giant component satisfies the conditions of (a).
14.4.8 Prove Part (iii) of Theorem 14.10 by showing that
 
2a b − 1
N(b, a) ≤ n ∑ 2t+a (n − b − a(k − 1) − t)!(k!)a+t
a − 1 t≥0
≤ ck (2k!)a (n − b − a(k − 1))!,
where ck depends only on k.
Then use (14.7).
14.4.9 In a directed k-uniform hypergraph, the vertices of each edge are totally
order. Thus each k-set has k! possible orientations. Given a permutation
i1 , i2 , . . . , in of [n] we construct a directed `-overlapping Hamilton cycle
~E1 = (i1 , . . . , ik ), ~E2 = (ik−`+1 . . . , i2k−` ), . . . , ~Em = (in−(k−`)+1 , . . . , i` ).
`
Let ~Hn,p:k be the directed hypergraph in which each possible directed
edge is included with probability p. Use the idea of McDiarmid in Sec-
tion 13.2 to show (see Ferber [296]) that
 14.4 Exercises 295

P(~Hn,p:k contains a directed r-overlapping Hamilton cycle)

≥ P(Hn,p;k contains an r-overlapping Hamilton cycle).
14.4.10 A hypergraph H = (V, E) is 2-colorable if there exists a partition of
V into two non-empty sets A, B such that e ∩ A 6= 0,
/ e ∩ B 6= 0/ for all
e ∈ E. Let m = nk p. Show that if c is sufficiently large and m = c2k n

then w.h.p. Hn,p;k is not 2-colorable.

14.4.11 Verify (14.19) and (14.20)
14.4.12 Given a hypergraph H, let a vertex coloring be strongly proper if no
edge contains two vertices of the same color. The strong chromatic
number χ1 (H) is the minimum number of color’s in a strongly proper
coloring. Suppose that k ≥ 3 and 0 < p < 1 is constant. Show that w.h.p.
d nk−1 p
χ1 (Hn,p;k ) ≈ where d = .
2 log d (k − 2)!
14.4.13 Let U1 ,U2 , . . . ,Uk denote k disjoint sets of size n. Let H Pn,m,k denote
the set of k-partite, k-uniform hypergraphs with vertex set V = U1 ∪
U2 ∪ · · · ∪Uk and m edges. Here each edge contains exactly one vertex
from each Ui , 1 ≤ i ≤ k. The random hypergraph HPn,m,k is sampled
uniformly from H Pn,m,k . Prove the k-partite analogue of Shamir’s
problem viz. there exists a constant K > 0 such that if m ≥ Kn log n
then
lim P(HPn,m,k has a 1-factor) = 1.
n→∞

14.4.14 Find the threshold, up to a log n factor for the existence of the following
structures in Gn,p : replace 4 by n in an appropriate way.

C4 -cycle, overlap 2 K4 -cycle, overlap 2

14.4.15 Verify (14.21).

14.4.16 Verify Claim 10.

(Hints: For I ⊆ X2 let κ(I) denote the number of components in the
graph induced by I and the vertices incident with I.
296 Hypergraphs

(a) |I| = t ≤ n/3 implies that |H ∩ hIi| ≤ (16)t n − t+c

 

2 − 1 !.
(b) If F ⊆ H2 and |F| = h then the number of
subgraphs of F with t
edges and c components is at most (8e)t 2hc .)

14.5 Notes
Components and cores
If H = (V, E) is a k-uniform hypergraph and 1 ≤ j ≤ k −1 then two sets J1 , J2 ∈
V

j are said to be j-connected if there is a sequence of serts E1 , E2 , . . . , E` such
that J1 ⊆ E1 , J2 ⊆ E`
and |Ei ∩ Ei+1 | ≥ j for 1 ≤ i < `. This defines an equiv-
alence relation on Vj and the equivalance classes are called j-components.
Karoński and Łuczak [477] studied the sizes of the 1-components of the ran-
dom hypergraph Hn,m;k and proved the existence of a phase transition at m ≈
n
k(k−1) . Cooley, Kang and Koch [199] generalised this to j-components and
(n)
proved the existence of a phase transition at m ≈  k k n . As usual, a
( j)−1 (k− j)
phase transition corresponds to the emergence of a unique giant, i.e. one of
order nj .

The notion of a core extends simply to hypergraphs and the sizes of cores in
random hypergraphs has been considered by Molloy [594]. The r-core is the
largest sub-hypergraph with minimum degree r. Molloy proved the existence
of a constant ck,r such that if c < cr,k then w.h.p. Hn,cn;k has no r-core and that if
c > cr,k then w.h.p. Hn,cn;k has a r-core. The efficiency of the peeling algorithm
for finding a core has been considered by Jiang, Mitzenmacher and Thaler
[458]. They show that w.h.p. the number of rounds in the peeling algorithm is
log log n
asymptotically log(k−1)(r−1) if c < cr,k and Ω(log n) if c > cr,k . Gao and Molloy
[364] show that for |c − cr,k | ≤ n−δ , 0 < δ < 1/2, the number of rounds grows
like Θ̃(nδ /2 ). In this discussion, (r, k) 6= (2, 2).

Chromatic number
Krivelevich and Sudakov [521] studied the chromatic number of the random
k-uniform hypergraph Hn,p;k . For 1 ≤ γ ≤ k − 1 we say that a set of vertices S
is γ-independent in a hypergraph H if |S ∩ e| ≤ γ. The γ-chromatic number of a
hypergraph H = (V, E) is the minimum number of
sets in a partition of V into
γ-independent sets. They show that if d (γ) = γ k−1
γ
n−1

k−1 p is sufficiently large
d (γ)
then w.h.p. (γ+1) log d (γ)
is a good estimate of the γ-chromatic number of Hn,p;k .
 14.5 Notes 297

Dyer, Frieze and Greenhill [271] extended the results of [5] to hypergraphs.
Let uk,` = `k−1 log `. They show that if uk,`−1 < c < uk,` then w.h.p. the (weak
(γ = k − 1)) chromatic number of Hn,cn;k is either k or k + 1.
Achlioptas, Kim,
Krivelevich and Tetali [2] studied the 2-colorability of H =
Hn,p;k . Let m = nk p be the expected number of edges in H. They show that if
m = c2k n and c > log2 2 then w.h.p. H is not 2-colorable. They also show that if
c is a small enough constant then w.h.p. H is 2-colorable.

Orientability
Gao and Wormald [366], Fountoulakis, Khosla and Panagiotou [316] and
Lelarge [534] discuss the orientability of random hypergraphs. Suppose that
0 < ` < k. To `-orient an edge e of a k-uniform hypergraph H = (V, E), we
assign positive signs to ` of its vertices and k − ` negative signs to the rest. An
(`, r)-orientation of H consists of an `-orientation of each of its edges so that
each vertex receives at most r positive signs due to incident edges. This notion
has uses in load balancing. The papers establish a threshold for the existence
of an (`, r)-orientation. Describing it it is somewhat complex and we refer the
reader to the papers themselves.

VC-dimension
Ycart and Ratsaby [754] discuss the VC-dimension of H = Hn,p;k . Let p =
cn−α for constants c, α. They give the likely VC-dimension of H for various
values of α. For example if h ∈ [k] and α = k − h(h−1)
h+1 then the VC-dimension
is h or h − 1 w.h.p.

Erdős-Ko-Rado
The famous Erdős-Ko-Rado theorem states that if n > 2k then the
maximum
size of a family of mutually intersecting k-subsets of [n] is n−1
k−1 and this is
achieved by all the subsets that contain the element 1. Such collections will
be called stars. Balogh, Bohman and Mubayi [48] considered this problem in
relation to the random hypergraph Hn,p;k . They consider for what values of k, p
is it true that maximum size intersecting family of edges is w.h.p. a star. More
recently Hamm and Kahn [400], [401] have answered some of these questions.
For many ranges of k, p the answer is as yet unknown.
Bohman, Cooper, Frieze, Martin and Ruszinko [112] and Bohman,
Frieze, Martin, Ruszinko and Smyth [113] studied the k-uniform hypergraph H
298 Hypergraphs

obtained by adding random k-sets one by one, only adding a set if it intersects
all previous sets. They prove that w.h.p. H is a star for k = o(n1/3 ) and were
able to analyse the structure of H for k = o(n5/12 ).

Perfect matchings and Hamilton cycles in regular hypergraphs

The perfect matching probelm turns out to be a much easier problem than that
discussed in Section 14.3. Cooper, Frieze, Molloy and Reed [227] used small
subgraph conditioning to prove that Hn,r;k has a perfect matching w.h.p. iff
log r
k > kr where kr = (r−1) log r + 1.
((r−1 ))
Dudek, Frieze, Ruciński and Šileikis [259] made some progress on loose Hamil-
ton cycles in random regular hypergraphs. Their approach was to find an em-
beddding of Hn,m;k in a random regular k-uniform hypergraph.
P ART THREE

OTHER MODELS
 15
Trees

The properties of various kinds of trees are one of the main objects of study in
graph theory mainly due to their wide range of application in various areas of
science. Here we concentrate our attention on the “average” properties of two
important classes of trees: labeled and recursive. The first class plays an im-
portant role in both the sub-critical and super-critical phase of the evolution of
random graphs. On the other hand random recursive trees serve as an example
of the very popular random preferential attachment models. In particular we
will point out, an often overlooked fact, that the first demonstration of a power
law for the degree distribution in the preferential attachment model was shown
in a special class of inhomogeneous random recursive trees.
The families of random trees, whose properties are analyzed in this chapter, fall
into two major categories according to the order of their heights: they are ei-
ther of square root (labeled trees) or logarithmic (recursive trees) height. While
most of square-root-trees appear in probability context, most log-trees are en-
countered in algorithmic applications.

15.1 Labeled Trees

Consider the family Tn of all nn−2 labeled trees on vertex set [n] = {1, 2, . . . , n}.
Let us choose a tree Tn uniformly at random from the family Tn . The tree Tn is
called a random tree (random Cayley tree).
The Prüfer code [656] establishes a bijection between labeled trees on vertex
set [n] and the set of sequences [n]n−2 of length n − 2 with items in [n]. Such
a coding also implies that there is a one-to-one correspondence between the
number of labeled trees on n vertices with a given degree sequence d1 , d2 , . . . , dn
and the number of ways in which one can distribute n − 2 particles into n cells,
such that ith cell contains exactly di − 1 particles.
If the positive integers di , i = 1, 2, . . . , n satisfy
d1 + d2 + · · · + dn = 2(n − 1),
then there exist  
n−2
(15.1)
d1 − 1, d2 − 1, . . . , dn − 1

301
302 Trees

trees with n labeled vertices, the ith vertex having degree di .

The following observation is a simple consequence of the Prüfer bijection.
Namely, there are
 
n−2
(n − 1)n−i−1 (15.2)
i−1
trees with n labeled vertices in which the degree of a fixed vertex v is equal to
i.
Let Xv be the degree of the vertex v in a random tree Tn , and let Xv∗ = Xv − 1.
Dividing the above formula by nn−2 , it follows that, for every i, Xi∗ has the
Bin(n − 2, 1/n) distribution, which means that the asymptotic distribution of
Xi∗ tends to the Poisson distribution with mean one.
This observation allows us to obtain an immediate answer to the question of
the limiting behavior of the maximum degree of a random tree. Indeed, the
proof of Theorem 3.4 yields:

Theorem 15.1 Denote by ∆ = ∆(Tn ) the maximum degree of a random tree.

Then w.h.p.
log n
∆(Tn ) ≈ .
log log n

The classical approach to the study of the properties of labeled trees chosen
at random from the family of all labeled trees was purely combinatorial, i.e.,
via counting trees with certain properties. In this way, Rényi and Szekeres
[658], using complex analysis, found the height of a random labeled tree on n
vertices (see also Stepanov [716], while for a general probabilistic context of
their result, see a survey paper by Biane, Pitman and Yor [90]).
Assume that a tree with vertex set V = [n] is rooted at vertex 1. Then there is
a unique path connecting the root with any other vertex of the tree. The height
of a tree is the length of the longest path from the root to any pendant vertex of
the tree. Pendant vertices are the vertices of degree one.

Theorem 15.2 Let h(Tn ) be the height of a random tree Tn . Then

 
h(Tn )
lim P √ < x = η(x),
n→∞ 2n
where
4π 5/2 ∞ 2 −(kπ/x)2
η(x) = ∑k e .
x3 k=1
 15.1 Labeled Trees 303

Moreover,
√ π(π − 3)
E h(Tn ) ≈ 2πn and Var h(Tn ) ≈ n.
3

We will now introduce a useful relationship between certain characteristics of

random trees and branching processes. Consider a Galton-Watson branching
process µ(t), t = 0, 1, . . . , starting with M particles, i.e., with µ(0) = M, in
which the number of offspring of a single particle is equal to r with probability
pr , ∑∞
r=0 pr = 1. Denote by ZM the total number of offspring in the process
µ(t). Dwass [268] (see also Viskov [738]) proved the following relationship.

Lemma 15.3 Let Y1 ,Y2 , . . . ,YN be a sequence of independent identically dis-

tributed random variables, such that
P(Y1 = r) = pr for r = 1, 2, . . . , N.
Then
M
P(ZM = N) = P(Y1 +Y2 + . . . +YN = N − M).
N

Now, instead of a random tree Tn chosen from the family of all labeled trees Tn
on n vertices, consider a tree chosen at random from the family of all (n+1)n−1
trees on n + 1 vertices, with the root labeled 0 and all other vertices labeled
from 1 to n. In such a random tree, with a natural orientation of the edges from
the root to pendant vertices, denote by Vt the set of vertices at distance t from
the root 0. Let the number of outgoing edges from a given vertex be called
its out-degree and Xr,t+ be the number of vertices of out-degree r in Vt . For our
branching process, choose the probabilities pr , for r = 0, 1, . . ., as equal to
λ r −λ
pr = e ,
r!
i.e., assume that the number of offspring has the Poisson distribution with mean
λ > 0. Note that λ is arbitrary here.
Let Zr,t be the number of particles in the tth generation of the process, having
exactly r offspring. Next let X = [mr,t ], r,t = 0, 1, . . . , n be a matrix of non-
negative integers. Let st = ∑nr=0 mr,t and suppose that the matrix X satisfies the
following conditions:
(i) s0 = 1,
st = m1,t−1 + 2m2,t−1 + . . . nmn,t−1 for t = 1, 2, . . . n.
(ii) st = 0 implies that st+1 = . . . = sn = 0.
 304 Trees

(iii) s0 + s1 + . . . + sn = n + 1.

Then, as proved by Kolchin [506], the following relationship holds between

the out-degrees of vertices in a random rooted tree and the number of offspring
in the Poisson process starting with a single particle.

Theorem 15.4

P([Xr,t+ ] = X) = P([Zr,t ] = X|Z = n + 1).

Proof In Lemma 15.3 let M = 1 and N = n + 1. Then,

1
P(Z1 = n + 1) = P(Y1 +Y2 + . . . +Yn+1 = n)
n+1
n+1 ri
1 λ −λ
= ∑ e
n + 1 r1 +...+rn+1 =n ∏
i=1 ri !

(n + 1)n λ n e−λ (n+1)

= .
(n + 1)!
Therefore

P([Zr,t ] = X|Z = n + 1) =
n st
m0,t mn,t
∏t=0 m0,t ,...,mn,t p0 . . . pn
=
P(Z = n + 1)
 mr,t
n st ! n λ r −λ
(n + 1)! ∏t=0 m0,t ! m1,t ! ... mn,t ! ∏r=0 r! e
=
(n + 1)n λ n e−λ (n+1)
(n + 1)! s1 ! s2 ! . . . sn ! n n 1
= ∏ ∏ mr,t ! (r!)mr,t . (15.3)
(n + 1)n t=0 r=0

On the other hand, one can construct all rooted trees such that [Xr,t+ ] = X in the
following manner. We first layout an unlabelled tree in the plane. We choose
a single point (0, 0) for the root and then points St = {(i,t) : i = 1, 2, . . . , st }
for t = 1, 2, . . . , n. Then for each t, r we choose mr,t points of St that will be
joined to r points in St+1 . Then, for t = 0, 1, . . . , n − 1 we add edges. Note that
Sn , if non-empty, has a single point corresponding to a leaf. We go through St
in increasing order of the first component. Suppose that we have reached (i,t)
and this has been assigned out-degree r. Then we join (i,t) to the first r vertices
of St+1 that have not yet been joined by an edge to a point in St . Having put
Sn
in these edges, we assign labels 1, 2, . . . , n to t=1 St . The number of ways of
 15.1 Labeled Trees 305

doing this is
n
st !
∏ ∏n × n!.
t=1 r=1 mr,t !

The factor n! is an over count. As a set of edges, each tree with [Xr,t+ ] = X
n
appears exactly ∏t=0 ∏nr=0 (r!)mr,t times, due to permutations of the trees below
each vertex. Summarising, the total number of tree with out-degrees given by
the matrix X is
n n
1
n! s1 ! s2 ! . . . sn ! ∏ ∏ mr,t ,
m
t=0 r=0 r,t ! (r!)

which, after division by the total number of labeled trees on n + 1 vertices,

i.e., by (n + 1)n−1 , results in an identical formula to that given for the random
matrix [X+r,t ] in the case of [Zr,t ], see (15.3). To complete the proof one has to
notice that for those matrices X which do not satisfy conditions (i) to (iii) both
probabilities in question are equal to zero.

Hence, roughly speaking, a random rooted labeled tree on n vertices has asymp-
totically the same shape as a branching process with Poisson, parameter one in
terms of family sizes. Grimmett [391] uses this probabilistic representation to
deduce the asymptotic distribution of the distance from the root to the nearest
pendant vertex in a random labeled tree Tn , n ≥ 2. Denote this random variable
by d(Tn ).

Theorem 15.5 As n → ∞,
( )
k−1
P(d(Tn ) ≥ k) → exp ∑ αi ,
i=1

where the αi are given recursively by

α0 = 0, αi+1 = eαi − e−1 − 1.

Proof Let k be a positive integer and consider the sub-tree of Tn induced by

the vertices at distance at most k from the root. Within any level (strata) of Tn ,
order the vertices in increasing lexicographic order, and then delete all labels,
excluding that of the root. Denote the resulting tree by Tnk .
Now consider the following branching process constructed recursively accord-
ing to the following rules:

(i) Start with one particle (the unique member of generation zero).
 306 Trees

(ii) For k ≥ 0, the (k + 1)th generation Ak+1 is the union of the families of de-
scendants of the kth generation together with one additional member which
is allocated at random to one of these families, each of the |Ak | families
having equal probability of being chosen for this allocation. As in Theorem
15.4, all family sizes are independent of each other and the past, and are
Poisson distributed with mean one.

Lemma 15.6 As n → ∞ the numerical characteristics of Tnk have the same

distribution as the corresponding characteristics of the tree defined by the first
k generations of the branching process described above.

Proof For a proof of Lemma 15.6, see the proof Theorem 3 of [391].

Let Yk be the size of the kth generation of our branching process and let Nk
be the number of members of the kth generation with no offspring. Let i =
(i1 , i2 , . . . , ik ) be a sequence of positive integers, and let

A j = {N j = 0} and B j = {Y j = i j } for j = 1, 2, . . . , k.

Then, by Lemma 15.6, as n → ∞,

P(d(Tn ) ≥ k) → P(A1 ∩ A2 ∩ . . . ∩ Ak ) .

Now,
k
P(A1 ∩ A2 ∩ . . . ∩ Ak ) = ∑ ∏ P(A j |A1 ∩ . . . ∩ A j−1 ∩ B1 ∩ . . . B j )
i j=1

× P(B j |A1 ∩ . . . ∩ A j−1 ∩ B1 ∩ . . . B j−1 ),

Using the Markov property,

k
P(A1 ∩ A2 ∩ . . . ∩ Ak ) = ∑ ∏ P(A j |B j ) P(B j |A j−1 ∩ B j−1 )
i j=1
k
i j −1
= ∑ ∏ 1 − e−1 C j (i j ), (15.4)
i j=1

where C j (i j ) = P(B j |A j−1 ∩ B j−1 ) is the coefficient of xi j in the probability

generating function D j (x) of Y j conditional upon Y j−1 = i j−1 and N j = 0. Thus

Y j = 1 + Z + R1 + . . . + Ri j−1 −1 ,
 15.1 Labeled Trees 307

where Z has the Poisson distribution and the Ri are independent random vari-
ables with Poisson distribution conditioned on being non-zero. Hence
e − 1 i j−1 −1
 x 
D j (x) = xex−1 .
e−1
Now,
∞
Dk (1 − e−1 )
∑ (1 − e−1 )ik −1Ck (ik ) = 1 − e−1
.
ik =1

We can use this to eliminate ik in (15.4) and give

P(A1 ∩ A2 ∩ . . . ∩ Ak ) =
!ik−1 −1
k−1
i −1 eβ1 − 1
∑ ∏ β1 j C j (i j )eβ1 −1 , (15.5)
(i1 ,...,ik−1 ) j=1
e−1

where β1 = 1 − e−1 . Eliminating ik−1 from (15.5) we get

P(A1 ∩ A2 ∩ . . . ∩ Ak ) =
!ik−2 −1
k−2
i −1 eβ2 − 1
∑ ∏ β1 j C j (i j )eβ1 +β2 −2 ,
(i1 ,...,ik−2 ) j=1
e−1

where β2 = (eβ1 − 1). Continuing we see that, for k ≥ 1,

( ) ( )
k k
P(A1 ∩ A2 ∩ . . . ∩ Ak ) = exp ∑ (βi − 1) = exp ∑ αi ,
i=1 i=1

where β0 , β1 , . . . are given by the recurrence

 
β0 = 1, βi+1 = eβi − 1 e−1 ,

and αi = βi − 1. One can easily check that βi remains positive and decreases
monotonically as i → ∞, and so αi → −1.
Another consequence of Lemma 15.3 is that, for a given N, one can associated
with the sequence Y1 ,Y2 , . . . ,YN , a generalized occupancy scheme of distribut-
ing n particles into N cells (see [506]). In such scheme, the joint distribution of
the number of particles in each cell (ν1 , ν2 , . . . , νN ) is given, for r = 1, 2, . . . , N
by
 N !

P(νr = kr ) = P Yr = kr  ∑ Yr = n .
 (15.6)
r=1
n
Now, denote by Xr+ = ∑t=0 Xr,t+ the number of vertices of out-degree r in a
308 Trees

random tree on n + 1 vertices, rooted at a vertex labeled 0. Denote by Z (r) =

n
∑t=0 Zr,t , the number of particles with exactly r offspring in the Poisson process
µ(t). Then by Theorem 15.4,

P(Xr+ = kr , r = 0, 1, . . . , n) = P(Z (r) = kr , r = 0, 1, . . . , n|Z1 = n + 1).

Hence by equation (15.1), the fact that we can choose λ = 1 in the process
µ(t) and (15.6), the joint distribution of out-degrees of a random tree coincides
with the joint distribution of the number of cells containing the given number
of particles in the classical model of distributing n particles into n + 1 cells,
where each choice of a cell by a particle is equally likely.
The above relationship, allows us to determine the asymptotic behavior of the
expectation of the number Xr of vertices of degree r in a random labeled tree
Tn .

Corollary 15.7
n
E Xr ≈ .
(r − 1)! e

15.2 Recursive Trees

We call a tree on n vertices labeled 1, 2, . . . , n a recursive tree (or increasing
tree) if the tree is rooted at vertex 1 and, for 2 ≤ i ≤ n, the labels on the unique
path from the root to vertex i form an increasing sequence. It is not difficult
to see that any such tree can be constructed “recursively”: Starting with the
vertex labeled 1 and assuming that vertices “arrive” in order of their labels, and
connect themselves by an edge to one of the vertices which “arrived” earlier.
So the number of recursive (increasing) trees on n vertices is equal to (n − 1)!.
A random recursive tree is a tree chosen uniformly at random from the family
of all (n − 1)! recursive trees. Or equivalently, it can be generated by a recur-
sive procedure in which each new vertex chooses a neighbor at random from
previously arrived vertices. We assume that our tree is rooted at vertex 1 and
all edges are directed from the root to the leaves.
Let Tn be a random recursive tree and let D+ n,i be the out-degree of the vertex
with label i, i.e the number of “children” of vertex i. We start with the exact
probability distribution of these random variables.
 15.2 Recursive Trees 309

Theorem 15.8 For i = 1, 2, . . . , n and r = 1, 2, . . . , n − 1,

(i − 1)! n−i k
 
P(D+
n,i = r) = ∑ r (i − 1)k−r |s(n − i, k)| (15.7)
(n − 1)! k=r
where s(n − i, k) is the Stirling number of the first kind.

Proof Conditioning on tree Tn−1 we see that, for r ≥ 1,

n−2 1
P(D+
n,i = r) = P(D+
n−1,i = r) + P(D+
n−1,i = r − 1). (15.8)
n−1 n−1
Fix i and let
n−i
Φn,i (z) = ∑ P(D+n,i = r)zr
r=0

be the probability generating function of D+

n,i .
r
Multiplying (15.8) by z and then summing over r ≥ 1 we see that

Φn,i (z) − P(D+

n,i = 0) =
n−2   z
Φn−1,i (z) − P(D+ n−1,i = 0) + Φn−1,i (z).
n−1 n−1
Notice, that the probability that vertex i is a leaf equals
n−1  
1 i−1
P(D+ n,i = 0) = ∏ 1 − = . (15.9)
j=i j n −1
Therefore
n−2 z
Φn,i (z) = Φn−1,i (z) + Φn−1,i (z).
n−1 n−1
With the boundary condition,
Φi,i (z) = P(D+
i,i = 0) = 1.

One can verify inductively that

n−i  
z+i+k−2
Φn,i (z) = ∏
k=1 i+k−1
(i − 1)!
= (z + i − 1)(z + i) . . . (z + n − 2). (15.10)
(n − 1)!
Recall the definition of Stirling numbers of the first kind s(n, k). For non-
negative integers n and k
n
z(z − 1) . . . (z − n + 1) = ∑ s(n, k)zk .
k=1
310 Trees

Hence
(i − 1)! n−i
Φn,i (z) = ∑ |s(n − i, k)|(z + i − 1)k
(n − 1)! k=1
(i − 1)! n−i k k r
 
= ∑ ∑ r z (i − 1)k−r |s(n − i, k)|
(n − 1)! k=1 r=0
!
n−i
(i − 1)! n−i k
 
=∑
(n − 1)! ∑ r (i − 1) |s(n − i, k)| zr .
k−r
r=0 k=r

It follows from (15.10), by putting z = 0, that the expected number of vertices

of out-degree zero is
n
i−1 n
∑ n−1 = 2.
i=1

Then (15.8) with i = r = 1 implies that P(D+

n,1 = 1) = 1/(n − 1). Hence, if Ln
is the number of leaves in Tn , then
n 1
E Ln = + . (15.11)
2 n−1

For a positive integer n, let ζn (s) = ∑nk=1 k−s be the incomplete Riemann zeta
function, and let Hn = ζ (1) = ∑nk=1 k−1 be the nth harmonic number, and let
δn,k denote the Kronecker function 1n=k .

Theorem 15.9 For 1 ≤ i ≤ n, let Dn,i be the degree of vertex i in a random

recursive tree Tn . Then

E Dn,i = Hn−1 − Hi−1 + 1 − δ1,i ,

while
Var Dn,i = Hn−1 − Hi−1 − ζn−1 (2) + ζi−1 (2).

Proof Let N j be the label of that vertex among vertices 1, 2, . . . j − 1 which is

the parent of vertex j. Then for j ≥ 1 and 1 ≤ i < j
n
Dn,i = ∑ δN j ,i . (15.12)
j=i+1

By definition N2 , N3 , . . . , Nn are independent random variables and for all i, j,

1
P(N j = i) = . (15.13)
j−1
 15.2 Recursive Trees 311

The expected value of Dn,i follows immediately from (15.12) and (15.13). To
compute the variance observe that
n  
1 1
Var Dn,i = ∑ 1− .
j=i+1 j − 1 j−1

From the above theorem it follows that Var Dn,i ≤ E Dn,i . Moreover, for fixed i
and n large, E Dn,i ≈ log n, while for i growing with n the expectation E Dn,i ≈
log n − log i. The following theorem, see Kuba and Panholzer [527], shows a
standard limit behavior of the distribution of Dn,i .

Theorem 15.10 Let i ≥ 1 be fixed and n → ∞. Then

p d
(Dn,i − log n)/ log n → N(0, 1).

Now, let i = i(n) → ∞ as n → ∞. If

(i) i=o(n), then

p d
(Dn,i − (log n − log i))/ log n − log i → N(0, 1),

(ii) i = cn, 0 < c < 1, then

d
Dn,i → Po(− log c),

(iii) n − i = o(n), then

P(D+
n,i = 0) → 1.

Now, consider another parameter of a random recursive tree.

Theorem 15.11 Let r ≥ 1 be fixed and let Xn,r be the number of vertices of
degree r in a random recursive tree Tn . Then, w.h.p.

Xn,r ≈ n/2r ,

and
Xn,r − n/2r d
√ → Yr ,
n

as n → ∞, where Yr has the N(0, σr2 ) distribution.

312 Trees

In place of proving the above theorem we will give a simple proof of its imme-
diate implication, i.e., the asymptotic behavior of the expectation of the random
variable Xn,r . The proof of asymptotic normality of suitably normalized Xn,r is
due to Janson and can be found in [433]. (In fact, in [433] a stronger statement
is proved, namely, that, asymptotically, for all r ≥ 1, random variables Xn,r are
jointly Normally distributed.)

Corollary 15.12 Let r ≥ 1 be fixed. Then

E Xn,r ≈ n/2r .

Proof Let us introduce a random variable Yn,r counting the number of vertices
of degree at least r in Tn . Obviously,

Xn,r = Yn,r −Yn,r+1 . (15.14)

Moreover, using a similar argument to that given for formula (15.7), we see
that for 2 ≤ r ≤ n,
n−2 1
E[Yn,r |Tn−1 ] = Yn−1,r + Yn−1,r−1 (15.15)
n−1 n−1
Notice, that the boundary condition for the recursive formula (15.15) is, triv-
ially given by
EYn,1 = n.

We will show, that EYn,r /n → 2−r+1 which, by (15.14), will imply the theorem.
Set
an,r := n2−r+1 − EYn,r . (15.16)

EYn,1 = n implies that an,1 = 0. We see from (15.11) that the expected number
of leaves in a random recursive tree on n vertices is given by
n 1
E Xn,1 = + .
2 n−1
Hence an,2 = 1/(n − 1) as EYn,2 = n − E Xn,1 .
Now we show that,

0 < an,1 < an,2 < · · · < an,n−1 . (15.17)

From the relationships (15.15) and (15.16) we get

n−2 1
an,r = an−1,r + an−1,r−1 . (15.18)
n−1 n−1
 15.2 Recursive Trees 313

Inductively assume that (15.17) holds for some n ≥ 3. Now, by (15.18), we get
n−2 1
an,r > an−1,r−1 + an−1,r−1 = an−1,r−1 .
n−1 n−1
Finally, notice that
2
an,n−1 = n22−n − ,
(n − 1)!
since there are only two recursive trees with n vertices and a vertex of degree
n − 1. So, we conclude that a(n, r) → 0 as n → ∞, for every r, and our theorem
follows.
Finally, consider the maximum degree ∆n = ∆n (Tn ) of a random recursive tree
Tn . It is easy to see that for large n, its expected value should exceed log n, since
it is as large as the expected degree of the vertex 1, which by Theorem 15.9
equals Hn−1 ≈ log n. Szymański [722] proved that the upper bound is O(log2 n)
(see Goh and Schmutz [383] for a strengthening of his result). Finally, Devroye
and Lu (see [244]) have shown that in fact ∆n ≈ log2 n . This is somewhat
surprising. While each vertex in [1, n1−o(1) ] only has a small chance of having
such a degree, there are enough of these vertices to guarantee one w.h.p..

Theorem 15.13 In a random recursive tree Tn , w.h.p.

∆n ≈ log2 n.

The next theorem was originally proved by Devroye [239] and Pittel [643].
Both proofs were based on an analysis of certain branching processes. The
proof below is related to [239].

Theorem 15.14 Let h(Tn ) be the height of a random recursive tree Tn . Then
w.h.p.
h(Tn ) ≈ e log n.
Proof
Upper Bound: For the upper bound we simply estimate the number ν1 of
vertices at height h1 = (1 + ε)e log n where ε = o(1) but is sufficiently large so
that claimed inequalities are valid. Each vertex at this height can be associated
with a path i0 = 1, i1 , . . . , ih of length h in Tn . So, if S = {i1 , . . . , ih } refers to
such a path, then
!h1
n
1 1 1
E ν1 = ∑ ∏ ≤ ∑i ≤
|S|=h i∈S
i − 1 h1 ! i=1
1
314 Trees
 h1
(1 + log n)e
= o(1), (15.19)
h1
assuming that h1 ε → ∞.
Explanation: If S = i1 = 1, i2 , . . . , ih1 then the term ∏hj=1
 1
1/i j is the proba-
bility that i j chooses i j−1 in the construction of Tn .
Lower Bound: The proof of the lower bound is more involved. We consider
a different model of tree construction and relate it to Tn . We consider a Yule
process. We run the process for a specific time t and construct a tree Y (t). We
begin by creating a single particle x1 at time 0 this will be the root of a tree
Y (t). New particles are generated at various times τ1 = 0, τ2 , . . . ,. Then at time
τk there will be k particles Xk = {x1 , x2 , . . . , xk } and we will have Y (t) = Y (τk )
for τk ≤ t < τk+1 . After xk has been added to Y (τk ), each x ∈ Xk is associated
with an exponential random variable Ex with mean one1 . If zk is the particle
in Xk that minimizes Ex , x ∈ Xk then a new particle xk+1 is generated at time
τk+1 = τk + Ezk and an edge {zk , xk+1 } is added to Y (τk ) to create Y (τk+1 ).
After this we independently generate new random variables Ex , x ∈ Xk+1 .

Remark 15.15 The memory-less property of the exponential random vari-

able, i.e. P(Z ≥ a + b | Z ≥ a) = P(Z ≥ b), implies that we could equally well
think that at time t ≥ τk the Ex are independent exponentials conditional on
being at least τk . In which case the choice of zk is uniformly random from Xk ,
even conditional on the processes prior history.

Suppose then that we focus attention on Y (y; s,t), the sub-tree rooted at y con-
taining all descendants of y that are generated after time s and before time t.
We observe three things:

(T1) The tree Y (τn ) has the same distribution as Tn . This is because each par-
ticle in Xk is equally likely to be zk .
(T2) If s < t and y ∈ Y (s) then Y (y; s,t) is distributed as Y (t − s). This follows
from Remark 15.15, because when zk ∈ / Y (y; s,t) it does not affect any of
the the variables Ex , x ∈ Y (y; s,t).
(T3) If x, y ∈ Y (s) then Y (x; s,t) and Y (y; s,t) are independent. This also fol-
lows from Remark 15.15 for the same reasons as in (T2).

It is not difficult to prove (see Exercise (vii) or Feller [294]) that if Pn (t) is the
1 An exponential random variable Z with mean λ is characterised by P(Z ≥ x) = e−x/λ .
 15.2 Recursive Trees 315

probability there are exactly n particles at time t then

Pn (t) = e−t (1 − e−t )n−1 . (15.20)
Next let
t1 = (1 − ε) log n.
Then it follows from (15.20) that if ν(t) is the number of particles in our Yule
process at time t then
 n−1
−t1 −t1 k−1 1
P(ν(t1 ) ≥ n) ≤ ∑ e (1 − e ) = 1 − 1−ε = o(1). (15.21)
k≥n n

We will show that w.h.p. the tree Tν(t1 ) has height at least
h0 = (1 − ε)et1
and this will complete the proof of the theorem.
We will choose s → ∞, s = O(logt1 ). It follows from (15.20) that if ν0 = εes
then
ν0
P(ν(s) ≤ ν0 ) = ∑ e−s (1 − e−s )k−1 ≤ ε = o(1). (15.22)
k=0

Suppose now that ν(s) ≥ ν0 and that the vertices of T1;0,s are
 1/2
x1 , x2 , . . . , xν(s) . Let σ = ν0 and consider the sub-trees
A j , j = 1, 2, . . . , τ of T1;0,t1 rooted at x j , j = 1, 2, . . . , ν(s). We will show that
1
P(Tx1 ;s,t1 has height at least (1 − ε)3 e log n) ≥ . (15.23)
2σ log σ
Assuming that (15.23) holds, since the trees A1 , A2 , . . . , Aτ are independent, by
T3, we have

P(h(Tn ) ≤ (1 − ε)3 e log n) ≤

 ν0
1
o(1) + P(h(T1;0,t1 ) ≤ (1 − ε)3 e log n) ≤ o(1) + 1 − = o(1).
2σ log σ
To prove all this we will associate a Galton-Watson branching process with
each of x1 , x2 , . . . , xτ . Consider for example x = x1 and let τ0 = log σ . The ver-
tex x will be the root of a branching process Π, which we now define. We will
consider the construction of Y (x; s,t) at times τi = s + iτ0 for i = 1, 2, . . . , i0 =
(t1 − s)/τ0 . The children of x in Π are the vertices at depth at least (1 − ε)eτ0 in
Y (x; s, τ1 ). In general, the particles in generation i will correspond to particles
at depth at least (1 − ε)eτ0 in the tree Y (ξ ; τi−1 , τi ) where ξ is a particle of
Y (x; s,t) included in generation i − 1 of Π.
316 Trees

If the process Π does not ultimately become extinct then generation i0 corre-
sponds to vertices in Y (t) that are at depth

i0 × (1 − ε)eτ0 = (1 − ε)e(t1 − s) ≥ (1 − ε)3 e log n.

We will prove that

1
P(Π does not become extinct) ≥ , (15.24)
2σ log σ
and this implies (15.23) and the theorem.
To prove (15.24) we first show that µ, the expected number of progeny of a
particle in Π satisfies µ > 1 and after that we prove (15.24).
Let D(h, m) denote the expected number of vertices at depth h in the tree Tm .
Then for any ξ ∈ Π,

µ ≥ D((1 − ε)eτ0 , σ ) × P(ν(τ0 ) ≥ σ ). (15.25)

It follows from (15.20) and σ = eτ0 that

∞
P(|Y (ξ , 0, τ0 )| ≥ σ ) = ∑ e−τ0 (1 − e−τ0 )k
k=σ
1
= (1 − e−τ0 )σ ≥ . (15.26)
2e
We show next that for m  h we have
(log m − log h − 1)h
D(h, m) ≥ . (15.27)
h!
To prove this, we go back to (15.19) and write

1 m 1 1
D(h, m) = ∑ i−1 ∑ ∏
h i=2 j−1
S∈([2,m]\{i}
h−1 )
j∈S\{i}

m m
1 1 1 1 1 1
= ∑ ∏ j−1 ∑ ≥ ∑ ∏ j−1 ∑
h j∈S 16=k∈S
k − 1 h j∈S k=h+1 k
S∈([2,m]
h−1 )
/ S∈([2,m]
h−1 )

log m − log h − 1
≥ D(h − 1, m). (15.28)
h
Equation (15.27) follows by induction since D(1, m) ≥ log m.
Explanation of (15.28): We choose a path of length h by first choosing a vertex
i and then choosing S ⊆ [2, m] \ {i}. We divide by h because each h-set arises
1
h times in this way. Each choice will contribute ∏ j∈S∪{i} j−1 . We change the
m 1 m 1
order of summation i, S and then lower bound ∑16=k∈S
/ k−1 by ∑k=h+1 k .
 15.2 Recursive Trees 317

We now see from (15.25), (15.26) and (15.27) that

(τ0 − log((1 − ε)eτ0 ) − 1)(1−ε)eτ0 1

µ≥ × ≥
((1 − ε)eτ0 )! 2e
1 1
√ ×  1,
2e 2π (1 − ε/2)(1−ε)eτ0
if we take ετ0 / log τ0 → ∞.
We are left to prove (15.24). Let G(z) be the probability generating function for
the random variable Z equal to the number of descendants of a single particle.
We first observe that for any θ ≥ 1,
∞
P(Z ≥ θ σ ) ≤ P(|Y (ξ , 0, τ0 )| ≥ θ σ ) = ∑ e−τ0 (1 − e−τ0 )k ≤ e−θ .
k=θ σ

Note that for 0 ≤ x ≤ 1, any k ≥ 0 and a ≥ k it holds that

 
k k
1− + xa ≥ xk .
a a
We then write for 0 ≤ x ≤ 1,
θσ θσ
G(x) ≤ ∑ pk xk + P(Z ≥ θ σ ) ≤ ∑ pk xk + e−θ
k=0 k=0
θσ  
k k
≤∑ 1− pk + pk xθ σ + e−θ
k=0 θσ θσ
∞   
k k
≤∑ 1− pk + pk x θσ
+ e−θ
k=0 θσ θσ
µ µ θσ
= H(x) = 1 − + x + e−θ .
θσ θσ
The function H is monotone increasing in x and so ρ = P(Π becomes extinct)
being the smallest non-negative solution to x = G(x) (see Theorem 24.1) im-
plies that ρ is at most the smallest non-negative solution q to x = H(x). The
convexity of H and the fact that H(0) > 0 implies that q is at most the value ζ
satisfying H 0 (ζ ) = 1 or
1
q≤ζ = < 1.
µ 1/(θ σ −1)
But ρ = G(ρ) ≤ G(q) ≤ H(q) and so
 
µ 1 µ −1 1
1−ρ ≥ 1 − θ σ /(θ σ −1) − e−θ ≥ − e−θ ≥ ,
θσ µ θσ 2σ log σ
after putting θ = 2 log σ and using µ  1.
318 Trees

Devroye, Fawzi and Fraiman [240] give another proof of the above theorem
that works for a wider class of random trees called scaled attachment random
recursive trees, where each vertex i attaches to the random vertex biXi c and
X0 , . . . , Xn is a sequence of independent identically distributed random vari-
ables taking values in [0, 1).

15.3 Inhomogeneous Recursive Trees

Plane-oriented recursive trees
This section is devoted to the study of the properties of a class of inhomoge-
neous recursive trees that are closely related to the Barabási-Albert “preferen-
tial attachment model”, see [57]. Bollobás, Riordan, Spencer and Tusnády gave
a proper definition of this model and showed how to reduce it to random plane-
oriented recursive trees, see [151]. In this section we present some results that
preceded [57] and created a solid mathematical ground for the further develop-
ment of general preferential attachment models, which will be discussed later
in the book (see Chapter 18).
Suppose that we build a recursive tree in the following way. We start as before
with a single vertex labeled 1 and add n − 1 vertices labeled 2, 3, . . . , n, one by
one. We assume that the children of each vertex are ordered (say, from left to
right). At each step a new vertex added to the tree is placed in a position “in
between” old vertices. A tree built in this way is called a plane-oriented recur-
sive tree. To study this model it is convenient to introduce an extension of a
plane-oriented recursive tree: given a plane-oriented recursive tree we connect
each vertex with external nodes, representing a possible insertion position for
an incoming new vertex. See Figure 15.3 for a diagram of all plane-oriented
recursive trees on n = 3 vertices, together with their extensions.
Assume now, as before that all the edges of a tree are directed toward the
leaves, and denote the out-degree of a vertex v by d + (v). Then the total number
of extensions of an plane-oriented recursive tree on n vertices is equal to

∑ (d + (v) + 1) = 2n − 1.
v∈V

So a new vertex can choose one those those 2n − 1 places to join the tree and
create a tree on n+1 vertices. If we assume that this choice in each step is made
uniformly at random then a tree constructed this way is called a random plane-
oriented recursive tree. Notice that the probability that the vertex labeled n + 1
+ (v)+1
is attached to vertex v is equal to d 2n−1 i.e., it is proportional to the degree
of v. Such random trees, called plane-oriented because of the above geometric
 15.3 Inhomogeneous Recursive Trees 319
00
11
1 11
00 00
11
1 11
00 1 11
00
00
11
00
11
00
11 00
11

2 11
00
00
11
00
11

00
11
3 11
00
00
11 2 11
00
00
11 11
00
00
113 3 11
00
00
11
00
11
11
00
002
11
00
11
00
11 00
11

00
11 1 11
00 1 11
00
00
11
00
11
1 11
00 00
11
00
11
00
11

2 11
00
00
11
00
11
2 11
00
00
11
00
11
11
00
003
11
00
11
3 11
00
00
11
00
11
11
00
002
11
00
11

00
11
3 11
00
00
11

Figure 15.1 Plane-oriented recursive trees and their extensions, n = 3

interpretation, were introduced by Szymański [721] under the name of non-

uniform recursive trees. Earlier, Prodinger and Urbanek [655] described plane-
oriented recursive trees combinatorially, as labeled ordered (or plane) trees
with the property that labels along any path down from the root are increasing.
Such trees are also known in the literature as heap-ordered trees (see Chen and
Ni [180], Prodinger [654], Morris, Panholzer and Prodinger [606]) or, more
recently, as scale-free trees. So, random plane-oriented recursive trees are the
simplest example of random preferential attachment graphs.
Denote by an the number of plane-oriented recursive trees on n vertices. This
number, for n ≥ 2 satisfies an obvious recurrence relation
an+1 = (2n − 1)an .
Solving this equation we get that
an = 1 · 3 · 5 · · (2n − 3) = (2n − 3)!!.
This is also the number of Stirling permutations, introduced by Gessel and
Stanley [373], i.e. the number of permutations of the multiset
{1, 1, 2, 2, 3, 3, . . . , n, n}, with the additional property that, for each value of
1 ≤ i ≤ n, the values lying between the two copies of i are greater than i.
There is a one-to-one correspondence between such permutations and plane-
oriented recursive trees, given by Koganov [501] and, independently, by Janson
320 Trees

[435]. To see this relationship consider a plane-oriented recursive tree on n + 1

vertices labelled 0, 1, 2, . . . , n, where the vertex with label 0 is the root of the
tree and is connected to the vertex labeled 1 only, and the edges of the tree are
oriented in the direction from the root. Now, perform a depth first search of
the tree in which we start from the root. Next we go to the leftmost child of
the root, explore that branch recursively, go to the next child in order etc., until
we stop at the root. Notice that every edge in such a walk is traversed twice. If
every edge of the tree gets a label equal to the label of its end-vertex furthest
from the root, then the depth first search encodes each tree by a string of length
2n, where each label 1, 2, . . . , n appears twice. So the unique code of each tree
is a unique permutation of the multiset {1, 1, 2, 2, 3, 3, . . . , n, n} with additional
property described above. Note also that the insertion of a pair (n + 1, n + 1)
into one of the 2n − 1 gaps between labels of the permutation of this multiset,
corresponds to the insertion of the vertex labeled n + 1 into a plane-oriented
recursive tree on n vertices.
Let us start with exact formulas for probability distribution of the out-degree
D+n,i of a vertex with label i, i = 1, 2, . . . , n in a random plane-oriented recursive
tree. Kuba and Panholzer [527] proved the following theorem.
Theorem 15.16 For i = 1, 2, . . . , n and r = 1, 2, . . . , n − 1,
r  
r Γ(n − 3/2)Γ(i − 1/2)
P(D+n,i = r) = ∑ k (−1)k Γ(i − 1 − k/2)Γ(n − 1/2) ,
k=0
R ∞ z−1 −t
where Γ(z) = 0 t e dt is the Gamma function. Moreover,
2i−2
n−i
i−1 4
E(D+
n,i ) = 2n−2

−1 (15.29)
n−1

For simplicity, we show below that the formula (15.29) holds for i = 1, i.e.,
the expected value of the out-degree of the root of a random plane-oriented
recursive tree, and investigate its behavior as n → ∞. It is then interesting to
compare the latter with the asymptotic behavior of the degree of the root of a
random recursive tree. Recall that for large n this is roughly log n (see Theorem
15.10).
The result below was proved by Mahmoud, Smythe and Szymański [566].

Corollary 15.17 For n ≥ 2 the expected value of the degree of the root of a
random plane-oriented recursive tree is
4n−1
E(D+
n,1 ) = 2n−2

− 1,
n−1
 15.3 Inhomogeneous Recursive Trees 321

and,
√
E(D+
n,1 ) ≈ πn.
Proof Denote by
n
4n 2i (2n)!!
un = 2n
= ∏ = .
− − 1)!!


n i=1 2i 1 (2n

Hence, in terms of un , we want to prove that E(D+ n,1 ) = un−1 − 1.

It is easy to see that the claim holds for n = 1, 2 and that
n−1  
2 1
P(D+ n,1 = 1) = ∏ 1 − = ,
i=1 2i − 1 2n − 3
while, for r > 1 and n ≥ 1,

P(D+
n+1,1 = r) =
 
r+1 r
1− P(D+
n,1 = r) + P(D+
n,1 = r − 1).
2n − 1 2n − 1
Hence
E(D+n+1,1 )
n  
2n − r − 2 r
= ∑r P(D+
n,1 = r) + P(D +
n,1 = r − 1) =
r=1 2n − 1 2n − 1
!
n−1 n−1
1
∑ r(2n − r − 2) P(D+
n,1 = r) + ∑ (r + 1) 2
P(D+
n,1 = r)
2n − 1 r=1 r=1
n
1
= ∑ (2nr + 1) P(D+n,1 = r).
2n − 1 r=1
So, we get the following recurrence relation
2n 1
E(D+
n+1,1 ) = E(D+ n,1 ) +
2n − 1 2n − 1
and the first part of the theorem follows by induction.
To see that the second part also holds one has to use the Stirling approximation
to check that
√ 3p
un = πn − 1 + π/n + · · · .
8

The next theorem, due to Kuba and Panholzer [527], summarizes the asymp-
totic behavior of the suitably normalized random variable D+
n,i .
322 Trees

Theorem 15.18 Let i ≥ 1 be fixed and let n → ∞. If

(i) i = 1, then
d 2 /2
n−1/2 D+
n,1 → D1 , with density f D1 (x) = (x/2)e
−x
,
√
i.e., is asymptotically Rayleigh distributed with parameter σ = 2,
d
(ii) i ≥ 2, then n−1/2 D+n,i → Di , with density

2i − 3
Z ∞
2 /4
fDi (x) = 2i−1
(t − x)2i−4 e−t dt.
2 (i − 2)! x

Let i = i(n) → ∞ as n → ∞. If
(i) i = o(n), then the normalized random variable (n/i)−1/2 D+n,i is asymptot-
ically Gamma distributed γ(α, β ), with parameters α = −1/2 and β = 1,
(ii) i = cn, 0 < c < 1, then the random variable D+n,i is asymptotically nega-
tive binomial distributed NegBinom(r, p) with parameters r = 1 an p =
√
c,
(iii) n − i = o(n), then P(D+ n,i = 0) → 1, as n → ∞.

We now turn our attention to the number of vertices of a given out-degree. The
next theorem shows a characteristic feature of random
graphs built by preferential attachment rule where every new vertex prefers to
attach to a vertex with high degree (rich get richer rule). The proportion of
vertices with degree r in such a random graph with n vertices grows like n/rα ,
for some constant α > 0, i.e., its distribution obeys a so called power law. The
next result was proved by Szymański [721] (see also [566] and [723]) and it
indicates such a behavior for the degrees of the vertices of a random plane-
oriented recursive tree, where α = 3.

Theorem 15.19 Let r be fixed and denote by Xn,r + the number of vertices of

out-degree r in a random plane-oriented recursive tree Tn . Then,

 
+ 4n 1
E Xn,r = +O .
(r + 1)(r + 2)(r + 3) r
Proof Observe first that conditional on Tn ,
+ + r+1 + r
E(Xn+1,r |Tn ) = Xn,r − X + X + + 1r=0 , (15.30)
2n − 1 n,r 2n − 1 n,r−1
which gives
+ 2n − r − 2 + r +
E Xn+1,r = E Xn,r + E Xn,r−1 + 1r=0 (15.31)
2n − 1 2n − 1
 15.3 Inhomogeneous Recursive Trees 323
+
for r ≥ 1, (Xn,−1 = 0).
We will show that the difference
de f + 4n
an,r = E Xn,r − .
(r + 1)(r + 2)(r + 3)
is asymptotically negligible with respect to the leading term in the statement
of the theorem. Substitute an,r in the equation (15.31) to get that for r ≥ 1,
2n − r − 2 r 1
an+1,r = an,r + an,r−1 − . (15.32)
2n − 1 2n − 1 2n − 1
2
We want to show that |an,r | ≤ max{r,1} , for all n ≥ 1, r ≥ 0. Note that this is true
for all n and r = 0, 1, since from (15.31) it follows (inductively) that for n ≥ 2

+ 2n − 1 1
E Xn,0 = and so an,0 = − .
3 3
For n ≥ 2,

+ n 1 3 1 3
E Xn,1 = − + and so an,1 = − + .
6 12 4(2n − 3) 12 4(2n − 3)
We proceed by induction on r. By definition
4r
ar,r = − ,
(r + 1)(r + 2)(r + 3)
and so,
2
|ar,r | < .
r
We then see from (15.32) that for and r ≥ 2 and n ≥ r that
2n − r − 2 2 r 2 1
|an+1,r | ≤ · + · − .
2n − 1 r 2n − 1 r − 1 2n − 1
r2
 
2 2 r
= − r+1− −
r (2n − 1)r r−1 2
2
≤ ,
r
which completes the induction and the proof of the theorem.
In fact much more can be proved.

Theorem 15.20 Let ε > 0 and r be fixed. Then, w.h.p.

+
Xn,r
(1 − ε)ar ≤ ≤ (1 + ε)ar , (15.33)
n
324 Trees

where
4
ar = .
(r + 1)(r + 2)(r + 3)
Moreover,
+ − na


Xn,r r d
√ → Yr , (15.34)
n
as n → ∞, jointly for all r ≥ 0, where the Yr are jointly Normally distributed
with expectations EYr = 0 and covariances σrs = Cov(Yr ,Ys ) given by
r s
(−1)k+l r
    
s 2(k + l + 4)! (k + 1)(l + 1)
σrs = 2 ∑ ∑ −1− .
k=0 l=0 k + l + 4 k l (k + 3)!(l + 3)! (k + 3)(l + 3)
Proof For the proof of asymptotic normality of a suitably normalized random
+ , i.e., for the proof of statement (15.34)) see Janson [433]. We will
variable Xn,r
give a short proof of the first statement (15.33), due to Bollobás, Riordan,
Spencer and Tusnády [151] (see also Mori [604]).
Consider a random plane-oriented recursive tree Tn as an element of a process
∞ . Fix n ≥ 1 and r ≥ 0 and for 0 ≤ t ≤ n define the martingale
(Tt )t=o
+ + +
Yt = E(Xn,r |Tt ) where Y0 = E(Xn,r ) and Yn = Xn,r .
One sees that the differences
|Yt+1 −Yt | ≤ 2.
For a proof of this, see the proof of Theorem 18.3. Applying the
Hoeffding- Azuma inequality (see Theorem 22.16) we get, for any fixed r,
+ +
| ≥ n log n) ≤ e−(1/8) log n = o(1).
p
P(|Xn,r − E Xn,r
√
+  n log n and (15.33)
But Theorem 15.19 shows that for any fixed r, E Xn,r
follows.
Similarly, as for uniform random recursive trees, Pittel [643] established the
asymptotic behavior of the height of a random plane-oriented recursive tree.

Theorem 15.21 Let h∗n be the height of a random plane-oriented recursive

tree. Then w.h.p.
log n
h∗n ≈ ,
2γ
where γ is the unique solution of the equation
γeγ+1 = 1,
1
i.e., γ = 0.27846.., so 2γ = 1.79556....
 15.3 Inhomogeneous Recursive Trees 325

Inhomogeneous recursive trees: a general model

As before, consider a tree that grows randomly in time. Each time a new vertex
appears, it chooses exactly one of the existing vertices and attaches to it. This
way we build a tree Tn of order n with n + 1 vertices labeled {0, 1, . . . , n},
where the vertex labeled 0 is the root. Now assume that for every n ≥ 0 there
is a probability distribution
n
P(n) = (p0 , p1 , . . . , pn ), ∑ p j = 1.
j=0

Suppose that Tn has been constructed for some n ≥ 1. Given Tn we add an

edge connecting one of its vertices with a new vertex labeled n + 1 and thus
forming a tree Tn+1 . A vertex vn ∈ {0, 1, 2, . . . , n} is chosen to be a neighbor of
the incoming vertex with probability

P(vn = j|Tn ) = p j , for j = 0, 1, . . . , n.

Note that for the uniform random recursive tree we have

1
pj = , for 0 ≤ j ≤ n.
n+1
We say that a random recursive tree is inhomogeneous if the attachment rule
of new vertices is determined by a non-uniform probability distribution. Most
often the probability that a new vertex chooses a vertex j ∈ {0, 1, . . . , n} is
proportional to w(dn ( j)), the value of a weight function w applied to the degree
dn ( j) of vertex j after n-th step. Then the probability distribution P(n) is defined
w(dn ( j))
pj = .
∑nk=0 w(dn (k))
Consider a special case when the weight function is linear and, for 0 ≤ j ≤ n,

w(dn ( j)) = dn ( j) + β , β > −1, (15.35)

so that the total weight

n
wn = ∑ (dn (k) + β ) = 2n + (n + 1)β . (15.36)
k=0

Obviously the model with such probability distribution is only a small gen-
eralisation of plane-oriented random recursive trees and we obtain the latter
when we put β = 0 in (15.35). Inhomogeneous random recursive trees of this
type are known in the literature as either scale free random trees or Barabási-
Albert random trees. For obvious reasons, we will call such graphs generalized
random plane-oriented recursive trees.
326 Trees

Let us focus the attention on the asymptotic behavior of the maximum degree
of such random trees. We start with some useful notation and observations.
Let Xn, j denote the weight of vertex j in a generalized plane-oriented random
recursive tree, with initial values X1,0 = X j, j = 1 + β for j > 0. Let
 
Γ n + β β+2
cn,k =   , n ≥ 1, k ≥ 0,
+k
Γ n + ββ +2

be a double sequence of normalising constants. Note that

cn+1,k wn
= , (15.37)
cn,k wn + k
and, for any fixed k,
cn,k = n−k/(β +2) (1 + O(n−1 )).
Let k be a positive integer and
 
Xn, j + k − 1
Xn, j;k = cn,k .
k

Lemma 15.22 Let Fn be
the σ -field generated by the first n steps. If n ≥
max{1, j}, then Xn, j;k , Fn is a martingale.
Proof Because Xn+1, j − Xn, j ∈ {0, 1}, we see that
 
Xn+1, j + k − 1
k
    
Xn, j + k − 1 Xn, j + k − 1 Xn+1, j − Xn, j
= +
k k−1 1
  
Xn, j + k − 1 k(Xn+1, j − Xn, j )
= 1+ .
k Xn, j
Hence, noting that
Xn, j
P(Xn+1, j − Xn, j = 1|Fn ) = ,
wn
and applying (15.37)
 
cn+1,k k
E(Xn+1, j;k |Fn ) = Xn, j;k 1+ = Xn, j;k ,
cn,k wn
we arrive at the lemma.
Thus, the random variable Xn, j;k , as a non-negative martingale, is
 15.3 Inhomogeneous Recursive Trees 327

bounded in L1 and it almost surely converges to X jk /k!, where X j is the limit

of Xn, j;1 . Since Xn, j;k ≤ cXn, j;2k , where the constant c does not depend on n, it
is also bounded in L2 , which implies that it converges in L1 . Therefore we can
determine all moments of the random variable X j . Namely, for j ≥ 1,

X jk
 
β +k
= lim E Xn, j;k = X j, j;k = c j,k . (15.38)
k! n→∞ k
Let ∆n be the maximum degree in a generalized random plane-oriented recur-
sive tree Tn and let, for j ≤ n,

∆n, j = max Xn,i;1 = max cn,1 Xn,i .

0≤i≤ j 0≤i≤ j

Note that since Xn,i is the weight of vertex i, i.e., its degree plus β , we find that
∆n,n = cn,1 (∆n + β ). Define

ξ j = max Xi and ξ = ξ∞ = sup X j . (15.39)

0≤i≤ j j≥0

Now we are ready to prove the following result, due to Móri [605].

Theorem 15.23
 
P lim n−1/(β +2) ∆n = ξ = 1.
n→∞

The limiting random variable ξ is almost surely finite and positive and it has
an absolutely continuous distribution. The convergence also holds in L p , for
all p, 1 ≤ p < ∞.

Proof In the proof we skip the part dealing with the positivity of ξ and the
absolute continuity of its distribution.
By Lemma 15.22, ∆n,n is the maximum of martingales, therefore
(∆n,n |F ) is a non-negative sub-martingale, and so
n ∞  
β +k ∞
E ∆kn,n ≤ ∑ E Xn,k j;1 ≤ ∑ E X jk = k! ∑ c j,k < ∞,
j=0 j=0 k j=0

if k > β + 2. (Note c0,k is defined here as equal to c1,k ). Hence (∆n,n |F ) is

bounded in Lk , for every positive integer k, which implies both almost sure
convergence and convergence in L p , for any p ≥ 1.
Assume that k > β + 2 is fixed. Then, for n ≥ k,
n
E(∆n,n − ∆n, j )k ≤ ∑ E Xn,i;1 .
i= j+1
 328 Trees

Take the limit as n → ∞ of both sides of the above inequality. Applying (15.39)
and (15.38), we get
∞   ∞
 k β +k
E lim n−1/(β +2) ∆n − ξ j ≤ ∑ E ξik = k! ∑ c j,k .
n→∞
i= j+1 k i= j+1

The right-hand side tends to 0 as j → ∞, which implies that

n−1/(β +2) ∆n tends to ξ , as claimed.
To conclude this section, setting β = 0 in Theorem 15.23, one can obtain the
asymptotic behavior of the maximum degree of a plane-oriented random re-
cursive tree.

15.4 Exercises
(i) Use the Prüfer code to show that there is one-to-one correspondence be-
tween the family of all labeled trees with vertex set [n] and the family of all
ordered sequences of length n − 2 consisting of elements of [n].
(ii) Prove Theorem 15.1.
(iii) Let ∆ be the maximum degree of a random labeled tree on n vertices. Use
(15.1) to show that for every ε > 0, P(∆ > (1 + ε) log n/ log log n) tends to
0 as n → ∞.
(iv) Let ∆ be defined as in the previous exercise and let t(n, k) be the number
of labeled trees on n vertices with maximum  degree at most k. Knowing
n
1
that t(n, k) < (n − 2)! 1 + 1 + 2!1 + . . . + (k−1)! , show that for every ε >
0, P(∆ < (1 − ε) log n/ log log n) tends to 0 as n → ∞.
(v) Determine a one-to-one correspondence between the family of permutations
on {2, 3, . . . , n} and the family of recursive trees on the set [n].
(vi) Let Ln denote the number of leaves of a random recursive tree with n ver-
tices. Show that E Ln = n/2 and Var Ln = n/12.
(vii) Prove (15.20).
(viii) Show that Φn,i (z) given in Theorem 15.8 is the probability generating func-
tion of the convolution of n − i independent Bernoulli random variables with
success probabilities equal to 1/(i + k − 1) for k = 1, 2, . . . , n − i.
(ix) Let Ln∗ denotes the number of leaves of a random plane-oriented recursive
tree with n vertices. Show that
2n − 1 2n(n − 2)
E Ln∗ = and Var Ln∗ = .
3 9(2n − 3)
(x) Prove that Ln∗ /n (defined above) converges in probability, to 2/3.
 15.5 Notes 329

15.5 Notes
Labeled trees
The literature on random labeled trees and their generalizations is very exten-
sive. For a comprehensive list of publications in this broad area we refer the
reader to a recent book of Drmota [254], to a chapter of Bollobás’s book [132]
on random graphs, as well as to the book by Kolchin [508]. For a review of
some classical results, including the most important contributions, forming the
foundation of the research on random trees, mainly due to Meir and Moon
(see, for example : [586], [587]and [589]), one may also consult a survey by
Karoński [475].

Recursive trees
Recursive trees have been introduced as probability models for system gen-
eration (Na and Rapoport [613]), spread of infection (Meir and Moon [588]),
pyramid schemes (Gastwirth [367]) and stemma construction in philology (Na-
jock and Heyde [617]). Most likely, the first place that such trees were intro-
duced in the literature, is the paper by Tapia and Myers [726], presented there
under the name “concave node-weighted trees”. Systematic studies of random
recursive trees were initiated by Meir and Moon ([588] and [603]) who inves-
tigated distances between vertices as well as the process of cutting down such
random trees. Observe that there is a bijection between families of recursive
trees and binary search trees, and this has opened many interesting directions
of research, as shown in a survey by Mahmoud and Smythe [565] and the book
by Mahmoud [563].
Early papers on random recursive trees (see, for example, [613], [367] and
[253]) were focused on the distribution of the degree of a given vertex and of
the number of vertices of a given degree. Later, these studies were extended
to the distribution of the number of vertices at each level, which is referred to
as the profile. Recall, that in a rooted tree, a level (strata) consists of all those
vertices that are at the same distance from the root.
The profile of a random recursive tree is analysed in many papers. For example,
Drmota and Hwang [255] derive asymptotic approximations to the correlation
coefficients of two level sizes in random recursive trees and binary search trees.
These coefficients undergo sharp sign-changes when one level is fixed and the
other is varying. They also propose a new means of deriving an asymptotic
estimate for the expected width, which is the number of nodes at the most
abundant level.
Devroye and Hwang [242] propose a new, direct, correlation-free approach
330 Trees

based on central moments of profiles to the asymptotics of width in a class of

random trees of logarithmic height. This class includes random recursive trees.
Fuchs, Hwang, Neininger [361] prove convergence in distribution for the pro-
file, normalized by its mean, of random recursive trees when the limit ratio
α of the level and the logarithm of tree size lies in [0, e). Convergence of all
moments is shown to hold only for α ∈ (0, 1) (with only convergence of finite
moments when α ∈ (1, e)).
van der Hofstadt, Hooghiemstra and Van Mieghem [418] study the covariance
structure of the number of nodes k and l steps away from the root in random
recursive trees and give an analytic expression valid for all k, l and tree sizes n.
For an arbitrary positive integer i ≤ in ≤ n − 1, a function of n, Su, Liu and
Feng [719] demonstrate the distance between nodes i and n in random recursive
trees Tn , is asymptotically normal as n → ∞ by using the classical limit theory
method.
Holmgren and Janson [420] proved limit theorems for the sums of functions
of sub-trees of binary search trees and random recursive trees. In particular,
they give new simple proofs of the fact that the number of fringe trees of size
k = kn in a binary search tree and the random recursive tree (of total size n)
asymptotically has a Poisson distribution if k → ∞, and that the distribution
√
is asymptotically normal for k = o( n). Recall that a fringe tree is a sub-tree
consisting of some vertex of a tree and all its descendants (see Aldous [14]).
For other results on that topic see Devroye and Janson [243].
Feng, Mahmoud and Panholzer [295] study the variety of sub-trees lying on the
fringe of recursive trees and binary search trees by analysing the distributional
behavior of Xn,k , which counts the number of sub-trees of size k in a random
tree of size n, with k = k(n). Using analytic methods, they characterise for both
tree families the phase change behavior of Xn,k .
One should also notice interesting applications of random recursive trees. For
example, Mehrabian [585] presents a new technique for proving logarithmic
upper bounds for diameters of evolving random graph models, which is based
on defining a coupling between random graphs and variants of random re-
cursive trees. Goldschmidt and Martin [384] describe a representation of the
Bolthausen-Sznitman coalescent in terms of the cutting of random recursive
trees.
Bergeron, Flajolet, Salvy [81] have defined and studied a wide class of ran-
dom increasing trees. A a tree with vertices labeled {1, 2, . . . , n} is increasing
if the sequence of labels along any branch starting at the root is increasing.
Obviously, recursive trees and binary search trees (as well as the general class
of inhomogeneous trees, including plane-oriented trees) are increasing. Such a
general model, which has been intensively studied, yields many important re-
 15.5 Notes 331

sults for random trees discussed in this chapter. Here we will restrict ourselves
to pointing out just a few papers dealing with random increasing trees authored
by Dobrow and Smythe [252], Kuba and Panholzer [527] and Panholzer and
Prodinger [631], as well as with their generalisations, i.e., random increasing
k-trees, published by Zhang, Rong, and Comellas [757], Panholzer and Seitz
[632] and Darrasse, Hwang and Soria [233],

Inhomogeneous recursive trees

Plane-oriented recursive trees
As we already mentioned in Section 15.3, Prodinger and Urbanek [655], and,
independently, Szymański [721] introduced the concept of plane-oriented ran-
dom trees (more precisely, this notion was introduced in an unpublished pa-
per by Dondajewski and Szymański [253]), and studied the vertex degrees of
such random trees. Mahmoud, Smythe and Szymański [566], using Pólya urn
models, investigated the exact and limiting distributions of the size and the
number of leaves in the branches of the tree (see [435] for a follow up). Lu
and Feng [544] considered the strong convergence of the number of vertices
of given degree as well as of the degree of a fixed vertex (see also [565]). In
Janson’s [433] paper, the distribution of vertex degrees in random recursive
trees and random plane recursive trees are shown to be asymptotically normal.
Brightwell and Luczak [159] investigate the number Dn,k of vertices of each
degree k at each time n, focusing particularly on the case where k = k(n) is a
growing function of n. They show that Dn,k is concentrated around its mean,
which is approximately 4n/k3 , for all k ≤ (n log n)−1/3 , which is best possible
up to a logarithmic factor.
Hwang [422] derives several limit results for the profile of random plane-
oriented recursive trees. These include the limit distribution of the normalized
profile, asymptotic bimodality of the variance, asymptotic approximation to
the expected width and the correlation coefficients of two level sizes.
Fuchs [360] outlines how to derive limit theorems for the number of sub-trees
of size k on the fringe of random plane-oriented recursive trees.
Finally, Janson, Kuba and Panholzer [437] consider generalized Stirling per-
mutations and relate them with certain families of generalized plane recursive
trees.

Generalized recursive trees

Móri [604] proves the strong law of large numbers and central limit theorem for
the number of vertices of low degree in a generalized random plane-oriented
recursive tree. Szymański [723] gives the rate of concentration of the number
332 Trees

of vertices with given degree in such trees. Móri [605] studies maximum de-
gree of a scale-free trees. Zs. Katona [488] shows that the degree distribution
is the same on every sufficiently high level of the tree and in [487] investigates
the width of scale-free trees.
Rudas, Toth, Valko [680], using results from the theory of general branching
processes, give the asymptotic degree distribution for a wide range of weight
functions. Backhausz and Móri [45] present sufficient conditions for the almost
sure existence of an asymptotic degree distribution constrained to the set of
selected vertices and describe that distribution.
Bertoin, Bravo [82] consider Bernoulli bond percolation on a large scale-free
tree in the super-critical regime, i.e., when there exists a giant cluster with high
probability. They obtain a weak limit theorem for the sizes of the next largest
clusters, extending a result in Bertoin [84] for large random recursive trees.
Devroye, Fawzi, Fraiman [240] study depth properties of a general class of
random recursive trees called attachment random recursive trees. They prove
that the height of such tree is asymptotically given by αmax log n where αmax
is a constant. This gives a new elementary proof for the height of uniform
random recursive trees that does not use branching random walk. For further
generalisations of random recursive trees see Mahmoud [564].
 16
Mappings

In the evolution of the random graph Gn,p , during its sub-critical phase, tree
components and components with exactly one cycle, i.e. graphs with the same
number of vertices and edges, are w.h.p. the only elements of its structure.
Similarly, they are the only graphs outside the giant component after the phase
transition, until the random graph becomes connected w.h.p. In the previous
chapter we studied the properties of random trees. Now we focus our attention
on random mappings of a finite set into itself. Such mappings can be repre-
sented as digraphs with the same number of vertices and edges. So the study of
their “average” properties may help us to better understand the typical structure
of classical random graphs. We start the chapter with a short look at the basic
properties of random permutations (one-to-one mappings) and then continue
to the general theory of random mappings.

16.1 Permutations
Let f be chosen uniformly at random from the set of all n! permutations on
the set [n], i.e., from the set of all one-to-one functions [n] → [n]. In this sec-
tion we will concentrate our attention on the properties of a functional digraph
representing a random permutation.
Let D f be the functional digraph ([n], (i, f (i))). The digraph D f consists of
vertex disjoint cycles of any length 1, 2, . . . , n. Loops represent fixed points,
see Figure 16.1.
Let Xn,t be the number of cycles of length t, t = 1, 2, . . . , n in the digraph D f .
Thus Xn,1 counts the number of fixed points of a random permutation. One can
easily check that

1 bn/tc (−1)i e−1/t

P(Xn,t = k) = ∑ → as n → ∞, (16.1)
k!t k i=0 t i i! t k k!
for k = 0, 1, 2, . . . , n. Indeed, convergence in (16.1) follows directly from Lemma
21.10 and the fact that
((t − 1)!)i (n − ti)!
 
Xn,t 1 n! 1
Bi = E = · = i .
i n! (t!)i (n − ti)! i! t i!

333
334 Mappings

x 1 2 3 4 5 6 7 8 9 10

f(x) 3 10 1 4 2 6 8 9 7 5

10
11
00
711
00
8
00
11
11
00
00
11
2 11
00 00
11 11
00 00
11
00
11 00
11 00
11 00
11
11
00
00
11
00
11
4
00
11 11
00
00
11
11
00
00
11 00
11
5 9

11
00 11
00 11
00
00
11
00
11 00
11 00
11
1 11
00 00
11
3 6

Figure 16.1 A permutation digraph example

This means that Xn,t converges in distribution to a random variable with Pois-
son distribution with mean 1/t.
Moreover, direct computation gives

P(Xn,1 = j1 , Xn,2 = j2 , . . . , Xn,n = jn )

n
1 n!
= n
n! ∏t=1 jt !(t!) jt ∏((t − 1)!) jt
t=1
n
1 jt 1
 
=∏ ,
t=1 t jt !

n
for non-negative integers j1 , j2 , . . . , jn satisfying ∑t=1 t jt = n.
Hence, asymptotically, the random variables Xn,t have independent Poisson
distributions with expectations 1/t, respectively (see Goncharov [387] and
Kolchin [505]).
Next, consider the random variable Xn = ∑nj=1 Xn, j counting the total number
of cycles in a functional digraph D f of a random permutation. It is not difficult
to show that Xn has the following probability distribution.
 16.1 Permutations 335

Theorem 16.1 For k = 1, 2, . . . , n,

|s(n, k)|
P(Xn = k) = ,
n!
where the s(n, k) are Stirling numbers of the first kind, i.e., numbers satisfying
the following relation:
n
x(x − 1) · · · (x − n + 1) = ∑ s(n, k)xk .
k=0

Moreover,
n n
1 1
E Xn = Hn = ∑ , Var Xn = Hn − ∑ .
j=1 j j=1 j2

Proof Denote by c(n, k) the number of digraphs D f (permutations) on n ver-

tices and with exactly k cycles. Consider a vertex n in D f . It either has a loop
(belongs to a unit cycle) or it doesn’t. If it does, then D f is composed of a
loop in n and a cyclic digraph (permutation) on n − 1 vertices with exactly
k − 1 cycles. and there are c(n − 1, k − 1) such digraphs (permutations). Other-
wise, the vertex n can be thought as dividing (lying on) one of the n − 1 arcs
which belongs to cyclic digraph on n − 1 vertices with k cycles and there are
(n − 1)c(n − 1, k) such permutations (digraphs) of the set [n]. Hence

c(n, k) = c(n − 1, k − 1) + (n − 1)c(n − 1, k) .

Now, multiplying both sides by xk , dividing by n! and summing up over all k,

we get
Gn (x) = (x + n − 1)Gn−1 (x).

where Gn (x) is the probability generating function of Xn . But G1 (x) = x, so

x(x + 1) · · · (x + n − 1)
Gn (x) = ,
n!
and the first part of the theorem follows. Note that
 
x+n−1 Γ(x + n)
Gn (x) = = ,
n Γ(x)Γ(n + 1)
where Γ is the Gamma function.

The results for the expectation and variance of Xn can be obtained by calculat-
ing the first two derivatives of Gn (x) and evaluating them at x = 1 in a standard
336 Mappings

way but one can also show them using only the fact that the cycles of functional
digraphs must be disjoint. Notice, for example, that
E Xn = ∑ P(S induces a cycle)
06/ =S⊂[n]
n  
n (k − 1)!(n − k)!
= ∑ = Hn .
k=1 k n!
Similarly one can derive the second factorial moment of Xn counting ordered
pairs of cycles (see Exercises 16.3.2 and 16.3.3) which implies the formula for
the variance.
Goncharov [387] proved a Central Limit Theorem for the number Xn of cy-
cles.
Theorem 16.2
  Zx
Xn − log n 2
lim P √ ≤x = e−t /2 dt,
n→∞ log n −∞

i.e., the standardized random variable Xn converges in distribution to the stan-

dard Normal random variable.
Another numerical characteristic of a digraph D f is the length Ln of its longest
cycle. Shepp and Lloyd [703] established the asymptotic behavior of the ex-
pected value of Ln .

Theorem 16.3
 
E Ln 1 −y
Z ∞ Z ∞
lim = exp −x − e dy dx = 0.62432965....
n→∞ n 0 x y

16.2 Mappings
Let f be chosen uniformly at random from the set of all nn mappings from
[n] → [n]. Let D f be the functional digraph ([n], (i, f (i))) and let G f be the
graph obtained from D f by ignoring orientation. In general, D f has unicyclic
components only, where each component consists of a directed cycle C with
trees rooted at vertices of C, see the Figure 16.2.
Therefore the study of functional digraphs is based on results for permutations
of the set of cyclical vertices (these lying on cycles) and results for forests
consisting of trees rooted at these cyclical vertices (we allow also trivial one
vertex trees). For example, to show our first result on the connectivity of G f
we will need the following enumerative result for the forests.
 16.2 Mappings 337

x 1 2 3 4 5 6 7 8 9 10

f(x) 3 10 5 4 2 5 8 9 7 5

10
11
00
711
00
8
00
11
11
00
00
11
2 11
00 00
11 11
00 00
11
00
11 00
11 00
11 00
11
11
00
00
11
00
11
4
00
11 11
00
00
11
1111
000000
11 00
11
000011
1111
0000
1111
00
00
11
00
11 5 9
0000
1111 00
11
00
11
0000
1111
0000 11
1111 00
000
111
0000
1111 00
11
00
11
000
111
000
111 00
11
00
11
0000
1111 300
11
0000
1111
0000
1111
0000
1111
00
11 6
0000
1111
0000
1111
0000
1111
00
11
00
11
00
11
1

Figure 16.2 A mapping digraph example

Lemma 16.4 Let T (n, k) denote the number of forests with vertex set [n],
consisting of k trees rooted at the vertices 1, 2, . . . , k. Then,
T (n, k) = knn−k−1 .
Proof Observe first that by (15.2) there are n−1

n−k
k−1 n trees with n+1 labelled
vertices in which the degree of a vertex n + 1 is equal to k. Hence there are
   
n − 1 n−k . n
n = knn−k−1
k−1 k
trees with n+1 labeled vertices in which the set of neighbors of the vertex n+1
is exactly [k]. An obvious bijection (obtained by removing the vertex n + 1
from the tree) between such trees and the considered forests leads directly to
the lemma.

Theorem 16.5
1 n (n)k
r
π
P(G f is connected ) = ∑ k ≈ .
n k=1 n 2n
338 Mappings

Proof If G f is connected then there is a cycle with k vertices say such that
after removing the cycle we have a forest consisting of k trees rooted at the
vertices of the cycle. Hence,
n  
n
P(G f is connected ) = n−n ∑ (k − 1)! T (n, k)
k=1 k

1 n (n)k 1 n k−1
 
j
= ∑ k = ∑ ∏ 1−
n k=1 n n k=1 j=0 n
1 n
= ∑ uk .
n k=1

If k ≥ n3/5 , then
   
k(k − 1) 1 1/6
uk ≤ exp − ≤ exp − n ,
2n 3

while, if k < n3/5 ,

k2
  3 
k
uk = exp − + O 2 .
2n n
So
3/5
1 + o(1) n 2
 1/6

P(G f is connected ) = ∑ e−k /2n + O ne−n /3
n k=1
1 + o(1)
Z ∞  
2 1/6
= e−x /2n dx + O ne−n /3
n 0
1 + o(1) ∞ −y2 /2
Z  
1/6
= √ e dy + O ne−n /3
n 0
r
π
≈ .
2n

Let Zk denote the number of cycles of length k in a random mapping. Then

1 k−1
   
n −k j uk
E Zk = (k − 1)! n = ∏ 1 − = .
k k j=0 n k

If Z = Z1 + Z2 + · · · + Zn , then
n
uk 1 −x2 /2n 1 ∞ 1 −s/2n 1
Z ∞ Z
EZ = ∑ ≈ e dx = e ds ≈ log n.
k=1 k x=1 x 2 s=1 s 2
1 R ∞ 1 −s/2n
(To estimate the integral 2 s=1 s e we break it into I1 + I2 + I3 where I1 =
 16.2 Mappings 339
R n/ω R ωn  
ωn
s=1 · · · ds ≈ log n, ω = log n, I2 = s=n/ω · · · ds ≤ log n/ω = o(log n) and
R∞
I3 = s=ωn · · · ds = o(1).)
Moreover the expected number of vertices of cycles in a random mapping is
equal to
!
n n
r
πn
E ∑ kZk = ∑ uk ≈ .
k=1 k=1 2

Note that the functional digraph of a random mapping can be interpreted as a

representation of a process in which vertex i ∈ [n] chooses its image indepen-
dently with probability 1/n. So, it is natural to consider a general model of a
random mapping fˆ : [n] → [n] where, independently for all i ∈ [n],

P fˆ(i) = j = p j , j = 1, 2, . . . , n,


(16.2)

and
p1 + p2 + . . . + pn = 1.

This model was introduced (in a slightly more general form) independently by
Burtin [170] and Ross [674]. We will first prove a generalisation of Theorem
16.5.

Theorem 16.6

P(G fˆ is connected ) =
!
=∑ p2i 1+ ∑ pj + ∑ ∑ p j pk + ∑ ∑ ∑ p j pk pl + · · · .
i j6=i j6=i k6=i, j j6=i k6=i, j l6=i, j,k

To prove this theorem we use the powerful “Burtin–Ross Lemma”. The short
and elegant proof of this lemma given here is due to Jaworski [448] (His gen-
eral approach can be applied to study other characteristics of a random map-
pings, not only their connectedness).

Lemma 16.7 (Burtin–Ross Lemma) Let fˆ be a generalized random mapping

defined above and let G fˆ[U] be the subgraph of G fˆ induced by U ⊂ [n]. Then

P(G fˆ[U] does not contain a cycle) = ∑ pk .

k∈[n]\U
340 Mappings

Proof The proof is by induction on r = |U|. For r = 0 and r = 1 it is obvious.

Assume that the result holds for all values less than r, r ≥ 2. Let 0/ 6= S ⊂ U and
denote by A the event that G fˆ[S] is the union of disjoint cycles and by B the
event that G fˆ[U \ S] does not contain a cycle . Notice that events A and B are
independent, since the first one depends on choices of vertices from S, only,
while the second depends on choices of vertices from U \ S. Hence
P(G fˆ[U] contains a cycle ) = ∑ P(A ) P(B).
06/ =S⊂U

But if A holds then fˆ restricted to S defines a permutation on S. So,

P(A ) = |S|! ∏ p j .
j∈S

Since |U \ S| < r, by the induction assumption we obtain

P(G fˆ[U] contains a cycle ) =
= ∑ |S|! ∏ p j ∑ pk
06/ =S⊂U j∈S k∈[n]\(U\S)
!
= ∑ |S|! ∏ p j 1 − ∑ pk
06/ =S⊂U j∈S k∈(U\S)

= ∑ |S|! ∏ pk − ∑ |S|! ∏ pk
S⊂U, |S|≥1 k∈S S⊂U, |S|≥2 k∈S

= ∑ pk ,
k∈U

completing the induction.

Before we prove Theorem 16.6 we will point out that Lemma 16.4 can be
immediately derived from the above result. To see this, in Lemma 16.7 choose
p j = 1/n, for j = 1, 2, · · · n, and U such that |U| = r = n − k. Then, on one
hand,
1 k
P(G f [U] does not contain a cycle) = ∑ = .
i∈[n]\U
n n

On the other hand,

T (n, k)
P(G f [U] does not contain a cycle) = ,
nn−k
where T (n, k) is the number of forests on [n] with k trees rooted in vertices
from the set [n] \ U. Comparing both sides we immediately get the result of
Lemma 16.4, i.e., that
T (n, k) = knn−k−1 .
 16.2 Mappings 341

Proof (of Theorem 16.6). Notice that G fˆ is connected if and only if there is
a subset U ⊆ [n] such that U spans a single cycle while there is no cycle on
[n] \U. Moreover, the events “U ⊆ [n] spans a cycle” and “there is no cycle on
[n] \U” are independent. Hence, by Lemma 16.7,
Pr(G fˆ is connected) =
= ∑ P(U ⊂ [n] spans a cycle) P(there is no cycle on [n] \U)
06/ =U⊆[n]

= ∑ (|U| − 1)! ∏ p j ∑ pk (16.3)

06/ =U⊂[n] j∈U k∈U
!
=∑ p2i 1+ ∑ pj + ∑ ∑ p j pk + ∑ ∑ ∑ p j pk pl + · · · .
i j6=i j6=i k6=i, j j6=i k6=i, j l6=i, j,k

Using the same reasoning as in the above proof, one can show the following
result due to Jaworski [448].
Theorem 16.8 Let X be the number of components in G fˆ and Y be the num-
ber of its cyclic vertices (vertices belonging to a cycle). Then for k = 1, 2, . . . , n,
P(X = k) = ∑ ∏ p j |s(|U|, k)| − ∑ ∏ p j |s(|U| − 1, k)||U|,
U⊂[n] j∈U U⊂[n] j∈U
|U|≥k |U|≥k+1

where s(·, ·) is the Stirling number of the first kind. On the other hand,
P(Y = k) = k! ∑ ∏ p j − (k + 1)! ∑ ∏ p j .
U⊂[n] j∈U U⊂[n] j∈U
|U|=k |U|=k+1

The Burtin–Ross Lemma has another formulation which we present below.

Lemma 16.9 (Burtin-Ross Lemma - the second version) Let ĝ : [n] → [n] ∪
{0} be a random mapping from the set [n] to the set [n] ∪ {0}, where, indepen-
dently for all i ∈ [n],
P (ĝ(i) = j) = q j , j = 0, 1, 2, . . . , n,
and
q0 + q1 + q2 + . . . + qn = 1.
Let Dĝ be the random directed graph on the vertex set [n] ∪ {0}, generated by
the mapping ĝ and let Gĝ denote its underlying simple graph. Then
P(Gĝ is connected ) = q0 .
342 Mappings

Notice that the event that Gĝ is connected is equivalent to the event that Dĝ is
a (directed) tree, rooted at vertex {0}, i.e., there are no cycles in Gĝ [[n]].
We will use this result and Lemma 16.9 to prove the next theorem (for more
general results, see [449]).

Theorem 16.10 Let D fˆ be the functional digraph of a mapping fˆ defined in

(16.2) and let ZR be the number of predecessors of a set R ⊂ [n], |R| = r, r ≥ 1,
of vertices of D fˆ , i.e.,

ZR = |{ j ∈ [n] : for some non-negative integer k, fˆ(k) ( j) ∈ R}|,

where fˆ(0) ( j) = j and for k ≥ 1, fˆ(k) ( j) = fˆ( fˆ(k−1) ( j)).

Then, for k = 0, 1, 2, . . . , n − r,

P(ZR = k + r) = ΣR ∑ (ΣU∪R )k−1 (1 − ΣU∪R )n−k ,

U⊂[n]\R
|U|=k

where for A ⊆ [n], ΣA = ∑ j∈A p j .

Proof The distribution of ZR follows immediately from the next observation

and the application of Lemma 16.9. Denote by A the event that there is a forest
spanned on the set W = U ∪R, where U ⊂ [n]\R, composed of r (directed) trees
rooted at vertices of R. Then

P(ZR = k + r) = ∑ P(A |B ∩ C ) P(B) P(C ), (16.4)

U⊂[n]\R
|U|=k

where B is the event that all edges that begin in U end in W , while C denotes
the event that all edges that begin in [n] \W end in [n] \W . Now notice that

P(B) = (ΣW )k , while P(C ) = (1 − ΣW )n−k .

Furthermore,
P(A |B ∩C) = P(Gĝ is connected ),

where ĝ : U → U ∪ {0}, where {0} stands for the set R collapsed to a single
vertex, is such that for all u ∈ U independently,
pj ΣR
q j = P(ĝ(u) = j) = , for j ∈ U, while q0 = .
ΣW ΣW
So, applying Lemma 16.9, we arrive at the thesis.

We will finish this section by stating the central limit theorem for the number
of components of G f , where f is a uniform random mapping f : [n] → [n]
 16.3 Exercises 343

(see Stepanov [717]). It is an analogous result to Theorem 16.2 for random

permutations.

Theorem 16.11
 
Xn − 21 log n
Z x
2
lim P  q ≤ x = e−t /2 dt,
n→∞ 1 −∞
2 log n

the standardized random variable Xn converges in distribution to the standard

Normal random variable.

16.3 Exercises
16.3.1 Prove directly that if Xn,t is the number of cycles of length t in a random
permutation then E Xn,t = 1/t.
16.3.2 Find the expectation and the variance of the number Xn of cycles in
a random permutation using fact that the rth derivative of the gamma
dr R∞ r x−1 e−t dt,
function equals (dx) r Γ(x) = 0 (logt) t

16.3.3 Determine the variance of the number Xn of cycles in a random per-

mutation (start with computation of the second factorial moment of Xn ,
counting ordered pairs of cycles).
16.3.4 Find the probability distribution for the length of a typical cycle in a
random permutation, i.e., the cycle that contains a given vertex (say
vertex 1). Determine the expectation and variance of this characteristic.
16.3.5 Find the probability distribution of the number of components in a
functional digraph D f of a uniform random mapping f : [n] → [n].
16.3.6 Show that the length of the cycle containing item 1 in a random permu-
tation is uniformly distributed in [n].
16.3.7 Show that if X denotes the number of cycles in a random permuta-
tion of [n] then P(X ≥ t) ≤ P(Bin(t, 1/2) ≤ dlog2 ne). Deduce that for
every constant K > 0, there exists a constant L > 0, such that P(X ≥
K log n) ≤ n−L .
16.3.8 Now let X denote the number of cycles in the digraph D f of a random
mapping f : [n] → [n]. Show that for every constant K > 0, there exists
a constant L > 0, such that P(X ≥ K log n) ≤ n−L .
 344 Mappings

16.3.9 Determine the expectation and variance of the number of components

in a functional digraph D fˆ of a generalized random mapping fˆ (see
Theorem 16.8)
16.3.10 Find the expectation and variance of the number of cyclic vertices in a
functional digraph D fˆ of a generalized random mapping fˆ (see Theo-
rem 16.8).
16.3.11 Prove Theorem 16.8.
16.3.12 Show that Lemmas 16.7 and 16.9 are equivalent.
16.3.13 Prove the Burtin-Ross Lemma for a bipartite random mapping, i.e. a
mapping with bipartition ([n], [m]), where each vertex i ∈ [n] chooses its
unique image in [m] independently with probability 1/m, and, similarly,
each vertex j ∈ [m] selects its image in [n] with probability 1/n.
16.3.14 Consider an evolutionary model of a random mapping (see [450],[451]),
i.e., a mapping fˆq [n] → [n], such that for i, j ∈ [n], P( fˆq (i) = j) = q if
i = j while, P( fˆq (i) = j) = (1 − q)/(n − 1) if i 6= j, where 0 ≤ q ≤ 1.
Find the probability that fˆq is connected.
16.3.15 Show that there is one-to-one correspondence between the family of nn
mappings f : [n] → [n] and the family of all doubly-rooted trees on the
vertex set [n] (Joyal bijection)

16.4 Notes
Permutations
Systematic studies of the properties of random permutations of n objects were
initiated by Goncharov in [386] and [387]. Golomb [385] showed that the ex-
pected length of the longest cycle of D f , divided by n is monotone decreas-
ing and gave a numerical value for the limit, while Shepp and Lloyd in [703]
found the closed form for this limit (see Theorem 16.3). They also gave the
corresponding result for kth moment of the rth longest cycle, for k, r = 1, 2, . . .
and showed the limiting distribution for the length of the rth longest cycle.
Kingman [495] and, independently, Vershik and Schmidt [736], proved that
for a random permutation of n objects, as n → ∞, the process giving the pro-
portion of elements in the longest cycle, the second longest cycle, and so on,
converges in distribution to the Poisson-Dirichlet process with parameter 1 (for
further results in this direction see Arratia, Barbour and Tavaré [40]). Arratia
and Tavaré [41] provide explicit bounds on the total variation distance between
the process which counts the sizes of cycles in a random permutations and a
process of independent Poisson random variables.
 16.4 Notes 345

For other results, not necessarily of a “graphical” nature, such as, for example,
the order of a random permutation, the number of derangements, or the number
of monotone sub-sequences, we refer the reader to the respective sections of
books by Feller [294], Bollobás [133] and Sachkov [688] or, in the case of
monotone sub-sequences, to a recent monograph by Romik [673].

Mappings
Uniform random mappings were introduced in the mid 1950’s by Rubin and
Sitgraves [675], Katz [489] and by Folkert [312]. More recently, much atten-
tion has been focused on their usefulness as a model for epidemic processes,
see for example the papers of Gertsbakh [372], Ball, Mollison and Scalia-
Tomba [54], Berg [79], Mutafchiev [612], Pittel [640] and Jaworski [451]. The
component structure of a random functional digraph D f has been studied by
Aldous [12]. He has shown, that the joint distribution of the normalized order
statistics for the component sizes of D f converges to the Poisson-Dirichlet dis-
tribution with parameter 1/2. For more results on uniform random mappings
we refer the reader to Kolchin’s monograph [507], or a chapter of Bollobás’
[133].
The general model of a random mapping fˆ, introduced by Burtin [170] and
Ross [674], has been intensively studied by many authors. The crucial Burtin-
Ross Lemma (see Lemmas: 16.7 and 16.9) has many alternative proofs (see
[37]) but the most useful seems to be the one used in this chapter, due to Ja-
worski [448]. His approach can also be applied to derive the distribution of
many other characteristics of a random digraph D f , as well as it can be used to
prove generalisations of the Burtin-Ross Lemma for models of random map-
pings with independent choices of images. (For an extensive review of results
in that direction see [449]). Aldous, Miermont, Pitman ([17],[18]) study the
asymptotic structure of D fˆ using an ingenious coding of the random mapping
fˆ as a stochastic process on the interval [0, 1] (see also the related work of Pit-
man [639], exploring the relationship between random mappings and random
forests).
Hansen and Jaworski (see [402], [403]) introduce a random mapping f D : [n] →
[n] with an in-degree sequence, which is a collection of exchangeable random
variables (D1 , D2 , . . . , Dn ). In particular, they study predecessors and succes-
sors of a given set of vertices, and apply their results to random mappings with
preferential and anti-preferential attachment.
 17
k-out

Several interesting graph properties require that the minimum degree of a graph
be at least a certain amount. E.g. having a Hamilton cycle requires that the min-
imum degree is at least two. In Chapter 6 we saw that Gn,m being Hamiltonian
and having minimum degree at least two happen at the same time w.h.p. One
is therefore interested in models of a random graph which guarantee a certain
minimum degree. We have already seen d-regular graphs in Chapter 11. In this
chapter we consider another simple and quite natural model Gk−out that gen-
eralises random mappings. It seems to have first appeared in print as Problem
38 of “The Scottish Book” [570]. We discuss the connectivity of this model
and then matchings and Hamilton cycles. We also consider a related model of
“Nearest Neighbor Graphs”.

17.1 Connectivity
For an integer k, 1 ≤ k ≤ n − 1, let ~Gk−out be a random digraph on vertex
set V = {1, 2, . . . , n} with arcs (directed edges) generated independently for
each v ∈ V by a random
choice of k distinct arcs (v, w), where w ∈ V \ {v}, so
that each of the n−1 k possible sets of arcs is equally likely to be chosen. Let
Gk−out be the random graph(multigraph) obtained from ~Gk−out by ignoring the
orientation of its arcs, but retaining all edges.
Note that ~G1−out is a functional digraph of a random mapping f : [n] → [n],
with a restriction that loops (fixed points) are not allowed. So for k = 1 the
following result holds.

Theorem 17.1
lim P(G1−out is connected ) = 0.
n→∞

The situation changes when each vertex is allowed to choose more than one
neighbor. Denote by κ(G) and λ (G) the vertex and edge connectivity of a
graph G respectively, i.e., the minimum number of vertices (respectively edges)

346
 17.1 Connectivity 347

the deletion of which disconnects G. Let δ (G) be the minimum degree of G.

The well known Whitney’s Theorem states that, for any graph G,
κ(G) ≤ λ (G) ≤ δ (G).
In the next theorem we show that for random k − out graphs these parameters
are equal w.h.p. It is taken from Fenner and Frieze [302]. The Scottish Book
[570] contains a proof that Gk−out is connected for k ≥ 2.

Theorem 17.2 Let κ = κ(Gk−out ), λ = λ (Gk−out ) and δ = δ (Gk−out ). Then,

for 2 ≤ k = O(1),
lim P(κ = λ = δ = k) = 1.
n→∞

Proof In the light of Whitney’s Theorem, to prove our theorem we have to

show that the following two statements hold:
lim P(κ(Gk−out ) ≥ k) = 1, (17.1)
n→∞

and
lim P(δ (Gk−out ) ≤ k) = 1. (17.2)
n→∞

Then, w.h.p.
k ≤ κ ≤ λ ≤ δ ≤ k,
and the theorem follows.
To prove statement (17.1) consider the deletion of r vertices from the random
graph Gk−out , where 1 ≤ r ≤ k − 1. If Gk−out can be disconnected by deleting r
vertices, then there exists a partition (R, S, T ) of the vertex set V , with | R |= r,
| S |= s and | T |= t = n − r − s, with k − r + 1 ≤ s ≤ n − k − 1, such that Gk−out
has no edge joining a vertex in S with a vertex in T . The probability of such an
event, for an arbitrary partition given above, is equal to
r+s−1
s n−s−1
n−r−s
! !
r + s sk n − s (n−r−s)k
   
k k
n−1
n−1

≤
k k
n n

Thus
b(n−r)/2c  sk  (n−r−s)k
n! r+s n−s
P(κ(Gk−out ) ≤ r) ≤ ∑
s=k−r+1 s!r!(n − r − s)! n n

We have replaced n − k − 1 by b(n − r)/2c because we can always interchange

S and T so that |S| ≤ |T |.
348 k-out

But, by Stirling’s formula,

n! nn
≤ αs s
s!r!(n − r − s)! s (n − r − s)n−r−s
where
 1/2
n 2c
αs = α(s, n, r) ≤ c ≤ ,
s(n − r − s) s1/2
for some absolute constant c > 0.
Thus
b(n−r)/2c  s  (n−r−s)
1 r+s n−s
P(κ(Gk−out ) ≤ r) ≤ 2c ∑ us
s=k−r+1 s1/2 s n−r−s

where
us = (r + s)(k−1)s (n − s)(k−1)(n−r−s) nn−k(n−r) .

Now,
 s  n−r−s
r+s n−s
≤ e2r ,
s n−r−s
and
(r + s)s (n − s)n−r−s

decreases monotonically, with increasing s, for s ≤ (n − r)/2. Furthermore, if

s ≤ n/4 then the decrease is by a factor of at least 2.
Therefore
!
n/4
2r −(k−1)(s−k+r−1) 2 n
P(κ(Gk−out ) ≤ r) ≤ 2ce ∑ 2 + 1/2 · uk−r+1
s=k−r+1 n 4
≤ 5ce2r n1/2 uk−r+1 ≤ 5ce2r an3/2−k(k−r) ,

where
a = (k + 1)(k−1)(k−r+1) .

It follows that

lim P(κ(Gk−out ) ≤ r) = lim P(κ(Gk−out ) ≤ k − 1) = 0,

n→∞ n→∞

which implies that

lim P(κ(Gk−out ) ≥ k) = 1,
n→∞
 17.2 Perfect Matchings 349

i.e., that equation (17.1) holds.

To complete the proof we have to show that equation (17.2) holds, i.e., that
P(δ (Gk−out ) = k) → 1 as n → ∞.
Since δ ≥ k in Gk−out , we have to show that w.h.p. there is a vertex of degree
k in Gk−out .
Let Ev be the event that vertex v has indegree zero in ~Gk−out . Thus the degree
of v in Gk−out is k if and only if Ev occurs. Now
n−2
n−1
!  n−1
k
P(Ev ) = k
n−1

= 1− → e−k .
k
n − 1

Let Z denote the number of vertices of degree k in Gk−out . Then we have shown
that E(Z) ≈ ne−k . Now the random variable Z is determined by kn independent
random choices. Changing one of these choices can change the value of Z by
at most one. Applying the Azuma-Hoeffding concentration inequality – see
Section 22.7, in particular Lemma 22.17 we see that for any t > 0
2t 2
 
P(Z ≤ E(Z) − t) ≤ exp − .
kn
Putting t = ne−k /2 we see that Z > 0 w.h.p. and the theorem follows.

17.2 Perfect Matchings

Non-bipartite graphs
Assuming that the number of vertices n of a random graph Gk−out is even,
Frieze [328] proved the following result.

Theorem 17.3
(
0 if k = 1
lim P(Gk−out has a perfect matching) =
n→∞
n even 1 if k ≥ 2.

We will only prove a weakening of the above result to where k ≥ 15. We follow
the ideas of Section 6.1. So, we begin by examining the expansion properties
of G = Ga−out , a ≥ 3.
Lemma 17.4 W.h.p. |NG (S)| ≥ |S| for all S ⊆ [n], |S| ≤ κa n where κa =
1 1 1/(a−2)


2 30 .
350 k-out

Proof The probability there exists a set S with insufficient expansion is at

most
κa n     as κa n    as
n n 2s ne 2s 2s
∑ s s−1 n ≤ ∑ s n
s=3 s=3
κa n  a−2
 s
s
=∑ e2 2a = o(1). (17.3)
s=3 n

Lemma 17.5 Let b = (1 + κa−2 )/2 . Then as n → ∞, n even, G(a+b)−out has

 

a perfect matching w.h.p.

Proof First note that G(a+b)−out contains H = Ga−out ∪ Gb−out in the follow-
ing sense. Start the construction of G(a+b)−out with H. If there is a v ∈ [n] that
chooses edge {v, w} in both Ga−out and Gb−out then add another random choice
for v.
Let us show that H has a perfect matching w.h.p. Enumerate the edges of
Gb−out as e1 , e2 , . . . , ebn . Here e(i−1)n+ j is the ith edge chosen by vertex j. Let
G0 = Ga−out and let Gi = G0 + {e1 , e2 , . . . , ei }. If Gi does not have a per-
fect matching, consider the sets A, A(x), x ∈ A defined prior to (6.6). It follows
from Lemma 17.4 that w.h.p. all of these sets are of size at least κa n. Thus,
P(y ∈ A(x)) ≥ κa n−b n . We subtract b to account for the previously inspected
edges associated with x’s choices.
It follows that

P(G(a+b)−out does not have a perfect matching)

≤ P(H does not have a perfect matching)
≤ P(Bin(bκa n, κa − b/n) ≤ n/2) = o(1).

Putting a = 8 gives b = 7 and a proof that G15−out , n even, has a perfect match-
ing w.h.p.

Bipartite graphs
We now consider the related problem of the existence of a perfect matching in
a random k-out bipartite graph.
Let U = {u1 , u2 , . . . , un },V = {v1 , v2 , . . . , vn } and let each vertex from U choose
independently and without repetition, k neighbors in V , and let each vertex
from V choose independently and without repetition k neighbors in U. Denote
 17.2 Perfect Matchings 351

by ~Bk−out the digraphs generated by the above procedure and let Bk−out be its
underlying simple bipartite graph.

Theorem 17.6
(
0 if k = 1
lim P(Bk−out has a perfect matching) =
n→∞ 1 if k ≥ 2.

We will give two different proofs. The first one - existential- of a combinatorial
nature is due to Walkup [741]. The second one - constructive- of an algorith-
mic nature, is due to Karp, Rinnooy-Kan and Vohra [485]. We start with the
combinatorial approach.

Existence proof
Let X denote the number of perfect matchings in Bk−out . Then
P(X > 0) ≤ E(X) ≤ n! 2n (k/n)n .
The above bound follows from the following observations. There are n! ways
of pairing the vertices of U with the vertices of V . For each such pairing there
are 2n ways to assign directions for the connecting edges, and then each possi-
ble matching has probability (k/n)n of appearing in Bk−out .
So, by Stirling’s formula,
P(X > 0) ≤ 3n1/2 (2k/e)n ,
which, for k = 1 tends to 0 as n → ∞, and the first statement of our theorem
follows.
To show that Bk−out has a perfect matching w.h.p. notice that since this is an
increasing graph property, it is enough to show that it is true for k = 2. Note
also, that if there is no perfect matching in Bk−out , then there must exist a set
R ⊂ U (or R ⊂ V ) such that the cardinality of neighborhood of S = N(R) of
R in U ( respectively, in V ) is smaller than the cardinality of the set R itself,
i.e., | S |<| R |. We will call such a pair (R, S) a bad pair, and, in particular, we
will restrict our attention to the “minimal bad pairs”, i.e., such that there is no
R0 ⊂ R for which (R0 , N(R0 )) is bad.
If (R, S) is a bad pair with R ⊆ U then (V \ S,U \ R) is also a bad pair. Given
this, we can concentrate on showing that w.h.p. there are no bad pairs (R, S)
with 2 ≤ |R| ≤ (n + 1)/2.
Every minimal bad pair has to have the following two properties:
 352 k-out

(i) | S |=| R | −1,

(ii) every vertex in S has at least two neighbors in R.
The first property is obvious. To see why property (ii) holds, suppose that there
is a vertex v ∈ S with at most one neighbor u in R. Then the pair (R \ {u} , S \
{v}) is also “bad pair” and so the pair (R, S) is not minimal.
Let r ∈ [2, (n + 1)/2] and let Yr be the number of minimal bad pairs (R, S),
with | R |= r in Bk−out . To complete the proof of the theorem we have to
show that ∑r EYr → 0 as n → ∞. By symmetry, choose (R, S), such that R =
{u1 , u2 , . . . ur } ⊂ U and S = {v1 , v2 , . . . vr−1 } ⊂ V is a minimal “bad pair”. Then
  
n n
EYr = 2 Pr Qr , (17.4)
r r−1
where
Pr = P((R, S) is bad)
and
Qr = P((R, S) is minimal | (R, S) is bad).
We observe that, for any fixed k,
!r !n−r+1
r−1
n−r

k k
Pr = n
n
.
k k

Hence, for k = 2,
 r 2r  n − r 2(n−r)
Pr ≤ . (17.5)
n n
Then we use Stirling’s formula to show,
    2
n n r n
=
r r−1 n−r+1 r
 2(n−r)
r n  n 2r n
≤ . (17.6)
n − r + 1 r(n − r) r n−r
To estimate Qr we have to consider condition (ii) which a minimal bad pair
has to satisfy. This implies that a vertex v ∈ S = N(R) is chosen by at least one
vertex from R (denote this event by Av ), or it chooses both its neighbors in R
(denote this event by Bv ). Then the events Av , v ∈ S are negatively correlated
(see Section 22.2) and the events Bv , v ∈ S are independent of other events in
this collection. Let S = {v1 , v2 , . . . , vr−1 }. Then we can write
!
r−1
\
Qr ≤ P (Avi ∪ Bvi )
i=1
 17.2 Perfect Matchings 353
 i−1 !
r−1 \
= ∏ P Avi ∪ Bvi  (Av j ∪ Bv j )
i=1 j=1
r−1
≤ ∏ P (Avi ∪ Bvi )
i=1

r−1
= 1 − P(Acv1 ) P(Bcv1 )
!!r−1
r

r − 2 2r
 
≤ 1− 1 − 2n

r−1 2
≤ η r−1 (17.7)

for some absolute constant 0 < η < 1 when r ≤ (n + 1)/2.

Going back to (17.4), and using (17.5), (17.6), (17.7)
(n+1)/2 (n+1)/2
η r−1 n
∑ E Xr ≤ 2 ∑ = o(1).
r=2 r=2 (n − r)(n − r + 1)

Hence ∑r E Xr → 0 as n → ∞, which means that w.h.p. there are no bad pairs,

implying that Bk−out has a perfect matching w.h.p.
Frieze and Melsted [350] considered the related question. Suppose that M, N
are disjoint sets of size m, n and that each v ∈ M chooses d ≥ 3 neighbors
in N. Suppose that we condition on each vertex in N being chosen at least
twice. They show that w.h.p. there is a matching of size equal to min {m, n}.
Fountoulakis and Panagiotou [315] proved a slightly weaker result, in the same
vein.

Algorithmic Proof
We will now give a rather elegant algorithmic proof of Theorem 17.6. It is
due to Karp, Rinnooy-Kan and Vohra [485]. We do this for two reasons. First,
because it is a lovely proof and second this proof is the basis of the proof that
2-in,2-out is Hamiltonian in [211]. In particular, this latter example shows that
constructive proofs can sometimes be used to achieve results not obtainable
through existence proofs alone.
Start with the random digraph ~B2−out and consider two multigraphs, GU and
GV with labeled vertices and edges, generated by ~B2−out on the sets of the
bipartition (U,V ) in the following way. The vertex set of the graph GU is U
and two vertices, u and u0 , are connected by an edge, labeled v, if a vertex v ∈ V
chooses u and u0 as its two neighbors in U. Similarly, the graph GV has vertex
set V and we put an edge labeled u between two vertices v and v0 , if a vertex
354 k-out

u ∈ U chooses v and v0 as its two neighbors in V . Hence graphs GU and GV are

random multigraphs with exactly n labeled vertices and n labeled edges.
We will describe below, a randomized algorithm which w.h.p. finds a perfect
matching in B2−out in O(n) expected number of steps.
Algorithm PAIR

• Step 0. Set HU = GU and let HV be empty graph on vertex set V . Initially

all vertices in HU are unmarked and all vertices in GV are unchecked. Let
CORE denote the set of edges of GU that lie on cycles in GU i.e. the edges
of the 2-core of GU .
• Step 1. If every isolated tree in HU contains a marked vertex, go to Step
5. Otherwise, select any isolated tree T in HU in which all vertices are un-
marked. Pick a random vertex u in T and mark it.
• Step 2. Add the edge {x, y} , x, y ∈ V that has label u to the graph HV .
• Step 3. Let Cx ,Cy be the components of HV just before the edge labeled u is
added. Let C = Cx ∪ Cy . If all vertices in C are checked, go to Step 6. Oth-
erwise, select an unchecked vertex v in C. If possible, select an unchecked
vertex v for which the edge labeled v in HU belongs to CORE.
• Step 4. Delete the edge labeled v from HU , return to Step 1.
• Step 5. STOP and declare success.
• Step 6. STOP and declare failure.

We next argue that Algorithm PAIR, when it finishes at Step 5, does indeed
produce a perfect matching in B2−out . There are two simple invariants of this
process that explain this:

(I1) The number of marked vertices plus the number of edges in HU is equal
to n.
(I2) The number of checked vertices is equal to the number of edges in HV .

For I1, we observe that each round marks one vertex and deletes one edge of
HU . Similarly, for I2, we observe that each round checks one vertex and adds
one edge to HV .
Lemma 17.7 Up until (possible) failure in Step 6, the components of HV are
either trees with a unique unchecked vertex or are unicyclic components with
all vertices checked. Also, failure in Step 6 means that PAIR tries to add an
edge to a unicyclic component.
Proof This is true initially, as initially HV has no edges and all vertices are
unchecked. Assume this to be the case when we add an edge {x, y} to HV . If
 17.2 Perfect Matchings 355

Cx 6= Cy are both trees then we will have a choice of two unchecked vertices
in C = Cx ∪Cy and C will be a tree. After checking one vertex, our claim will
still hold. The other possibilities are that Cx is a tree and Cy is unicyclic. In
this case there is one unchecked vertex and this will be checked and C will be
unicyclic. The other possibility is that C = Cx = Cy is a tree. Again there is
only one unchecked vertex and adding {x, y} will make C unicyclic.
Lemma 17.8 If HU consists of trees and unicyclic components then all the
trees in HU contain a marked vertex.
Proof Suppose that HU contains k trees with marked vertices and ` trees with
no marked vertices and that the rest of the components are unicyclic. It follows
that HU contains n − k − ` edges and then (I1) implies that ` = 0.
Lemma 17.9 If the algorithm stops in Step 5, then we can extract a perfect
matching from HU , HV .
Proof Suppose that we arrive at Step 5 after k rounds. Suppose that there are
k trees with a marked vertex. Let the component sizes in HU be n1 , n2 , . . . , nk
for the trees and m1 , m2 , . . . , m` for the remaining components. Then,
n1 + n2 + · · · + nk + m1 + m2 + · · · + m` = |V (HU )| = n.
|E(HU )| = n − k,
from I1 and so
(n1 − 1) + (n2 − 1) + · · · + (nk − 1)+
(≥ m1 ) + (≥ m2 ) + (≥ m` ) = n − k.
It follows that the components of HU that are not trees with a marked vertex
have as many edges as vertices and so are unicyclic.
We now show, given that HU , HV only contain trees and unicyclic components,
that we can extract a perfect matching. The edges of HU define a matching
of B2−out of size n − k. Consider a tree T component with marked vertex ρ.
Orient the edges of T away from ρ. Now consider an edge {x, y} of T , oriented
from x to y. Suppose that this edge has label z ∈ V . We add the edge {y, z} to
M1 . These edges are disjoint: z appears as the label of exactly one edge and y
is the head of exactly one oriented edge.
For the unicyclic components, we orient the unique cycle
C = (u1 , u2 , . . . , us ) arbitrarily in one of two ways. We then consider the trees
attached to each of the ui and orient them away from the ui . An oriented edge
{x, y} with label z yields a matching edge {y, z} as before.
The remaining k edges needed for a perfect matching come from HV . We ex-
tract a set of k matching edges out of HV in the same way we extracted n − k
356 k-out

edges from HU . We only need to check that these k edges are disjoint from
those chosen from HU . Let {y, z} be such an edge, obtained from the edge
{x, y} of HV , which has label z. z is marked in HU and so is the root of a tree
and does not appear in any matching edge of M1 . y is a checked vertex and so
the edge labelled y has been deleted from HU and this prevents y appearing in
a matching edge of M1 .

Lemma 17.10 W.h.p. Algorithm PAIR cannot reach Step 6 in fewer than
0.49n iterations.

Proof It follows from Lemma 2.10 that w.h.p. after ≤ 0.499n rounds, HV
only contains trees and unicyclic components. The lemma now follows from
Lemma 17.7.
To complete our analysis, it only remains to show

Lemma 17.11 W.h.p., at most 0.49n rounds are needed to make HU the union
of trees and unicyclic components.

Proof Recall that each edge of HU corresponds to an unchecked vertex of HV ,

the edges corresponding to checked vertices having been deleted. Moreover,
each tree component T of HV has one unchecked vertex, uT say. If uT is the
label of an edge of HU belonging to CORE then due to the choice rule for
vertex checking in Step 3, every vertex of T must be the label of an edge of
CORE. Hence the number of edges left in CORE, after a given iteration of
the algorithm, is equal to the number of tree components of HV , where every
vertex labels an edge of CORE. We use this to estimate the number of edges
of CORE that remain in HU after .49n iterations.
Let xe−x = 2e−2 , where 0 < x < 1. One can easily check that

2 0.40 < x < 0.41.
It follows from Lemma 2.16 that w.h.p. |CORE| ≈ 1 − 2x n, which implies,
that 0.63 n ≤ |CORE| ≤ 0.64 n.
Let Z be the number of tree components in HV made up of vertices which are
the labels of edges belong to CORE. Then, after at most 0.49n rounds,
(log n)2 
(0.64)k
  
n k−2 0.49n
E Z ≤ o(1) + ∑ k (k − 1)!
k−1 ×
k=1 k k−1 n
2
!.49n−(k−1)
k(n − k)
× 1− n
(17.8)
2
(log n)2
kk−2
≤ (1 + o(1))n ∑ (0.64)k e−0.98k
k=1 k!
≤ (1 + o(1))n×
 17.2 Perfect Matchings 357
"
k #
(0.64θ )2 (0.64θ )3 2(0.64θ )4 (0.64)e.02
 ∞
0.64θ + + + +∑
2 2 3 k=5 2k5/2
where θ = e−0.98  
1
≤ (1 + o(1))n 0.279 +
2 × 55/2 (1 − (0.64)e.02 )
≤ (1 + o(1))n [0.279 + 0.026]
≤ (0.305)n.

Explanation of (17.8); The o(1) term corresponds to components of size greater

than (log n)2 and w.h.p. there are none of these. For the summand, we
choose k
vertices and a tree on these k vertices in nk kk−2 ways. The term 0.49n


k−1 (k − 1)!
gives the number of sequences of edge choices that lead to a given tree. The

−(k−1)
term n2 is the probability that these edges exist and (0.64)k bounds the
probability that the vertices of the tree correspond to edges in CORE. The final
term is the probability that the tree is actually a component.
So after 0.49n rounds, in expectation, the number of edges left in CORE, is
at most 0.305
0.63 < 0.485 of its original size, and the Chebyshev inequality (ap-
plied to Z) can be used to show that w.h.p. it is at most 0.49 of its original
size. However, randomly deleting approximately 0.51 fraction of the edges of
CORE will w.h.p. leave just trees and unicyclic components in HU . To see this,
observe that if we delete 0.505n random edges from GU then we will have a
random graph in the sub-critical stage and so w.h.p. it will consist of trees and
unicyclic components. But deleting 0.505n random edges will w.h.p. delete
less than a 0.51 fraction of CORE.
This completes the proof that w.h.p. Algorithm PAIR finishes before 0.49n
rounds with a perfect matching. In summary,

Theorem 17.12 W.h.p. the algorithm PAIR finds a perfect matching in the
random graph B2−out in at most .49n steps.

One can ask whether one can w.h.p. secure a perfect matching in a bipartite
random graph having more edges then B1−out , but less than B2−out . To see that
it is possible, consider the following two-round procedure. In the first round
assume that each vertex from the set U chooses exactly one neighbor in V and,
likewise, every vertex from the set V chooses exactly one neighbor in U. In the
next round, only those vertices from U and V which have not been selected in
the first round get a second chance to make yet another random selection. It
358 k-out

is easy to see that, for large n, such a second chance is, on the average, given
to approximately n/e vertices on each side. I.e, that the average out-degree of
vertices in U and V is approximately 1 + 1/e. Therefore the underlying simple
graph is denoted as B(1+1/e)−out , and Karoński and Pittel [478] proved that the
following result holds.

Theorem 17.13 With probability 1 − O(n−1/2 ) a random graph B(1+1/e)−out

contains a perfect matching.

17.3 Hamilton Cycles

Bohman and Frieze [115] proved the following:
Theorem 17.14
(
0 if k ≤ 2
lim P(Gk−out has a Hamiltonian Cycle) =
n→∞ 1 if k ≥ 3.

To see that this result is best possible note that one can show that w.h.p. the
random graph G2−out contains a vertex adjacent to three vertices of degree two,
which prevents the existence of a Hamiltonian Cycle. The proof that G3−out
w.h.p. contains a Hamiltonian Cycle is long and complicated, we will therefore
prove the weaker result given below which has a straightforward proof, using
the ideas of Section 6.2. It is taken from Frieze and Łuczak [345].

Theorem 17.15
lim P(Gk−out has a Hamiltonian Cycle) = 1, if k ≥ 5.
n→∞

Proof Let H = G0 ∪G1 ∪G2 where Gi = Gki −out , where (i) k0 = 1, k1 = k2 = 2

and (ii) G0 , G1 , G2 are generated independently of each other. Then we can
couple the construction of H and G5−out so that H ⊆ G5−out . This is because
in the construction of H, some random choices in the construction of the as-
sociated digraphs might be repeated. In which case, having constructed H, we
can give G5−out some more edges.
It follows from Theorem 17.3 that w.h.p. Gi , i = 1, 2 contain perfect matchings
Mi , i = 1, 2. Here we allow n to be odd and so a perfect matching may leave
one vertex isolated. By symmetry M1 , M2 are uniform random
 17.3 Hamilton Cycles 359

matchings. Let M = M1 ∪ M2 . The components of M are cycles. There could be

degenerate 2-cycles consisting of two copies of the same edge and there may
be a path in the case n is odd.

Lemma 17.16 Let X be the number of components of M. Then w.h.p.

X ≤ 3 log n.

Proof Let C be the cycle containing vertex 1. We show that

 n 1
P |C| ≥ ≥ . (17.9)
2 2
To see this note that
k−1  
n − 2i 1 1
P(|C| = 2k) = ∏ < .
i=1 n − 2i + 1 n − 2k + 1 n − 2k +1
Indeed, consider the M1 -edge {1 = i1 , i2 } ∈ C containing vertex 1. Let {i2 , i3 } ∈
C be the M2 -edge containing i2 , Now, P(i3 6= 1) = (n−2)/(n−1). Assume that
i3 6= 1 and let {i3 , i4 } ∈ C be the M1 -edge containing i3 . Let {i4 , i5 } ∈ C be the
M2 -edge containing i4 . Then P(i5 6= 1) = (n − 4)/(n − 3), and so on until we
close the cycle with probability 1/(n − 2k + 1). Hence
 n  bn/4c 1 1
P |C| < < ∑ < ,
2 k=1 n − 2k + 1 2
and the bound given in (17.9) follows.
Consider next the following experiment. Choose the size s of the cycle contain-
ing vertex 1. Next choose the size of the cycle containing a particular vertex
from the remaining n − s vertices. Continue until the cycle chosen contains
all remaining vertices. Observe now, that deleting any cycle from M leaves a
random pair of matchings of the remaining vertices. So, by this observation
and the fact that the bound (17.9) holds, whatever the currently chosen cycle
sizes, with probability at least 1/2, the size of the remaining vertex set halves,
at least. Thus,
P(X ≥ 3 log n) ≤ P(Bin(3 log n, 1/2) ≤ log2 n) = o(1).

We use rotations as in Section 6.2. Lemma 17.16 enables us to argue that we

only need to add random edges trying to find x, y where y ∈ END(x), at most
O(log n) times. We show next that H1 = G1 ∪ G2 has sufficient expansion.
360 k-out

Lemma 17.17 W.h.p. S ⊆ [n], |S| ≤ n/1000 implies that |NH1 (S)| ≥ 2|S|.

Proof Let X be the number of vertex sets that violate the claim. Then,
 !2 k
n/1000    3k

n n  2
EX ≤ ∑ n−1


k=1 k 2k 2
n/1000  3 3 k
e n81k4
≤ ∑ 4k3 n4
k=1
n/1000  k
81e3 k
= ∑
k=1 4n
= o(1).

If n is even then we begin our search for a Hamilton cycle by choosing a cycle
of H1 and removing an edge. This will give us our current path P. If n is odd
we use the path P joining the two vertices of degree one in M1 ∪ M2 . We can
ignore the case where the isolated vertex is the same in M1 and M2 because
this only happens with probability 1/n. We run Algorithm Pósa of Section 6.2
and observe the following: At each point of the algorithm we will have a path P
plus a collection of vertex disjoint cycles spanning the vertices not in P. This is
because in Step (d) the edge {u, v} will join two cycles, one will be the newly
closed cycle and the other will be a cycle of M. It follows that w.h.p. we will
only need to execute Step (d) at most 3 log n times.
We now estimate the probability that we reach the start of Step (d) and fail
to close a cycle. Let the edges of G0 be {e1 , e2 , . . . , en } where ei is the edge
chosen by vertex i. Suppose that at the beginning of Step (d) we have identified
END. We can go through the vertices of END until we find x ∈ END such that
ex = {x, y} where y ∈ END(x). Because G0 and H1 are independent, we see
by Lemma 17.17 that we can assume P(y ∈ END(x)) ≥ 1/1000. Here we use
the fact that adding edges to H1 will not decrease the size of neighborhoods. It
follows that with probability 1 − o(1/n) we will examine fewer than (log n)2
edges of G0 before we succeed in closing a cycle.
Now we tryclosing cycles O(log n) times and w.h.p. each time we look at
O((log n)2 ) edges of G0 . So, if we only examine an edge of G0 once, we
will w.h.p. still always have n/1000 − O((log n)3 ) edges to try. The proba-
bility we fail to find a Hamilton cycle this way, given that H1 has sufficient ex-
 17.4 Nearest Neighbor Graphs 361

pansion, can therefore be bounded by P(Bin(n/1000 − O((log n)3 ), 1/1000) ≤

3 log n) = o(1).

17.4 Nearest Neighbor Graphs

Consider the complete graph Kn , on vertex set V = {1, 2, , . . . , n}, in which each
edge is assigned a cost Ci, j , i 6= j, and the costs are independent identically
distributed continuous random variables. Color an edge green if it is one of
the k shortest edges incident to either end vertex, and color it blue otherwise.
The graph made up of the green edges only is called the k-th nearest neighbor
graph and is denoted by Gk−nearest . Note that in the random graph Gk−nearest
the edges are no longer independent, as in the case of Gk−out or in the classical
model Gn,p . Assume without loss of generality that the Ci, j are exponential
random variables of mean one. Cooper and Frieze [210] proved

Theorem 17.18

0

 if k = 1,
lim P(Gk−nearest is connected ) = γ if k = 2,
n→∞ 
if k ≥ 3,

1

where 0.99081 ≤ γ ≤ 0.99586.

A similar result holds for a random bipartite k-th nearest neighbor graph, gen-
erated in a similar way as Gk−nearest but starting with the complete bipartite
graph Kn,n with vertex sets V1 ,V2 = {1, 2, , . . . , n}, and denoted by Bk−nearest .
The following result is from Pittel and Weishar [647].
Theorem 17.19

0

 if k = 1,
lim P(Bk−nearest is connected ) = γ if k = 2,
n→∞ 
if k ≥ 3,

1

where 0.996636 ≤ γ.
The paper [647] contains an explicit formula for γ.

Consider the related problem of the existence of a perfect matching in the

bipartite k-th nearest neighbor graph Bk−nearest . For convenience, to simplify
computations, we will assume here that the Ci, j are iid exponential random
362 k-out

variables with rate 1/n. Coppersmith and Sorkin [230] showed that the ex-
pected size of the largest matching in B1−nearest (which itself is a forest) is
w.h.p. asymptotic to
 
−1 −e−1
2 − e−e − e−e n ≈ 0.807n.

The same expression was obtained independently in [647]. Also, w.h.p.,

B2−nearest does not have a perfect matching. Moreover, w.h.p., in a maximal
matching there are at least 132log
log n
log n unmatched vertices, see [647].
The situation changes when each vertex chooses three, instead of one or two,
of its “green” edges. Then the following theorem was proved in [647]:

Theorem 17.20 B3−nearest has a perfect matching, w.h.p.

Proof The proof is analogous to the proof of Theorem 17.6 and uses Hall’s
Theorem. We use the same terminology. We can, as in Theorem 17.6, consider
only bad pairs of “size” k ≤ n/2. Consider first the case when k < εn, where
ε < 1/(2e2 ), i.e., “small” bad pairs. Notice, that in a bad pair, each of the k
vertices from V1 must choose its neighbors from the set of k − 1 vertices from
V2 . Let Ak be the number of such sets. Then,
  3k  3k  2 k
n2k
 
n n k k ke
E Ak ≤ 2 ≤2 ≤ 2 .
k k−1 n (k!)2 n n
(The factor 2 arises from allowing R to be chosen from V1 or V2 .)
Let Pk be the probability that there is a bad pair of size k in B3−nearest . Then
the probability that B3−nearest contains a bad pair of size less than t = bεnc is,
letting l = b(log n)2 c, at most
t t  2 k
ke
P
∑ k ∑ n ≤ 2
k=4 k=4
l  2 k t  2 k
ke ke
=2∑ +2 ∑
k=4 n k=l+1 n
l k t
le2
 

k
≤2∑ + 2 ∑ εe2
k=4 n k=l+1
2l 2 e8
l
≤ + εe2 .
n4
So, if ε < 1/(2e2 ), then
bεnc
∑ Pk → 0.
k=4
 17.4 Nearest Neighbor Graphs 363

It suffices to show that

n/2
∑ Pk → 0.
k=bεnc+1

To prove that there are no “large” bad pairs, note that for a pair to be bad it must
be the case that there is a set of n − k + 1 vertices of V2 that do not choose any
of the k vertices from V1 . Let R ⊂ V1 , |R| = k and S ⊂ V2 , |S| = k − 1. Without
loss of generality, assume that R = {1, 2, . . . k}, S = {1, 2, . . . k − 1}. Then let
Yi , i = 1, 2, . . . k be the smallest weight in Kn,n among the weights of edges
connecting vertex i ∈ R with vertices from V2 \ S, and let Z j , j = k, k + 1, . . . n
be the smallest weight among the weights of edges connecting vertex j ∈ V2 \ S
with vertices from R. Then, each Yi has an exponential distribution with rate
(n − k + 1)/n and each Z j has the exponential distribution with rate k/n.
Notice that in order for there not to be any edge in B3−nearest between respective
sets R and V2 \ S the following property should be satisfied: each vertex i ∈ R
has at least three neighbors in Kn,n with weights smaller than Yi and each vertex
j ∈ V2 \ S has at least three neighbors in Kn,n with weights smaller than the
corresponding Z j . If we condition on the value Yi = y, then the probability that
vertex i has at least three neighbors with respective edge weight smaller than
Yi , is given by
 k−1   k−2
Pn,k (y) = 1 − e−y/n − (k − 1) 1 − e−y/n e−y/n
 
k−1  2  k−3
− 1 − e−y/n e−y/n
2
Putting a = k/n
1
Pn,k (y) ≈ f (a, y) = 1 − e−ay − aye−ay − a2 y2 e−ay .
2
Similarly, the probability that there are three neighbors of vertex j ∈ V2 \ S with
edge weights smaller than Z j is ≈ f (1 − a, Z j ).
So, the probability that there is a bad pair in B3−nearest can be bounded by
  
n n
Pk ≤ 2 Ek ,
k k−1
where, by the Cauchy-Schwarz inequality and independence separately of Y1 ,
. . . ,Yn and Z1 , . . . , Zn ,
!
k n
Ek = E ∏ f (a,Yi ) ∏ f (1 − a, Z j )
i=1 j=k
364 k-out
!!1/2 !!1/2
k n
2 2
≤ E ∏f (a,Yi ) ∏f (1 − a, Z j )
i=1 j=k
k n
= ∏ E( f 2 (a,Yi ))1/2 ∏ E( f 2 (1 − a, Z j ))1/2
i=1 j=k
2
= E( f (a,Y1 )) k/2
E( f 2 (1 − a, Zn ))(n−k+1)/2 .

Asymptotically, Y1 has an exponential (1 − a) distribution, so

E( f 2 (a,Y1 ))
1 2 2 −ay 2
Z ∞ 
−ay −ay
≈ 1 − e − aye − a y e (1 − a)e−(1−a)y dy
0 2
Z ∞
= (1 − a) (e−(1−a)y − 2ey − 2aye−y − a2 y2 e−y + e−(1+a)y
0
1
+ 2aye−(1+a)y + 2a2 y2 e−(1+a)y + a3 y3 e−(1+a)y + a4 y4 e−(1+a)y )dy.
4
Now using
i!
Z ∞
yi e−cy dy = ,
0 ci+1
we obtain

1 1
E( f 2 (a,Y1 )) = (1 − a) − 2 − 2a − 2a2 +
1−a 1+a
4a2 6a3 6a4

2a
+ 2
+ 3
+ 4
+
(1 + a) (1 + a) (1 + a) (1 + a)5
2a6 (10 + 5a + a2 )
= .
(1 + a)5
Letting
g(a) = E( f 2 (a,Y1 ))a/2 ,

we have
    n
n n g(a)g(1 − a)
Pk ≤ 2 (g(a)g(1 − a))n ≈ 2 2a = 2h(a)n .
k k−1 a (1 − a)2(1−a)
Numerical examination of the function h(a) shows that it is bounded below 1
for a in the interval [δ , 0.5], which implies that the expected number of bad
pairs is exponentially small for any k > δ n, with k ≤ n/2. Taking δ < ε <
1/(2e2 ), we conclude that, w.h.p., there are no bad pairs in B3−nearest , and so
we arrive at the theorem.
 17.5 Exercises 365

17.5 Exercises
17.5.1 Let p = log n+(m−1)n log log n+ω where ω → ∞. Show that w.h.p. it is pos-
sible to orient the edges of Gn,p to obtain a digraph D such that the
minimum out-degree δ + (D) ≥ m.
17.5.2 The random digraph Dk−in,`−out is defined as follows: each vertex v ∈
[n] independently randomly chooses k-in-neighbors and `-out-neighbors.
Show that w.h.p. Dm−in,m−out is m-strongly connected for m ≥ 2 i.e. to
destroy strong connectivity, one must delete at least m vertices.
17.5.3 Show that w.h.p. the diameter of Gk−out is asymptotically equal to
log2k n for k ≥ 2.
17.5.4 For a graph G = (V, E) let f : V → V be a G-mapping if (v, f (v)) is an
edge of G for all v ∈ V . Let G be a connected graph with maximum de-
gree d. Let H = ki=0 Hi where (i) k ≥ 1, (ii) H0 is an arbitrary spanning
S

tree of G and (iii) H1 , H2 , . . . , Hk are independent uniform random G-

2k
mappings. Let θk = 1 − 1 − d1 and let α = 16/θk . Show that w.h.p.
for every A ⊂ V , we have
θk
|eH (A)| ≥ · |eG (A)|.
16 log n
where eG (A) (resp. eH (A)) is the number of edges of G (resp. H) with
exactly one endpoint in A.
17.5.5 Let G be a graph with n vertices and minimum degree ( 12 +ε)n for some
fixed ε > 0. Let H = ki=1 Hi where (i) k ≥ 2 and (ii) H1 , H2 , . . . , Hk
S

are independent uniform random G-mappings. Show that w.h.p. H is

connected.
17.5.6 Show that w.h.p. Gk−out contains k edge disjoint spanning trees. (Hint:
Use the Nash-Williams condition [618] – see Frieze and Łuczak [346]).

17.6 Notes
k-out process
Jaworski and Łuczak [452] studied the following process that generates Gk−out
along the way. Starting with the empty graph, a vertex v is chosen uniformly
at random from the set of vertices of minimum out-degree. We then add the
arc (v, w) where w is chosen uniformly at random from the set of vertices that
are not out-neighbors of v. After kn steps the digraph in question is precisely
~Gk−out . Ignoring orientation, we denote the graph obtained after m steps by
366 k-out

U(n, m). The paper [452] studied the structure of U(n, m) for n ≤ m ≤ 2m.
These graphs sit between random mappings and G2−out .

Nearest neighbor graphs

There has been some considerable research on the nearest neighbor graph gen-
erated by n points X = {X1 , X2 , . . . , Xn } chosen randomly in the unit square.
Given a positive integer k we define the k-nearest neighbor graph GX ,k by
joining vertex X ∈ X to its k nearest neighbors in Euclidean distance. We
first consider the existence of a giant component. Teng and Yao [727] showed
that if k ≥ 213 then there is a giant component w.h.p. Balister and Bollobás
[51] reduced this number to 11. Now consider connectivity. Balister, Bollobás,
Sarkar and Walters [53] proved that there exists a critical constant c∗ such that
if k ≤ c log n and c < c∗ then w.h.p. GX ,k is not connected and if k ≥ c log n
and if c > c∗ then w.h.p. GX ,k is connected. The best estimates for c∗ are given
in Balister, Bollobás, Sarkar and Walters [52] i.e. 0.3043 < c∗ < 0.5139.
When distances are independently generated then the situation is much clearer.
Cooper and Frieze [210] proved that if k = 1 then the k-nearest neighbor
graph O1 is not connected; the graph O2 is connected with probability γ ∈
[.99081, .99586]; for k ≥ 2, the graph Ok is k-connected w.h.p.

Directed k-in, `-out

There is a natural directed version of Gk−out called Dk−in,`−out where each
vertex randomly chooses k in-neighbors and ` out-neighbors.
Cooper and Frieze [208] studied the connectivity of such graphs. They prove
for example that if 1 ≤ k, ` ≤ 2 then

lim P(Dk−in,`−out is strongly connected) =

n→∞
(1 − (2 − k)e−` )(1 − (2 − `)e−k ).

In this result, one can in a natural way allow k, ` ∈ [1, 2]. Hamiltonicity was
discussed in [211] where it was shown that w.h.p. D2−in,2−out is Hamiltonian.
The random digraph Dn,p as well as ~Gk−out are special cases of a random di-
graph where each vertex, independently of others, first chooses its out-degree d
according to some probability distribution and then the set of its images - uni-
formly from all d-element subsets of the vertex set. If d is chosen according
to the binomial distribution then it is Dn,p while if d equals k with probabil-
ity 1, then it is ~Gk−out . Basic properties of the model (monotone properties,
 17.6 Notes 367

k-connectivity), were studied in Jaworski and Smit [454] and in Jaworski and
Palka [453] .

k-out subgraphs of large graphs

Just as in Section 6.5, we can consider replacing the host graph Kn by graphs of
large degree. Let an n vertex graph G be strongly Dirac if its minimum degree
is at least cn for some constant c > 1/2. Frieze and Johansson [339] consider
the subgraph Gk obtained from G by letting each vertex independently choose k
neighbors in G. They show that w.h.p. Gk is k-connected for k ≥ 2 and that Gk is
Hamiltonian for k sufficiently large. The paper by Frieze, Goyal, Rademacher
and Vempala [335] shows the use of Gk as a cut-sparsifier.

k-out with preferential attachment

Peterson and Pittel [638] considered the following model: Vertices
1, 2, . . . , n in this order, each choose k random out-neighbors one at a time,
subject to a “preferential attachment” rule: the current vertex selects vertex i
with probability proportional to a given parameter α = α(n) plus the number
of times i has already been selected. Intuitively, the larger α gets, the closer
the resulting k-out mapping is to the uniformly random k-out mapping. They
prove that α = Θ(n1/2 ) is the threshold for α growing “fast enough” to make
the random digraph approach the uniformly random digraph in terms of the
total variation distance. They also determine an exact limit of this distance for
α = β n1/2 .
 18
Real World Networks

There has recently been an increased interest in the networks that we see
around us in our every day lives. Most prominent are the Internet or the World
Wide Web or social networks like Facebook and Linked In. The networks are
constructed by some random process. At least we do not properly understand
their construction. It is natural to model such networks by random graphs.
When first studying so-called “real world networks”, it was observed that of-
ten the degree sequence exhibits a tail that decays polynomially, as opposed
to classical random graphs, whose tails decay exponentially. See, for example,
Faloutsos, Faloutsos and Faloutsos [292]. This has led to the development of
other models of random graphs such as the ones described below.

18.1 Preferential Attachment Graph

Fix an integer m > 0, constant and define a sequence of graphs G1 , G2 , . . . , Gt .
The graph Gt has vertex set [t] and G1 consists of m loops on vertex 1. Suppose
we have constructed Gt . To obtain Gt+1 we apply the following rule. We add
vertex t + 1 and connect it to m randomly chosen vertices y1 , y2 , . . . , ym ∈ [t] in
such a way that for i = 1, 2, . . . , m,

deg (w, Gt )
P(yi = w) = .
2mt

In this way, Gt+1 is obtained from Gt by adding vertex t + 1 and m randomly

chosen edges, in such a way that the neighbors of t + 1 are biased towards
higher degree vertices.
When m = 1, Gt is a tree and this is basically a plane-oriented recursive tree as
considered in Section 15.5.

This model was considered by Barabási and Albert [57]. This was followed
by a rigorous analysis of a marginally different model in Bollobás, Riordan,
Spencer and Tusnády [151].

368
 18.1 Preferential Attachment Graph 369

Expected Degree Sequence: Power Law

Fix t and let Vk (t) denote the set of vertices of degree k in Gt , where m ≤ k =
Õ(t 1/2 ). Let Dk (t) = |Vk (t)|. Then (compare with (15.30) when m = 1)

E(Dk (t + 1)|Gt ) =
 
(k − 1)Dk−1 (t) kDk (t)
Dk (t) + m − + 1k=m + ε(k,t). (18.1)
2mt 2mt
Explanation of (18.1): The total degree of Gt is 2mt and so
(k−1)Dk−1 (t)
2mt is the probability that yi is a vertex of degree k − 1, creating a
k (t)
new vertex of degree k. Similarly, kD2mt is the probability that yi is a vertex of
degree k, destroying a vertex of degree k. At this point t + 1 has degree m and
this accounts for the term 1k=m . The term ε(k,t) is an error term that accounts
for the possibility that yi = y j for some i 6= j.
Thus
  
m k
ε(k,t) = O = Õ(t −1/2 ). (18.2)
2 mt
Taking expectations over Gt , we obtain
 
(k − 1)D̄k−1 (t) kD̄k (t)
D̄k (t + 1) = D̄k (t) + 1k=m + m − + ε(k,t). (18.3)
2mt 2mt
Under the assumption D̄k (t) ≈ dk t (justified below) we are led to consider the

recurrence

(k−1)dk−1 −kdk
1k=m +

 2 if k ≥ m,
dk = (18.4)


0 if k < m,
or

k−1 2·1k=m
 k+2 dk−1 + k+2

 if k ≥ m,
dk =


0 if k < m.

Therefore
2
dm =
m+2
k
l −1 2m(m + 1)
dk = dm ∏ = . (18.5)
l=m+1 l + 2 k(k + 1)(k + 2)
370 Real World Networks

So for large k, under our assumption D̄k (t) ≈ dk t, we see that

2m(m + 1)
D̄k (t) ≈ t.
k3
We now show that the assumption D̄k (t) ≈ dk t can be justified. Note that the

following theorem is vacuous for k  t 1/6 .

Theorem 18.1

|D̄k (t) − dk t| = Õ(t 1/2 ) for k = Õ(t 1/2 ).

Proof Let
∆k (t) = D̄k (t) − dk t.

Then, replacing D̄k (t) by ∆k (t) + dk t in (18.3) and using (18.2) and (18.4) we
get
 
k−1 k
∆k (t + 1) = ∆k−1 (t) + 1 − ∆k (t) + Õ(t −1/2 ). (18.6)
2t 2t
Now assume inductively on t that for every k ≥ 0

|∆k (t)| ≤ At 1/2 (logt)β ,

where (logt)β is the hidden power of logarithm in Õ(t −1/2 ) of (18.6) and A is
an unspecified constant.
This is trivially true for k < m also for small t if we make A large enough. So,
replacing Õ(t −1/2 ) in (18.6) by the more explicit αt −1/2 (logt)β we get

∆k (t + 1) ≤
    
k −1 k
∆k (t) + αt −1/2 (logt)β
  
≤  ∆k−1 (t) +  1 −
2t 2t
 
k − 1 1/2 k
≤ At (logt)β + 1 − At 1/2 (logt)β + αt −1/2 (logt)β
2t 2t
≤ (logt)β (At 1/2 + αt −1/2 ).

Note that if t is sufficiently large then

1 1/2
 
1/2 1/2 1
(t + 1) = t 1+ ≥ t 1/2 + 1/2 ,
t 3t
and so
   
β 1/2 1 α
∆k (t + 1) ≤ (log(t + 1)) A (t + 1) − 1/2 + 1/2
3t t
 18.1 Preferential Attachment Graph 371

≤ A (log(t + 1))β (t + 1)1/2 ,

assuming that A ≥ 3α.
In the next section, we will justify our bound of Õ(t 1/2 ) for vertex degrees.
After that we will prove concentration of the number of vertices of degree k,
for small k.

Maximum Degree
Fix s ≤ t and let Xl be the degree of vertex s in Gl for s ≤ l ≤ t.

Lemma 18.2
P(Xt ≥ Aem (t/s)1/2 (log(t + 1))2 ) = O(t −A ).
1
Proof Note first that Xs = m. If 0 < λ < εt = log(t+1) then,
 
E eλ Xl+1 |Xl
m  
Xl k Xl m−k λ k
  
λ Xl m
=e ∑ 1− e
k=0 k 2ml 2ml
m  
Xl k Xl m−k
  
λ Xl m
≤e ∑ 1− (1 + kλ (1 + kλ ))
k=0 k 2ml 2ml
mλ Xl (m − 1)λ 2 Xl2
 
λ Xl
=e 1+ +
2m 4ml 2
 
λ Xl Xl
≤e 1 + λ (1 + mλ ) , since Xl < ml,
2l
 
(1+mλ )
λ 1+ 2l Xl
≤e .
We define a sequence λ = (λl , λl+1 , . . . , λt ) where
 
1 + mλ j
λ j+1 = 1 + λ j < εt .
2j
Here our only choice will be λl . We show below that we can find a suitable
value for this, but first observe that if we manage this then
     
E eλ Xt ≤ E eλl+1 Xt−1 · · · ≤ E eλt Xl ≤ 1 + o(1).

Now  
1 + mεt
λ j+1 ≤ 1 + λ j,
2j
372 Real World Networks

implies that
( )
t   t  t 1/2
1 + mεt 1 + mεt
λt = λl ∏ 1 + ≤ λl exp ∑ ≤ em λl .
j=l 2j j=l 2j l

So a suitable choice for λ = λl is

 1/2
l
λl = e−m εt .
t
This gives
 n o
E exp e−m εt (l/t)1/2 Xt ≤ 1 + o(1).

So,
 
P Xt ≥ Aem (t/l)1/2 (log(t + 1))2 ) ≤
m (t/l)1/2 (log(t+1)2 )
 
e−λ Ae E eλ Xt = O(t −A ).

Thus with probability 1 − o(1) as t → ∞ we have that the maximum degree in

Gt is O(t 1/2 (logt)2 ). This is not best possible. One can prove that w.h.p. the
maximum degree is O(t 1/2 ω(t)) and Ω(t 1/2 /ω(t)) for any ω(t) → ∞, see for
example Flaxman, Frieze and Fenner [308].

Concentration of Degree Sequence

Fix a value k for a vertex degree. We show that Dk (t) is concentrated around
its mean D̄k (t).

Theorem 18.3
u2
 
P(|Dk (t) − D̄k (t)| ≥ u) ≤ 2 exp − . (18.7)
8mt
Proof Let Y1 ,Y2 , . . . ,Ymt be the sequence of edge choices made in the con-
struction of Gt , and for Y1 ,Y2 , . . . ,Yi let
Zi = Zi (Y1 ,Y2 , . . . ,Yi ) = E(Dk (t) | Y1 ,Y2 , . . . ,Yi ).
We will prove next that |Zi − Zi−1 | ≤ 4 and then (18.7) follows directly from
the Azuma-Hoeffding inequality, see Section 22.7. Fix Y1 ,Y2 , . . . ,Yi and Ŷi 6= Yi .
We define a map (measure preserving projection) ϕ of
Y1 ,Y2 , . . . ,Yi−1 ,Yi ,Yi+1 , . . . ,Ymt
 18.1 Preferential Attachment Graph 373
000
111
111
000
000
111
000
111
111
000
000
111
v 00
11
11
00
00
11
00
11
11
00
00
11
v
000
111
000
111 000
111
000
111 00
11
00
11 00
11
00
11

111
000 11
00
000
111
000
111 00
11
00
11
000
111
000
111 00
11
00
11

111
000
000
111 11
00
00
11
000
111
000
111 00
11
00
11
000
111 00
11
111
000
000
111 111
000
000
111
000
111
000
111 000
111
000
111
000
111 000
111
111
000
000
111 11
00
00
11
000
111 00
11
000
111
000
111 00
11
00
11

G G

~ from G.
Figure 18.1 Constructing G

to
Y1 ,Y2 , . . . ,Yi−1 , Ŷi , Ŷi+1 , . . . , Ŷmt

such that
|Zi (Y1 ,Y2 , . . . ,Yi ) − Zi (Y1 ,Y2 , . . . , Ŷi )| ≤ 4.

In the preferential attachment model we can view vertex choices in the graph
G as random choices of arcs in a digraph G, ~ which is obtained by replacing
every edge of G by a directed 2-cycle (see Figure 18.1).
Indeed, if we choose a random arc and choose its head then v will be cho-
sen with probability proportional to the number of arcs with v as head i.e. its
degree. Hence Y1 ,Y2 , . . . can be viewed as a sequence of arc choices. Let

Yi = (x, y) where x > y

Ŷi = (x̂, ŷ) where x̂ > ŷ.

Note that x = x̂ if i mod m 6= 1.

Now suppose j > i and Y j = (u, v) arises from choosing (w, v). Then we define
(
Yj (w, v) 6= Yi
ϕ(Y j ) =
(w, ŷ) (w, v) = Yi

This map is measure preserving since each sequence ϕ(Y1 ,Y2 , . . . ,Yt ) occurs
with probability ∏tm −1
j=i+1 j . Only x, x̂, y, ŷ change degree under the map ϕ so
Dk (t) changes by at most four.
374 Real World Networks

18.2 A General Model of Web Graphs

In the model presented in the previous section a new vertex is added at time t
and this vertex chooses m random neighbors, with probability proportional to
their current degree. Cooper and Frieze [213] generalise this in the following
ways: they allow (a) new edges to be inserted between existing vertices, (b)
a variable number of edges to be added at each step and (c) endpoint vertices
to be chosen by a mixture of uniform selection and copying. This results in
a large number of parameters, which will be described below. We first give a
precise description of the process.
Initially, at step t = 0, there is a single vertex v0 . At any step t = 1, 2, . . . , T, . . . ,
there is a birth process in which either new vertices or new edges are added.
Specifically, either a procedure NEW is followed with probability 1 − α, or
a procedure OLD is followed with probability α. In procedure NEW, a new
vertex v is added to Gt−1 with one or more edges added between v and Gt−1 .
In procedure OLD, an existing vertex v is selected and extra edges are added at
v.
The recipe for adding edges at step t typically permits the choice of initial
vertex v (in the case of OLD) and of terminal vertices (in both cases) to be made
from Gt−1 either u.a.r (uniformly at random) or according to vertex degree, or
a mixture of these two based on further sampling. The number of edges added
to vertex v at step t by the procedures (NEW, OLD) is given by distributions
specific to the procedure.
Notice that the edges have an intrinsic direction, arising from the way they are
inserted, which one can ignore or not. Here the undirected model is considered
with a sampling procedure based on vertex degree. The process allows multiple
edges, and self-loops can arise from the OLD procedure. The NEW procedure,
as described, does not generate self-loops, although this could easily be modi-
fied.

Sampling parameters, notation and main properties

Our undirected model Gt has sampling parameters α, β , γ, δ , p, q whose mean-
ing is given below:
Choice of procedure at step t.
α: Probability that an OLD vertex generates edges.
1 − α: Probability that a NEW vertex is created.
Procedure NEW
p = (pi : i ≥ 1): Probability that the new node generates i new edges.
β : Probability that choices of terminal vertices are made uniformly.
 18.2 A General Model of Web Graphs 375

1 − β : Probability that choices of terminal vertices are made according to

degree.
Procedure OLD
q = (qi : i ≥ 1): Probability that the old node generates i new edges.
δ : Probability that the initial node is selected uniformly.
1 − δ : Probability that the initial node is selected according to degree.
γ: Probability that choices of terminal vertices are made uniformly.
1 − γ: Probability that choices of terminal vertices are made according to
degree.
The models require α < 1 and p0 = q0 = 0. It is convenient to assume a finite-
ness condition for the distributions {p j }, {q j }. This means that there exist
j0 , j1 such that p j = 0, j > j0 and q j = 0, j > j1 . Imposing the finiteness
condition helps simplify the difference equations used in the analysis.
The model creates edges in the following way: An initial vertex v is selected. If
the terminal vertex w is chosen u.a.r, we say v is assigned uniformly to w. If the
terminal vertex w is chosen according to its vertex degree, we say v is copied
to w. In either case the edge has an intrinsic direction (v, w), which we may
choose to ignore. Note that sampling according to vertex degree is equivalent
to selecting an edge u.a.r and then selecting an endpoint u.a.r.
Let
j0 j1
µp = ∑ j p j , µq = ∑ jq j
j=1 j=1

be the expected number of edges added by NEW or OLD and let

θ = 2((1 − α)µ p + α µq ).

To simplify subsequent notation, we introduce new parameters as follows:

αγ µq αδ
a = 1 + β µp + + ,
1−α 1−α
(1 − α)(1 − β )µ p α(1 − γ)µq α(1 − δ )
b= + + ,
θ θ θ
αγ µq
c = β µp + ,
1−α
(1 − α)(1 − β )µ p α(1 − γ)µq
d= + ,
θ θ
αδ
e= ,
1−α
α(1 − δ )
f= .
θ
376 Real World Networks

We note that
c + e = a − 1 and b = d + f . (18.8)

Now define the sequence (d0 , d1 , . . . , dk , . . .) by d0 = 0, and for k ≥ 1

k−1
dk (a + bk) = (1 − α)pk + (c + d(k − 1))dk−1 + ∑ (e + f (k − j))q j dk− j .
j=1
(18.9)
For convenience we define dk = 0 for k < 0. Since a ≥ 1, this system of equa-
tions has a unique solution.
The main quantity we study is the random variable Dk (t), the number of ver-
tices of degree k at step t. Cooper and Frieze [213] prove that, as t → ∞, for
small k, Dk (t) ≈ dk t.

Theorem 18.4 There exists a constant M > 0 such that almost surely for all
t, k ≥ 1
|Dk (t) − tdk | ≤ Mt 1/2 logt.

This will be proved in Section 18.2.

It is shown in (18.10), that the number of vertices ν(t) at step t is w.h.p. asymp-
totic to (1 − α)t. It follows that the proportion of vertices of degree k is w.h.p.
asymptotic to
dk
d¯k = .
1−α
The next theorem summarises what is known about the sequence (dk ) defined
by (18.9).

Theorem 18.5 There exist constants C1 ,C2 ,C3 ,C4 > 0 such that

(i) C1 k−ζ ≤ dk ≤ C2 min{k−1 , k−ζ / j1 } where ζ = (1 + d + f µq )/(d + f ).

(ii) If j1 = 1 then dk ≈ C3 k−(1+1/(d+ f )) .
(iii) If f = 0 then dk ≈ C4 k−(1+1/d) .

Evolution of the degree sequence of Gt

Let ν(t) = |V (t)| be the number of vertices and let η(t) = |2E(t)| be the total
degree of the graph at the end of step t. E ν(t) = (1 − α)t and E η(t) = θt. The
 18.2 A General Model of Web Graphs 377

random variables ν(t), η(t) are sharply concentrated provided t → ∞. Indeed

ν(t) has distribution Bin(t, 1 − α) and so by Theorem 22.6 and its corollaries,

P(|ν(t) − (1 − α)t| ≥ t 1/2 logt) = O(t −K ) (18.10)

for any constant K > 0.

Similarly, η(t) has expectation θt and is the sum of t independent random vari-
ables, each bounded by max{ j0 , j1 }. Hence, by Theorem 22.6 and its corollar-
ies,
P(|η(t) − θt| ≥ t 1/2 logt) = O(t −K ) (18.11)

for any constant K > 0.

These results are almost sure in the sense that they hold for all t ≥ t0 with
probability 1 − O(t0−K+1 ). Thus we can focus on processes such that this is
true.
We remind the reader that Dk (t) is the number of vertices of degree k at step t
and that Dk (t) is its expectation. Here D j (t) = 0 for all j ≤ 0,t ≥ 0, D1 (0) =
1, Dk (0) = 0, k ≥ 2.
Using (18.10) and (18.11) we see that

Dk (t + 1) = Dk (t) + (1 − α)pk + O(t −1/2 logt) (18.12)

j0   
jDk−1 (t) jDk (t)
+ (1 − α) ∑ p j β −
j=1 (1 − α)t (1 − α)t
 
j(k − 1)Dk−1 (t) jkDk (t)
+ (1 − β ) − (18.13)
θt θt
 
δ Dk (t) (1 − δ )kDk (t)
−α +
(1 − α)t θt
j1
δ Dk− j (t) (1 − δ )(k − j)Dk− j (t)
 
+α ∑ qj + (18.14)
j=1 (1 − α)t θt
j1   
Dk−1 (t) Dk (t)
+ α ∑ jq j γ − +
j=1 (1 − α)t (1 − α)t
 
(k − 1)Dk−1 (t) kDk (t)
(1 − γ) − . (18.15)
θt θt
Here (18.13), (18.14), (18.15) are (respectively) the main terms of the change
in the expected number of vertices of degree k due to the effect on: terminal
vertices in NEW, the initial vertex in OLD and the terminal vertices in OLD.
Rearranging the right hand side, we find:

Dk (t + 1) = Dk (t) + (1 − α)pk + O(t −1/2 logt)

378 Real World Networks

Dk (t) αγ µq αδ (1 − α)(1 − β )µ p k
− β µp + + + +
t 1−α 1−α θ

α(1 − γ)µq k α(1 − δ )k
+ +
θ θ
Dk−1(t)

αγ µq (1 − α)(1 − β )µ p (k − 1)
+ β µp + + +
t 1−α θ

α(1 − γ)µq (k − 1)
+
θ
j1  
Dk− j (t) αδ α(1 − δ )(k − j)
+ ∑ qj + .
j=1 t 1−α θ

Thus for all k ≥ 1 and almost surely for all t ≥ 1,

Dk (t + 1) = Dk (t) + (1 − α)pk + O(t −1/2 logt) (18.16)

1
+ ((1 − (a + bk))Dk (t) + (c + d(k − 1))Dk−1 (t)
t
j1
+ ∑ q j (e + f (k − j))Dk− j (t)).
j=1

The following Lemma establishes an upper bound on dk given in Theorem

18.5(i).

C2
Lemma 18.6 The solution of (18.9) satisfies dk ≤ k .

Proof Assume that k > k0 where k0 is sufficiently large, and thus pk = 0.

Smaller values of k can be dealt with by adjusting C2 . We proceed by induction
on k. From (18.9),
j1
C2 C2
(a + bk)dk ≤ (c + d(k − 1)) + ∑ (e + f (k − j))q j
k − 1 j=1 k− j
C2 (c + e)
≤ C2 (d + f ) +
k − j1
C2 (a − 1)
= C2 b + ,
k − j1
from (18.8). So
C2 C2 b C2 (a − 1) C2
dk − ≤ + −
k a + bk (k − j1 )(a + bk) k
 18.2 A General Model of Web Graphs 379
C2 (a − 1) C2 a
= −
(k − j1 )(a + bk) k(a + bk)
≤ 0,

for k ≥ j1 a.
We can now prove Theorem 18.4, which is restated here for convenience.

Theorem 18.7 There exists a constant M > 0 such that almost surely for
t, k ≥ 1,
|Dk (t) − tdk | ≤ Mt 1/2 logt. (18.17)

Proof Let ∆k (t) = Dk (t) − tdk . It follows from (18.9) and (18.16) that
 
a + bk − 1
∆k (t + 1) = ∆k (t) 1 − + O(t −1/2 logt)
t
!
j1
1
+ (c + d(k − 1))∆k−1 (t) + ∑ (e + f (k − j))q j ∆k− j (t) . (18.18)
t j=1

Let L denote the hidden constant in O(t −1/2 logt). We can adjust M to deal  with
small values of t, so we assume that t is sufficiently large. Let k0 (t) = t+1−b

a .
t max{ j0 , j1 }
If k > k0 (t) then we observe that (i) Dk (t) ≤ k (t) = O(1) and (ii) tdk ≤
0
t kC(t)
2
= O(1) follows from Lemma 18.6, and so (18.17) holds trivially.
0
Assume inductively that ∆κ (τ) ≤ Mτ 1/2 log τ for κ + τ ≤ k + t and that k ≤
k0 (t). Then (18.18) and k ≤ k0 implies that for M large,
logt
|∆k (t + 1)| ≤ L + Mt 1/2 logt ×
t 1/2 !!
j1
1
1+ c + dk + ∑ (e + f k)q j − (a + bk − 1)
t j=1
logt
=L + Mt 1/2 logt
t 1/2
≤ M(t + 1)1/2 log(t + 1)

provided M  2L. We have used (18.8) to obtain the second line.

This completes the proof by induction.

A general power law bound for dk

The following lemma completes the proof of Theorem 18.5(i).
Lemma 18.8 For k > j0 we have,
380 Real World Networks

(i) dk > 0.
(ii) C1 k−(1+d+ f µq )/b ≤ dk ≤ C2 k−(1+d+ f µq )/b j1 .

Proof (i) Let κ be the first index such that pκ > 0, so that, from (18.9), dκ > 0.
It is not possible for both c and d to be zero. Therefore the coefficient of dk−1
in (18.9) is non-zero and thus dk > 0 for k ≥ κ.
(ii) Re-writing (18.9) we see that for k > j0 , pk = 0 and then dk satisfies
j1
c + d(k − 1) e + f (k − j)
dk = dk−1 + ∑ dk− j q j , (18.19)
a + bk j=1 a + bk

which is a linear difference equation with rational coefficients (see [593]).

We let di = 0 for i < 0 to handle the cases where k − j < 0 in the above sum.
Let y = 1 + d + f µq , then
j1
c + d(k − 1) e + f (k − j) y
+ ∑ qj = 1− ≥0
a + bk j=1 a + bk a + bk

and thus
 
y
1− min{dk−1 , . . . , dk− j1 } ≤ dk ≤
a + bk
 
y
1− max{dk−1 , . . . , dk− j1 }. (18.20)
a + bk
It follows that
k  
y
d j0 ∏ 1− ≤ dk ≤
j= j0 +1 a+bj
b(k− j0 )/ j1 c  
y
max{d1 , d2 , . . . , d j0 } ∏ 1− . (18.21)
s=0 a + b(k − s j1 )

The lower bound in (18.21) is proved by induction on k. It is trivial for k = j0

and for the inductive step we have
( )
  i 
y y
dk ≥ d j0 1 − min 1−
a + bk i= j0 ,...,k−1 j=∏
j0 +1 a+bj
k  
y
= d j0 ∏ 1− .
j= j0 +1 a + bj

The upper bound in (18.21) is proved as follows: Let di1 = max{dk−1 , . . . ,

dk− j1 }, and in general, let dit+1 = max{dit −1 , . . . , dit − j1 }. Using (18.20) we see
 18.2 A General Model of Web Graphs 381

there is a sequence k − 1 ≥ i1 > i2 > · · · > i p > j0 ≥ i p+1 such that |it − it−1 | ≤
j1 for all t, and p ≥ b(k − j0 )/ j1 c. Thus
p  
y
dk ≤ di p+1 ∏ 1 − ,
t=0 a + bit

and the RHS of (18.21) now follows.

Now consider the product in the LHS of (18.21).
k  
y
∏ 1 −
j= j0 +1 a+bj
( 2 !)
k 
y 1 y
= exp ∑ −a+b j − 2 a+b j −···
j= j0 +1
( )
k
y
= exp O(1) − ∑
j= j0 +1 a + b j

= C1 k−y/b .

This establishes the lower bound of the lemma. The upper bound follows sim-
ilarly, from the upper bound in (18.21).

The case j1 = 1
We prove Theorem 18.5(ii). When q1 = 1, p j = 0, j > j0 = Θ(1), the general
value of dk , k > j0 can be found directly, by iterating the recurrence (18.9).
Thus
1
dk = (dk−1 ((a − 1) + b(k − 1)))
a + bk
 
1+b
= dk−1 1 −
a + bk
k  
1+b
= d j0 ∏ 1− .
j= j0 +1 a + jb

Thus, for some constant C6 > 0,

dk ≈ C6 (a + bk)−x

where
1 2
x = 1+ = 1+ .
b α(1 − δ ) + (1 − α)(1 − β ) + α(1 − γ)
382 Real World Networks

The case f = 0
We prove Theorem 18.5(iii). The case ( f = 0) arises in two ways. Firstly if
α = 0 so that a new vertex is added at each step. Secondly, if α 6= 0 but δ = 1
so that the initial vertex of an OLD choice is sampled u.a.r.
Observe that b = d now, see (18.8).
We first prove that for a sufficiently large absolute constant A2 > 0 and for all
sufficiently large k, that
dk 1+d ξ (k)
= 1− + 2 (18.22)
dk−1 a + dk k
where |ξ (k)| ≤ A2 .
We first re-write (18.9) as
j1 k−1
dk c + d(k − 1) eq j dt−1
= +∑ ∏ dt . (18.23)
dk−1 a + dk j=1 a + dk t=k− j+1

(We assume here that k > j0 , so that pk = 0.)

Now use induction to write
k−1
dt−1 d +1 ξ ∗ ( j, k)
∏ = 1 + ( j − 1) + (18.24)
t=k− j+1 dt a + dk k2

where |ξ ∗ ( j, k)| ≤ A3 for some constant A3 > 0. (We use the fact that j1 is
constant here.)
Substituting (18.24) into (18.23) gives
dk c + d(k − 1) e e(µq − 1)(d + 1) ξ ∗∗ (k)
= + + +
dk−1 a + dk a + dk (a + dk)2 (a + dk)k2
where |ξ ∗∗ (k)| ≤ eA3 .
Equation (18.22) follows immediately from this and c + e = a − 1. On iterating
(18.22) we see that for some constant C7 > 0,
1
dk ≈ C7 k−(1+ d ) .

18.3 Small World

In an influential paper Milgram [592] describes the following experiment. He
chose a person X to receive mail and then randomly chose a person Y to send
it. If Y did not know X then Y was to send the mail to someone he/she thought
more likely to know X and so on. Surprisingly, the mail got through in 64 out of
 18.3 Small World 383

296 attempts and the number of links in the chain was relatively small, between
5 and 6. More recently, Kleinberg [498] described a model that attempts to
explain this phenomenon.

Kleinberg’s Model
The model can be generalized significantly, but to be specific we consider
the following. We start with the n × n grid G0 which has vertex set [n]2 and
where (i, j) is adjacent to (i0 , j0 ) iff d((i, j), (i0 , j0 )) = 1 where d((i, j), (k, `)) =
|i − k| + | j − `|. In addition, each vertex u = (i, j) will choose another random
neighbor ϕ(u) where
d(u, v)−2
P(ϕ(u) = v = (k, `)) =
Du
where
Dx = ∑ d(x, y)−2 .
y6=x

The random neighbors model “long range contacts”. Let the grid G0 plus the
extra random edges be denoted by G.
It is not difficult to show that w.h.p. these random contacts reduce the diam-
eter of G to order log n. This however, would not explain Milgram’s success.
Instead, Kleinberg proposed the following decentralized algorithm A for find-
ing a path from an initial vertex u0 = (i0 , j0 ) to a target vertex uτ = (iτ , jτ ):
when at u move to the neighbor closest in distance to uτ .

Theorem 18.9 Algorithm A finds a path from initial to target vertex of order
O((log n)2 ), in expectation.
Proof Note that each step of A finds a node closer to the target than the
current node and so the algorithm must terminate with a path.
Observe next that for any vertex x of G we have
2n−2 2n−2
Dx ≤ ∑ 4 j × j−2 = 4 ∑ j−1 ≤ 4 log(3n).
j=1 j=1

As a consequence, v is the long range contact of vertex u, with probability at

least (4 log(3n)d(u, v)2 )−1 .
For 0 < j ≤ log2 n, we say that the execution of A is in Phase j if the distance
of the current vertex u to the target is greater than 2 j , but at most 2 j+1 . We say
that A is in Phase 0 if the distance from u to the target is at most 2.
384 Real World Networks

Let B j denote the set of nodes at distance 2 j or less from the target. Then
2j
|B j | ≥ 1 + ∑ i > 22 j−1 .
i=1

Note that by the triangle inequality, each member of B j is within distance

2 j+1 + 2 j < 22 j+2 of u.
Let X j ≤ 2 j+1 be the time spent in Phase j. Assume first that
log2 log2 n ≤ j ≤ log2 n. Phase j will end if the long range contact of the current
vertex lies in B j . The probability of this is at least
22 j−1 1
= .
4 log(3n)22 j+4 128 log(3n)
We can reveal the long range contacts as the algorithm progresses. In this way,
the long range contact of the current vertex will be independent of the previous
contacts of the path. Thus
∞ ∞  i
1
E X j = ∑ P(X j ≥ i) ≤ ∑ 1 − < 128 log(3n).
i=1 i=1 128 log(3n)

Now if 0 ≤ j ≤ log2 log2 n then X j ≤ 2 j+1 ≤ 2 log2 n. Thus the expected length
of the path found by A is at most 128 log(3n) × log2 n.
In the same paper, Kleinberg showed that replacing d(u, v)−2 by
d(u, v)−r for r 6= 2 led to non-polylogarithmic path length.

18.4 Exercises
18.4.1 Show that w.h.p. the Preferential Attachment Graph of Section 18.1 has
diameter O(log n). (Hint: Using the idea that vertex t chooses a random
edge of the current graph, observe that half of these edges appeared at
time t/2 or less).
18.4.2 For the next few questions we modify the Preferential Attachment Graph
of Section 18.1 in the following way: First let m = 1 and preferentially
generate a sequence of graphs Γ1 , Γ2 , . . . , Γmn . Then if the edges of Γmn
are (ui , vi ), i = 1, 2, . . . , mn let the edges of Gn be (udi/me , vdi/me ), i =
1, 2, . . . , mn. Show that (18.1) continues to hold.
18.4.3 Show that Gn of the previous question can also be generated in the
following way:
(a) Let π be a random permutation of [2mn]. Let
X = {(ai , bi ), i = 1, 2, . . . , mn} where ai = min {π(2i − 1), π(2i)}
and bi = max {π(2i − 1), π(2i)}.
 18.4 Exercises 385

(b) Let the edges of Gn be (adi/me , bdi/me ), i = 1, 2, . . . , mn.

This model was introduced in [151].
18.4.4 Show that the edges of the graph in the previous question can be gen-
erated as follows:

(a) Let ζ1 , ζ2 , . . . , ζ2mn be independent uniform [0, 1] random variables.

Let {xi < yi } = {ζ2i−1 , ζ2i } for i = 1, 2, . . . , mn. Sort the yi in in-
creasing order R1 < R2 < · · · < Rmn and let R0 = 0. Then let

W j = Rm j and I j = (W j−1 ,W j ] for j = 1, 2, . . . , n.

This model was introduced in [150].

(b) The edges of Gn are (ui , vi ), i = 1, 2, . . . , mn where xi ∈ Iui , yi ∈ Ivi .
18.4.5 Prove that (R1 , R2 , . . . , Rmn ) can be generated as
 1/2
ϒi
Ri =
ϒmn+1

where ϒN = ξ1 + ξ2 + · · · + ξN for N ≥ 1 and ξ1 , ξ2 , . . . , ξmn+1 are in-

dependent exponential copies of EXP(1).
18.4.6 Let L be a large constant and let ω = ω(n) → ∞ arbitrarily slowly. Then
let E be the event that
k
ϒk ≈ k for ∈ [ω, n] or k = mn + 1.
m
Show that

(a) P(¬E ) = o(1).

(b) Let ηi = ξ(i−1)m+1 + ξ(i−1)m+2 + · · · + ξim . If E occurs then
 1/2
i
(1) Wi ≈ for ω ≤ i ≤ n, and
n
ηi
(2) wi = Wi −Wi−1 ≈ for ω ≤ i ≤ n.
2m(in)1/2
(c) ηi ≤ log n for i ∈ [n] w.h.p.
(d) ηi ≤ log log n for i ∈ [(log n)10 ] w.h.p.
ηi
(e) If ω ≤ i < j ≤ n then P(edge i j exists) ≈ 2(i j)1/2
.
1 n n 1/2


(f) ηi ≥ log log n and i ≤ ω(log n)3
implies the degree dn (i) ≈ ηi i .
386 Real World Networks

18.5 Notes
There are by now a vast number of papers on different models of “Real World
Networks”. We point out a few additional results in the area. The books by
Durrett [266] and Bollobás, Kozma and Miklós [148] cover the area. See also
van der Hofstadt [416].

Preferential Attachment Graph

Perhaps the most striking result is due to Bollobás and Riordan [150]. There
they prove that the diameter of Gn is asymptotic to logloglogn n w.h.p. To prove
this they introduced the model in question 4 above. Cooper [205] and Peköz,
Röllin and Ross [635] discuss the degree distribution of Gn in some detail.
Flaxman, Frieze and Fenner [308] show that the if ∆k , λk are the kth largest
1/2
degree, eigenvalue respectively, then k ≈ ∆k for k = O(1). The proof follows
ideas from Mihail and Papadimitriou [591] and Chung, Lu and Vu [186], [187].
Cooper and Frieze [213] discussed the likely proportion of vertices visited by
a random walk on a growing preferential attachment graph. They show that
w.h.p. this is just over 40% at all times. Borgs, Brautbar, Chayes, Khanna and
Lucier [154] discuss “local algorithms” for finding a specific vertex or the
largest degree vertex. Frieze and Pegden [353] describe an algorithm for the
same problem, but with reduced storage requirements.

Geometric models
Some real world graphs have a geometric constraint. Flaxman, Frieze and
Vera [309], [310] considered a geometric version of the preferential attach-
ment model. Here the vertices X1 , X2 , . . . , Xn are randomly chosen points on
the unit sphere in R3 . Xi+1 chooses m neighbors and these vertices are chosen
with probability P(deg, dist) dependent on (i) their current degree and (ii) their
distance from Xi+1 . van den Esker [289] added fitness to the models in [309]
and [310]. Jordan [461] considered more general spaces than R3 . Jordan and
Wade [462] considered the case m = 1 and a variety of definitions of P that
enable one to interpolate between the preferential attachment graph and the
on-line nearest neighbor graph.
The SPA model was introduced by Aiello, Bonato, Cooper, Janssen and Pralat
[7]. Here the vertices are points in the unit hyper-cube D in Rm , equipped with
a toroidal metric. At time t each vertex v has a domain of attraction S(v,t) of
−
volume A1 deg t(v,t)+A2 . Then at time t we generate a uniform random point Xt+1
as a new vertex. If the new point lies in the domain S(v,t) then we join Xt+1 to
 18.5 Notes 387

v by an edge directed to v, with probability p. The paper [7] deals mainly with
the degree distribution. The papers by Jannsen, Pralat and Wilson [446], [447]
show that for graphs formed according to the SPA model it is possible to infer
the metric distance between vertices from the link structure of the graph. The
paper Cooper, Frieze and Pralat [226] shows that w.h.p. the directed diameter
c1 logt
at time t lies between log logt and c2 logt.
Random Apollonian networks were introduced by Zhou, Yan and Wang [759].
Here we build a random triangulation by inserting a vertex into a randomly
chosen face. Frieze and Tsourakakis [357] studied their degree sequence and
eigenvalue structure. Ebrahimzadeh, Farczadi, Gao, Mehrabian, Sato, Wormald
and Zung [274] studied their diameter and length of the longest path. Cooper
and Frieze [221] gave an improved longest path estimate and this was further
improved by Collevecchio, Mehrabian and Wormald [197].

Interpolating between Erdős-Rényi and Preferential Attachment

Pittel [644] considered the following model: G0 , G1 , . . . , Gm is a random (multi)
graph growth process Gm on a vertex set [n]. Gm+1 is obtained from Gm by in-
serting a new edge e at random. Specifically, the conditional probability that e
joins two currently disjoint vertices, i and j, is proportional to (di +α)(d j +α),
where di , d j are the degrees of i, j in Gm , and α > 0 is a fixed parameter. The
limiting case α = ∞ is the Erdős-Rényi graph process. He shows that w.h.p.
Gm contains a unique giant component
α  > cα =∗ α/(1 + α), and
 iff c := ∗2m/n
the size of this giant is asymptotic to n 1 − α+cα+c , where c < cα is the root
c c∗
of (α+c)2+α = (α+c∗ )2+α . A phase transition window is proved to be contained,
essentially, in [cα − An−1/3 , cα + Bn−1/4 ], and he conjectured that 1/4 may be
replaced with 1/3. For the multigraph version, MGm , he showed that MGm is
−1
connected w.h.p. iff m  mn := n1+α . He conjectured that, for α > 1, mn is
the threshold for connectedness of Gm itself.
 19
Weighted Graphs

There are many cases in which we put weights Xe , e ∈ E on the edges of a graph
or digraph and ask for the minimum or maximum weight object. The optimi-
sation questions that arise from this are the backbone of Combinatorial Opti-
misation. When the Xe are random variables we can ask for properties of the
optimum value, which will be a random variable. In this chapter we consider
three of the most basic optimisation problems viz. minimum weight spanning
trees; shortest paths and minimum weight matchings in bipartite graphs.

19.1 Minimum Spanning Tree

Let Xe , e ∈ E(Kn ) be a collection of independent uniform [0, 1] random vari-
ables. Consider Xe to be the length of edge e and let Ln be the length of the
minimum spanning tree (MST) of Kn with these edge lengths.
Frieze [326] proved the following theorem. The proof we give utilises the
rather lovely integral formula (19.1) due to Janson [430], (see also the related
equation (7) from [348].

Theorem 19.1
∞
1
lim E Ln = ζ (3) =
n→∞
∑ k3 = 1.202 · · ·
k=1

Proof Suppose that T = T ({Xe }) is the MST, unique with probability one.
We use the identity
Z 1
a= 1{x≤a} dx.
0
Therefore
Ln = ∑ Xe
e∈T
Z 1
= ∑ 1{p≤Xe } d p
e∈T p=0
Z 1
= ∑ 1{p≤Xe } d p
p=0 e∈T

388
 19.1 Minimum Spanning Tree 389
Z 1
= |{e ∈ T : Xe ≥ p}|d p
p=0
Z 1
= (κ(G p ) − 1)d p,
p=0

where κ(G p ) denote the number of components of graph G p . Here G p is the

graph induced by the edges e with Xe ≤ p, i.e., G p ≡ Gn,p . The last line may be
considered to be a consequence of the fact that the greedy algorithm solves the
minimum spanning tree problem. This algorithm examines edges in increasing
order of edge weight. It builds a tree, adding one edge at a time. It adds the
edge to the forest F of edges accepted so far, only if the two endpoints lie in
distinct components of F. Otherwise it moves onto the next edge. Thus the
number of edges to be added given F, is κ(F) − 1 and if the longest edge in
e ∈ F has Xe = p then κ(F) = κ(G p ), which follows by an easy induction.
Hence
Z 1
E Ln = (E κ(G p ) − 1)d p. (19.1)
p=0

We therefore estimate E κ(G p ). We observe first that

6 log n
p≥ ⇒ E κ(G p ) = 1 + o(1).
n
Indeed, 1 ≤ E κ(G p ) and

E κ(G p ) ≤ 1 + n P(G p is not connected)

n/2  
n k−2 k−1
≤ 1+n ∑ k p (1 − p)k(n−k)
k=1 k

n n/2 ne 6k log n 1 k
 
≤ 1+ ∑
p k=1 k n n3
= 1 + o(1).
6 log n
Hence, if p0 = n then
Z p0
E Ln = (E κ(G p ) − 1)d p + o(1)
p=0
Z p0
= E κ(G p )d p + o(1).
p=0

Write
(log n)2 (log n)2
κ(G p ) = ∑ Ak + ∑ Bk +C,
k=1 k=1
390 Weighted Graphs

where Ak stands for the number of components which are k vertex trees, Bk is
the number of k vertex components which are not trees and, finally, C denotes
the number of components on at least (log n)2 vertices. Then, for 1 ≤ k ≤
(log n)2 and p ≤ p0 ,
 
n k−2 k−1 k
E Ak = k p (1 − p)k(n−k)+(2)−k+1
k
kk−2 k−1
= (1 + o(1))nk p (1 − p)kn .
k!

   
n k−2 k k
E Bk ≤ k p (1 − p)k(n−k)
k 2
≤ (1 + o(1))(npe1−np )k
≤ 1 + o(1).

n
C≤ .
(log n)2

Hence
6 log n (log n)2
6 log n
Z
n
∑ E Bk d p ≤ (log n)2 (1 + o(1)) = o(1),
p=0 k=1 n

and
6 log n
6 log n n
Z
n
Cd p ≤ = o(1).
p=0 n (log n)2

So
(log n)2 6 log n
kk−2
Z
n
E Ln = o(1) + (1 + o(1)) ∑ nk pk−1 (1 − p)kn d p.
k=1 k! p=0

But
(log n)2
kk−2
Z 1
∑ nk pk−1 (1 − p)kn d p
k=1 k! p= 6 log
n
n

(log n)2
kk−2
Z 1
≤ ∑ nk n−6k d p
k=1 k! p= 6 log
n
n

= o(1).
 19.2 Shortest Paths 391

Therefore
(log n)2
kk−2
Z 1
E Ln = o(1) + (1 + o(1)) ∑ nk pk−1 (1 − p)kn d p
k=1 k! p=0

(log n)2
kk−2 (k − 1)!(kn))!
= o(1) + (1 + o(1)) ∑ nk
k=1 k! (k(n + 1))!
(log n)2 k
1
= o(1) + (1 + o(1)) ∑ nk kk−3 ∏
k=1 i=1 kn +i
(log n)2
1
= o(1) + (1 + o(1)) ∑
k=1 k3
∞
1
= o(1) + (1 + o(1)) ∑ 3
.
k=1 k

One can obtain the same result if the uniform [0, 1] random variable is re-
placed by any random non-negative random variable with distribution F hav-
ing a derivative equal to one at the origin, e.g. an exponential variable with
mean one, see Steele [714].

19.2 Shortest Paths

Let the edges of the complete graph Kn on [n] be given independent lengths
Xe , e ∈ [n]2 . Here Xe is exponentially distributed with mean 1. The following
theorem was proved by Janson [432]:
Theorem 19.2 Let Xi j be the distance from vertex i to vertex j in the complete
graph with edge weights independent EXP(1) random variables. Then, for
every ε > 0, as n → ∞,
(i) For any fixed i, j,
  
 Xi j 
P 
 − 1 ≥ ε → 0.

log n/n
(ii) For any fixed i,
  
 max j Xi j 
P 
 − 2 ≥ ε → 0.

log n/n
(iii)
  
 maxi, j Xi j 
P 
 − 3 ≥ ε → 0.

log n/n
 392 Weighted Graphs

Proof First, recall the following two properties of the exponential X:

(P1) P(X > α + β |X > α) = P(X > β ).

(P2) If X1 , X2 , . . . , Xm are independent EXP(1) exponential random variables then
min{X1 , X2 , . . . , Xm } is an exponential with mean 1/m.

Suppose that we want to find shortest paths from a vertex s to all other vertices
in a digraph with non-negative arc-lengths. Recall Dijkstra’s algorithm. After
several iterations there is a rooted tree T such that if v is a vertex of T then
the tree path from s to v is a shortest path. Let d(v) be its length. For x ∈ /T
let d(x) be the minimum length of a path P that goes from s to v to x where
v ∈ T and the sub-path of P that goes to v is the tree path from s to v. If
d(y) = min {d(x) : x ∈ / T } then d(y) is the length of a shortest path from s to y
and y can be added to the tree.
Suppose that vertices are added to the tree in the order v1 , v2 , . . . , vn and that
Y j = dist(v1 , v j ) for j = 1, 2, . . . , n. It follows from property P1 that

Yk+1 = min [Yi + Xvi ,v ] = Yk + Ek

i=1,2,...,k
v6=v1 ,...,vk

1
where Ek is exponential with mean k(n−k) and is independent of Yk .
This is because Xvi ,v j is distributed as an independent exponential X condi-
tioned on X ≥ Yk −Yi . Hence
n−1
1
EYn = ∑ k(n − k)
k=1
n−1  
1 1 1
= ∑ k + n−k
n k=1
2 n−1 1
= ∑k
n k=1
2 log n
= + O(n−1 ).
n
Also, from the independence of Ek ,Yk ,
n−1
VarYn = ∑ Var Ek
k=1
n−1  2
1
=∑
k=1 k(n − k)
n/2  2
1
≤2∑
k=1 k(n − k)
 19.2 Shortest Paths 393

8 n/2 1
≤ ∑ k2
n2 k=1
= O(n−2 )
and we can use the Chebyshev inequality (21.3) to prove (ii).
Now fix j = 2. Then if i is defined by vi = 2, we see that i is uniform over
{2, 3, . . . , n}. So
1 n i−1 1
E X1,2 = ∑ ∑ k(n − k)
n − 1 i=2 k=1

1 n−1 n − k
= ∑ k(n − k)
n − 1 k=1
1 n−1 1
= ∑k
n − 1 k=1
log n
= + O(n−1 ).
n
For the variance of X1,2 we have
X1,2 = δ2Y2 + δ3Y3 + · · · + δnYn ,
where
1
δi ∈ {0, 1}; δ2 + δ3 + · · · + δn = 1; P(δi = 1) = .
n−1
n
Var X1,2 = ∑ Var(δiYi ) + ∑ Cov(δiYi , δ jY j )
i=2 i6= j
n
≤ ∑ Var(δiYi ).
i=2

The last inequality holds since

Cov(δiYi , δ jY j ) = E(δiYi δ jY j ) − E(δiYi ) E(δ jY j )

= − E(δiYi ) E(δ jY j ) ≤ 0.
So
n
Var X1,2 ≤ ∑ Var(δiYi )
i=2
n 2
1 i−1

1
≤∑ ∑
i=2 n − 1 k=1 k(n − k)
= O(n−2 ).
394 Weighted Graphs

We can now use the Chebyshev inequality.

We turn now to proving (iii). We beginnwith a lower bound. o Let
 (1−ε) log n
Yi = min Xi, j : i 6= j ∈ [n] . Let A = i : Yi ≥ n . Then we have that
for i ∈ [n],
 
(1 − ε) log n
Pr(i ∈ A) = exp −(n − 1) = n−1+ε+o(1) . (19.2)
n
An application of the Chebyshev inequality shows that |A| ≈ nε+o(1) w.h.p.
Now the expected number of paths from a1 ∈ A to a2 ∈ A of length at most
(3−2ε) log n
n can be bounded by

log2 n
n2ε+o(1) × n2 × n−3ε+o(1) × = n−ε+o(1) . (19.3)
n2
Explanation for (19.3): The first factor n2ε+o(1) is the expected number of
pairs of vertices a1 , a2 ∈ A. The second factor is a bound on the number of
choices b1 , b2 for the neighbors of a1 , a2 on the path. The third factor F3 is a
bound on the expected number of paths of length at most α log n
n
from b1 to b2 ,
α = 1 − 3ε. This factor comes from
 `+1
` α log n 1
F3 ≤ ∑ n .
`≥0 n (` + 1)!

Here ` is the number of internal vertices on the path. There will be at most n`
choices for the sequence of vertices on the path. We then use the fact that the
exponential mean one random variable stochastically dominates the uniform
[0, 1] random variable U. The final two factors are the probability that the sum
of ` + 1 independent copies of U sum to at most α log n
n . Continuing we have
!`
α log n e1+o(1) α log n
F3 ≤ ∑
`≥0 n(` + 1) `
!
10 log n
α log n −`
≤ ∑ α+o(1)
n + ∑ e = n−1+α+o(1) .
n `=0 `>10 log n

The final factor in (19.3) is a bound on the probability that Xa1 b1 + Xa2 b2 ≤
(2+ε) log n
n . For this we use the fact that Xai bi , i = 1, 2 is distributed as (1−ε)n log n +
Ei where E1 , E2 are independent exponential mean one. Now Pr(E1 +E2 ≤ t) ≤
(1 − e−t )2 ≤ t 2 and taking t = 3ε log n
n
justifies the final factor of (19.3).
It follows from (19.3) that w.h.p. the shortest distance between a pair of vertices
in A is at least (3−2ε) n
log n
w.h.p., completing our proof of the lower bound in
(iii).
 19.3 Minimum Weight Assignment 395

We now consider the upper bound. Let now Y1 = dk3 where k3 = n1/2 log n. For
t < 1 − 1+o(1
n we have that
( )!
k3 k 
(1 + o(1))t −1

tnY1 tn
E(e ) = E exp ∑ = ∏ 1−
i=1 i(n − i) i=1 i

Then for any α > 0 and for we have

k3 
(1 + o(1))t −1
  
α log n
Pr Y1 ≥ ≤ E(etnY1 −tα log n ) ≤ e−tα log n ∏ 1 −
n i=1 i
(  )
k3   
(1 + o(1))t 1 1
= e−tα log n exp ∑ +O 2 = O(1)×exp + o(1) − α t log n .
i=1 i i 2

It follows, on taking α = 3/2 + o(1) that w.h.p.

(3 + o(1)) log n
Yj ≤ for all j ∈ [n].
2n
Letting T j be the set corresponding to Sk3 when we execute Dijkstra’s algoritm
starting at j, then we have that for j 6= k where T j ∩ Tk = 0,
/
   2 
log n k3 log n
Pr 6 ∃e ∈ (T j : Tk ) : Xe ≤ ≤ exp −
n n
2n
= e−(2+o(1)) log = o(n−2 )

and this is enough to complete the proof of (iii).

We can as for Spanning Trees, replace the exponential random variables by
random variables that behave like the exponential close to the origin. The paper
of Janson [432] allows for any random variable X satisfying P(X ≤ t) = t +o(t)
as t → 0.

19.3 Minimum Weight Assignment

Consider the complete bipartite graph Kn,n and suppose that its edges are as-
signed independent exponentially distributed weights, with rate 1. (The rate
of an exponential variable is one over its mean). Denote the minimum total
weight of a perfect matching in Kn,n by Cn . Aldous [13], [16] proved that
1
limn→∞ ECn = ζ (2) = ∑∞ k=1 k2 . The following theorem was conjectured by
Parisi [634]. It was proved independently by Linusson and Wästlund [543]
and Nair, Prabhakar and Sharma [616]. The proof given here is from Wästlund
[743].
396 Weighted Graphs

Theorem 19.3
n
1 1 1 1 1
ECn = ∑ k2 = 1 + 4 + 9 + 16 + · · · + n2 (19.4)
k=1

From the above theorem we immediately get the following corollary, first
proved by Aldous [16].

Corollary 19.4
∞
1 π2
lim ECn = ζ (2) = ∑ k2 = = 1.6449 · · ·
n→∞
k=1 6

Let E(λ ) denote an exponential random variable of rate λ i.e. Pr(E(λ ) ≥ x) =

e−lx . Consider the complete bipartite graph Kn,n , with bipartition (A, B), where
A = {a1 , a2 , . . . , an } and B = {b1 , b2 , . . . , bn }, and with edge weights which
are independent copies of E(1). We add a special vertex b∗ to B, with edges
to all n vertices of A. Each edge adjacent to b∗ is assigned an E(λ ) weight
independently, λ > 0.
For r ≥ 1 we let Mr be the minimum weight matching of Ar = {a1 , a2 , . . . , ar }
into B and Mr∗ be the minimum weight matching of Ar into B∗ = B ∪ {b∗ }. (As
λ → 0 it becomes increasingly unlikely that any of the extra edges are actually
used in the minimum weight matching.) We denote this matching by Mr∗ and
we let B∗r denote the corresponding set of vertices of B∗ that are covered by
Mr∗ . We let C(n, r) denote the weight of Mr .
Define P(n, r) as the normalized probability that b∗ participates in Mr∗ , i.e.
Pr(b∗ ∈ B∗r )
P(n, r) = lim . (19.5)
λ →0 λ
Its importance lies in the following lemma:
Lemma 19.5
P(n, r)
E(C(n, r) −C(n, r − 1)) = . (19.6)
r
Proof Choose i randomly from [r] and let Bbi ⊆ Br be the B-vertices in the
minimum weight matching of (Ar \ {ai }) into B∗ . Let X = C(n, r) and let Y =
C(n, r − 1). Let wi be the weight of the edge (ai , b∗ ), and let Ii denote the
indicator variable for the event that the minimum weight of an Ar matching
 19.3 Minimum Weight Assignment 397

that contains this edge is smaller than the minimum weight of an Ar matching
that does not use b∗ . We can see that Ii is the indicator variable for the event
{Yi + wi < X}, where Yi is the minimum weight of a matching from Ar \ {ai }
to B. Indeed, if (ai , b∗ ) ∈ Mr∗ then wi < X −Yi . Conversely, if wi < X −Yi and
no other edge from b∗ has weight smaller than X −Yi , then (ai , b∗ ) ∈ Mr∗ , and
when λ → 0, the probability that there are two distinct edges from b∗ of weight
smaller than X − Yi is of order O(λ 2 ). Indeed, let F denote the existence of
two distinct edges from b∗ of weight smaller than X and let Fi, j denote the
event that (ai , b∗ ) and a j , b∗ ) both have weight smaller than X.
Then,
Pr(F ) ≤ n2 EX (max Pr(Fi, j | X)) = n2 E((1−e−λ X )2 ) ≤ n2 λ 2 E(X 2 ), (19.7)
i, j

and since E(X 2 ) is finite and independent of λ , this is O(λ 2 ).

Note that Y and Yi have the same distribution. They are both equal to the min-
imum weight of a matching of a random (r − 1)-set of A into B. As a conse-
quence, E(Y ) = E(Yi ) = 1r ∑ j∈Ar E(Y j ). Since wi is E(λ ) distributed, as λ → 0
we have from (19.7) that
!
1
P(n, r) = lim
λ →0 λ j∈A
∑ Pr(w j < X −Y j ) + O(λ ) =
r
!
1  
lim E
λ →0
∑ 1 − e−λ (X−Y j ) = ∑ E(X −Yi ) = r E(X −Y ).
λ j∈A r j∈Ar

We now proceed to estimate P(n, r). Fix r and assume that b∗ ∈ / B∗r−1 . Suppose
∗ ∗
that Mr is obtained from Mr−1 by finding an augmenting path P = (ar , . . . , aσ , bτ )
from ar to B \ Br−1 of minimum additional weight. We condition on (i) σ , (ii)
the lengths σ , b j ), b j ∈ B \ Br−1 and
 of all edges other than (a ∗ will
(iii) min w(aσ , b j ) : b j ∈ B \ Br−1 . With this conditioning Mr−1 = Mr−1
0
be fixed and so will P = (ar , . . . , aσ ). We can now use the following fact: Let
X1 , X2 , . . . , XM be independent exponential random variables of rates λ1 , λ2 , . . . , λM .
Then the probability that Xi is the smallest of them is λi /(λ1 + λ2 + · · · + λM ).
Furthermore, the probability stays the same if we condition on the value of
min {X1 , X2 , . . . , XM }. Thus
λ
Pr(b∗ ∈ B∗r | b∗ ∈
/ B∗r−1 ) = .
n−r+1+λ
Lemma 19.6
1 1 1
P(n, r) = + +···+ . (19.8)
n n−1 n−r+1
398 Weighted Graphs

Proof
 
−1 ∗ n n−1 n−r+1
lim λ Pr(b ∈ B∗r ) =
lim λ −1
1− · ···
λ →0 λ →0 n+λ n−1+λ n−r+1+λ
 −1  −1 !
λ λ
= lim λ −1 1 − 1 + ··· 1+
λ →0 n n−r+1
  
1 1 1
= lim λ −1 + +···+ λ + O(λ 2 )
λ →0 n n−1 n−r+1
1 1 1
= + +···+ . (19.9)
n n−1 n−r+1

It follows from Lemmas 19.5 and 19.6 that

n r
1 1
ECn = ∑ r ∑ n−i+1.
r=1 i=1

It follows that

E(Cn+1 −Cn ) =
1 n+1 n
1 r
 
1 1 1
= +
∑ n−i+2 ∑ r ∑ n−i+2 n−i+1 −
n + 1 i=1 r=1 i=1
n+1 n  
1 1 1 1 1
= + −
∑ (n + 1)(n − i + 2) ∑ r n + 1 n − r + 1 ,
+
(n + 1)2 i=2 r=1
1
= . (19.10)
(n + 1)2
E(C1 ) = 1 and so (19.4) follows from (19.10).

19.4 Exercises
19.4.1 Suppose that the edges of the complete bipartite graph Kn,n are given
(b)
independent uniform [0, 1] edge weights. Show that if Ln is the length
of the minimum spanning tree, then
(b)
lim E Ln = 2ζ (3).
n→∞

19.4.2 Let G = Kαn,β n be the complete unbalanced bipartite graph with bipar-
tition sizes αn, β n. Suppose that the edges of G are given independent
 19.4 Exercises 399
(b)
uniform [0, 1] edge weights. Show that if Ln is the length of the mini-
mum spanning tree, then
i −1 i −1
(b) 1 (i1 + i2 − 1)! γ i1 i12 i21
lim E Ln = γ + + ∑ ,
n→∞ γ i1 ≥1,i2 ≥1 i1 !i2 ! (i1 + γi2 )i1 +i2

where γ = α/β .
19.4.3 Tighten Theorem 19.1 and prove that
 
1
E Ln = ζ (3) + O .
n

19.4.4 Suppose that the edges of Kn are given independent uniform [0, 1] edge
weights. Let Zk denote the minimum total edge cost of the union of k
edge-disjoint spanning trees. Show that limk→∞ Zk /k2 = 1.
19.4.5 Suppose that the edges of Gn,p where 0 < p ≤ 1 is a constant, are given
exponentially distributed weights with rate 1. Show that if Xi j is the
shortest distance from i to j then
1 For any fixed i, j,
  
 Xi j 1 
P 
 − ≥ ε → 0.
log n/n p 

2
  
 max j Xi j 2 
P 
 −  ≥ ε → 0.

log n/n p

19.4.6 The quadratic assignment problem is to

Minimise
Z = ∑ni, j,p,q=1 ai j pq xip x jq
Sub ject to
∑ni=1 xip = 1 p = 1, 2, . . . , n

∑np=1 xip = 1 i = 1, 2, . . . , n

xip = 0/1.

Suppose now that the ai j pq are independent uniform [0, 1] random vari-
ables. Show that w.h.p. Zmin ≈ Zmax where Zmin (resp. Zmax ) denotes
the minimum (resp. maximum) value of Z, subject to the assignment
constraints.
 400 Weighted Graphs

19.4.7 The 0/1 knapsack problem is to

Maximise
Z = ∑ni=1 ai xi
Sub ject to
∑ni=1 bi xi ≤ L

xi = 0/1 for i = 1, 2, . . . , n.

Suppose that the (ai , bi ) are chosen independently and uniformly from
[0, 1]2 and that L = αn. Show that w.h.p. the maximum value of Z,
Zmax , satisfies
 1/2

 2
 α n
α ≤ 14 .
(8α−8α 2 −1)n 1 1
Zmax ≈ 2 4 ≤α ≤ 2
.

n
 1
2 α≥ 2

19.4.8 Suppose that X1 , X2 , . . . , Xn are points chosen independently and uni-

formly at random from [0, 1]2 . Let Zn denote the total Euclidean length
of the shortest tour (Hamilton cycle) through each point. Show that
there exist constants c1 , c2 such that c1 n1/2 ≤ Zn ≤ c2 n1/2 w.h.p.
19.4.9 Prove equation (19.11) below. See the preceding paragraph to explain
the 3-dimensional assignment problem. The implied lower bound is
easy.
For the upper bound let X = V1 ×V2 ×V3 and let
H = {(it , jt , kt ),t = 1, 2, . . . , n : V1 = {i1 , . . . , in } etc.}. Let p = C/n2
for large C and let W0 ,W1 , . . . ,Wi , . . . be as in Claim 9 of Section 14.3.
Show that w.h.p. each Wi contributes O(1/n) × (1 − γ)i to the weight
of the final assignment.
19.4.10 Prove equation (19.12) below.

19.5 Notes
Shortest paths
There have been some strengthenings and generalisations of Theorem 19.2. For
example, Bhamidi and van der Hofstad [87] have found the (random) second-
order term in (i), i.e., convergence in distribution with the correct norming.
They have also studied the number of edges in the shortest path.
 19.5 Notes 401

Spanning trees
Beveridge, Frieze and McDiarmid [86] considered the length of the minimum
spanning tree in regular graphs other than complete graphs. For graphs G of
large degree r they proved that the length MST (G) of an n-vertex randomly
edge weighted graph G satisfies MST (G) = nr (ζ (3) + or (1)) w.h.p., provided
some mild expansion condition holds. For r regular graphs of large girth g they
proved that if
∞
r 1
cr = ∑ ,
(r − 1)2 k=1 k(k + ρ)(k + 2ρ)

then w.h.p. |MST (G) − cr n| ≤ 3n2g .

Frieze, Ruszinko and Thoma [355] replaced expansion in [86] by connectivity
and in addition proved that MST (G) ≤ nr (ζ (3) + 1 + or (1)) for any r-regular
graph.
Cooper, Frieze, Ince, Janson and Spencer [222] show that Theorem 19.1 can
be improved to yield E Ln = ζ (3) + cn1 + c2n+o(1)
4/3 for explicit constants c1 , c2 .
Bollobás, Gamarnik, Riordan and Sudakov [145] considered the
Steiner Tree problem on Kn with independent random edge weights, Xe , e ∈
E(Kn ). Here they assume that the Xe have the same distribution X ≥ 0 where
P(X ≤ x) = x + o(x) as x → 0. The main result is that if one fixes k = o(n)
vertices then w.h.p. the minimum length W of a sub-tree of Kn that includes
these k points satisfies W ≈ k−1 n
n log k .
Angel, Flaxman and Wilson [36] considered the minimum length of a spanning
tree of Kn that has a fixed root and bounded depth k. The edges weights Xe are
independent exponential mean one. They prove that if k ≥ log2 log n + ω(1)
then w.h.p. the minimum length tends to ζ (3) as in the unbounded case. On the
other hand, if k ≤ log2 log n−ω(1) then w.h.p. the weight is doubly exponential
in log2 log n − k. They also considered bounded depth Steiner trees.
Using Talagrand’s inequality, McDiarmid [575] proved that for any real t > 0
we have P(|Ln − ζ (3)| ≥ t) ≤ e−δ1 n where δ1 = δ2 (t). Flaxman [307] proved
that P(|Ln − ζ (3)| ≤ ε) ≥ e−δ2 n where δ1 = δ2 (ε).

Assignment problem
Walkup [742] proved if the weights of edges are independent uniform [0, 1]
then ECn ≤ 3 (see (19.3)) and later Karp [483] proved that ECn ≤ 2. Dyer,
Frieze and McDiarmid [272] adapted Karp’s proof to something more general:
402 Weighted Graphs

Let Z be the optimum value to the linear program:

n
Minimise ∑ c j x j , subject to x ∈ P = {x ∈ Rn : Ax = b, x ≥ 0} ,
j=1

where A is an m×n matrix. As a special case of [272], we have that if c1 , c2 , . . . ,

cn are independent uniform [0, 1] random variables and x∗ is any member of P,
then E(Z) ≤ m(max j x∗j ). Karp’s result can easily be deduced from this.
The assignment problem can be generalized to multi-dimensional versions: We
replace the complete bipartite graph Kn,n by the complete k-partite hypergraph
(k)
Kn with vertex partition V = V1 tV2 t · · · tVk where each Vi is of size n. We
(k)
give each edge of Kn an independent exponential mean one value. Assume for
example that k = 3. In one version of the 3-dimensional assignment problem
we ask for a minimum weight collection of hyper-edges such that each vertex
v ∈ V appears in exactly one edge. The optimal total weight Z of this collection
satisfies  
1
Z=Θ w.h.p. (19.11)
n
(The upper bound uses the result of [460], see Section 14.3).
Frieze and Sorkin [356] give an O(n3 ) algorithm that w.h.p. finds a solution of
1
value n1−o(1) .
In another version of the 3-dimensional assignment problem we ask for a min-
imum weight collection of hyper-edges such that each pair of vertices v, w ∈ V
from different sets in the partition appear in exactly one edge. The optimal total
weight Z of this collection satisfies
Ω(n) ≤ Z ≤ O(n log n) w.h.p. (19.12)
(The upper bound uses the result of [272] to greedily solve a sequence of re-
stricted assignment problems).
 20
Brief notes on uncovered topics

There are several topics that we have not been able to cover and that might be
of interest to the reader. For these topics, we provide some short synopses and
some references that the reader may find useful.

Contiguity
Suppose that we have two sequences of probability models on graphs G1,n , G2,n
on the set of graphs with vertex set [n]. We say that the two sequences are
contiguous if for any sequence of events An we have

lim P(G1,n ∈ An ) = 0 ⇔ lim P(G2,n ∈ An ) = 0.

n→∞ n→∞

This for example, is useful for us, if we want to see what happens w.h.p. in
the model G1,n , but find it easier to work with G2,n . In this context, Gn,p and
Gn,m=n2 p/2 are almost contiguous.
Interest in this notion in random graphs was stimulated by the results of Robin-
son and Wormald [668], [671] that random r-regular graphs, r ≥ 3, r = O(1)
are Hamiltonian. As a result, we find that other non-uniform models of ran-
dom regular graphs are contiguous to Gn,r e.g. the union rMn of r random
perfect matchings when n is even. (There is an implicit conditioning on rMn
being simple here). The most general result in this line is given by Wormald
[751], improving on earlier results of Janson [431] and Molloy, Robalewska,
Robinson and Wormald [596] and Kim and Wormald [494]. Suppose that
r = 2 j + ∑r−1
i=1 iki , with all terms non-negative. Then Gn,r is contiguous to the
sum jHn + ∑r−1 i=1 ki Gn,i , where n is restricted to even integers if ki 6= 0 for any
odd i. Here jHn is the union of j edge disjoint Hamilton cycles etc.
Chapter 8 of [441] is devoted to this subject.

Edge Colored Random Graphs

Suppose that we color the edges of a graph G. A set of edges S is said to
be rainbow colored if each edge of S has a different color. Consider first the

403
404 Brief notes on uncovered topics

existence of a rainbow spanning tree. We consider the graph process where

the edges are randomly colored using q ≥ n − 1 colors. Let τa be the hitting
time for n − 1 colors to appear in the process and let τc be the hitting time for
connectivity and let τ ∗ = max {τa , τc }. Frieze and McKay [349] showed that
w.h.p. Gτ ∗ contains a rainbow spanning tree. This is clearly best possible. Bal,
Bennett, Frieze and Pralat [47] consider the case where each edge has a choice
of k random colors. This reduces τa , but the result still holds.
The existence of rainbow Hamilton cycles is different. The existence of a
rainbow spanning tree can be checked in polynomial time and this leads to
a simple criterion for non-existence. This is clearly not likely for Hamilton
cycles. Cooper and Frieze [212] proved that if m ≥ Kn log n and q ≥ Kn then
w.h.p. Gn,m contains a rainbow Hamilton cycle. This was improved to m ≥
1+o(1)
2 n log n and q ≥ (1 + o(1))n by Frieze and Loh [344]. Bal and Frieze [49]
show that if m ≥ Kn log n and q = n and n is even there is a rainbow Hamil-
ton cycle w.h.p. Ferber [296] removed the requirement that n be even. Bal and
Frieze also considered rainbow perfect matchings in k-uniform hypergraphs.
Janson and Wormald [444] considered random coloring’s of r-regular graphs.
They proved that if r ≥ 4, r = O(1) and the edges of Gn,2r are randomly col-
ored so that each color is used r times, then w.h.p. there is a rainbow Hamil-
ton cycle. Ferber, Kronenberg, Mousset and Shikhelman [299] give results on
packing rainbow structures such as Hamilton cycles. Ferber, Nenadov and Pe-
ter prove that if p  n−1/d (log n)1/d and H is a fixed graph of density at most
d then w.h.p. Gn,p contains a rainbow copy of H if it is randomly colored with
(1 + ε)|E(H)| colors, for any fixed ε > 0.
Cooper and Frieze [209] found the threshold for the following property: If
k = O(1) and Gn,m is arbitrarily edge colored so that no color is used more
than k times, then Gn,m contains a rainbow Hamilton cycle.

Games
Positional games can be considered to be a generalisation of the game of
“Noughts and Crosses” or “Tic-Tac-Toe”. There are two players A (Maker)
and B (Breaker) and in the context for this section, the board will be a graph
G. Each player in turn chooses an edge and at the end of the game, the winner
is determined by the partition of the edges claimed by the players. As a typical
example, in the connectivity game, player A is trying to ensure that the edges
she collects contain a spanning tree of G and player B is trying to prevent this.
See Chvátal and Erdős [190] for one of the earliest papers on the subject and
books by Beck [65] and Hefetz, Krivelevich, Stojaković and Szabó [411]. Most
 Brief notes on uncovered topics 405

of the analyses have considered G = Kn and to make the problem interesting

[190] introduced the notion of bias. Thus in the connectivity game, player B is
allowed to collect b edges for each edge of A. Now the question becomes what
is the largest value of b for which A has a winning strategy. There is a strik-
ing though somewhat mysterious connection between the optimal values of b
for various games and thresholds for associated properties in random graphs.
For example in the connectivity game, the threshold bias b ≈ logn n i.e. player A
collects about 12 n log n edges, see Gebauer and Szabó [368]. Another example
is the biased H-game where Maker wins if she can create a copy of some fixed
graph H with at least two adjacent edges. The optimal threshold bias b for this
1/m (H)


game is of order Θ n 2 , Bednarska and Łuczak [66]. For sufficiently
small constant c > 0, if b ≤ cn1/m2 (H) , then Maker can create Θ(EXH ) copies
of H in Kn , where XH is the number of copies of H in Gn,1/b . Furthermore, if
Maker plays randomly, she achieves this goal w.h.p.
Recently Stojaković and Szabó [718] began research on random boards i.e.
where G is a random graph. Ben-Shimon, Ferber, Hefetz and Krivelevich [78]
prove a hitting time result for the b = 1 Hamilton cycle game on the graph
process. Assuming that player A wants to build a Hamilton cycle and player
B starts first, player A will have a winning strategy in Gm iff m ≥ m∗4 . This is
best possible. Biased Hamiltonicity games on Gn,p were considered in Ferber,
Glebov, Krivelevich and Naor [297] where it was shown that for p  logn n , the
np
threshold bias bHAM satisfies bHAM ≈ log n w.h.p. The H-game where A wins if
she can create a copy of some fixed graph H was first studied by Stojakovic and
Szabo [718] in the case of H is a clique on k vertices. This was strengthened
by Müller and Stojaković [611]. They show that if p ≤ cn−2/(k+1) , then w.h.p.
B can win this game. For p ≥ Cn−2/(k+1) one can use the results of [672] to
argue that A wins w.h.p. This result was generalised to arbitrary graphs H
(satisfying certain mild conditions) by Nenadov, Steger and Stojaković [622]
where they showed that the threshold is where one would expect it to be - at
the 2-density of H. As we have seen there are other models of random graphs
and Beveridge, Dudek, Frieze, Müller and Stojaković [85] studied these games
on random geometric graphs.
The game chromatic number χg (G) of a graph G can be defined as follows.
Once again there are two players A,B and they take it in turns to properly
color vertices of G with one of q colors. Thus if {u, v} is an edge and u is
colored with color c and v is uncolored at the start of any turn, then v may not
be colored with c by either player. The goal of A is to ensure that the game
ends with every vertex colored and the goal of B is to prevent this by using all
q colors in the neighborhood of some uncolored vertex. The game chromatic
406 Brief notes on uncovered topics

number is the minimum q for which A can win. For a survey on results on
this parameter see Bartnicki, Grytczuk, Kierstead and and Zhu [63]. Bohman,
Frieze and Sudakov [119] studied χg for dense random graphs and proved that
for such graphs, χg is within a constant factor of the chromatic number. Keusch
and Steger [491] proved that this factor is asymptotically equal to two. Frieze,
Haber and Lavrov [337] extended the results of [119] to sparse random graphs.

Graph Searching
Cops and Robbers
A collection of cops are placed on the vertices of a graph by player C and then
a robber is placed on a vertex by player R. The players take turns. C can move
all cops to a neighboring vertex and R can move the robber. The cop number
of a graph is them minimum number of cops needed so that C can win. The
basic rule being that if there is a cop occupying the same vertex as the robber,
then C wins. Łuczak and Pralat [558] proved a remarkable “zigzag” theorem
giving the cop number of a random graph. This number being nα where α =
α(p) follows a saw-toothed curve. Pralat and Wormald [650] proved that the
cop number of the random regular graph Gn,r is O(n1/2 ). It is worth noting
that Meyniel has conjectured O(n1/2 ) as a bound on the cop number of any
connected n-vertex graph. There are many variations on this game and the
reader is referred to the monograph by Bonato and Pralat [153].

Graph Cleaning
Initially, every edge and vertex of a graph G is dirty, and a fixed number of
brushes start on a set of vertices. At each time-step, a vertex v and all its
incident edges that are dirty may be cleaned if there are at least as many
brushes on v as there are incident dirty edges. When a vertex is cleaned, ev-
ery incident dirty edge is traversed (that is, cleaned) by one and only one
brush, and brushes cannot traverse a clean edge. The brush number b(G) is
the minimum number of brushes needed to clean G. Pralat [651], [652] proved
−2d
that w.h.p.b(Gn,p ) ≈ 1−e4  n for p = dn where d < 1 and w.h.p. b(Gn,p ) ≤
−2d
(1 + o(1)) d + 1 − 1−e2d n
for d > 1. For the random d-regular graph Gn,d ,
4
 
23/2
Alon, Pralat and Wormald [24] proved that w.h.p. b(Gn,d ) ≥ dn
4 1 − d 1/2 .
 Brief notes on uncovered topics 407

Acquaintance Time
Let G = (V, E) be a finite connected graph. We start the process by placing
one agent on each vertex of G. Every pair of agents sharing an edge are de-
clared to be acquainted, and remain so throughout the process. In each round
of the process, we choose some matching M in G. The matching M need not be
maximal; perhaps it is a single edge. For each edge of M, we swap the agents
occupying its endpoints, which may cause more agents to become acquainted.
We may view the process as a graph searching game with one player, where
the player’s strategy consists of a sequence of matchings which allow all agents
to become acquainted. Some strategies may be better than others, which leads
to a graph optimisation parameter. The acquaintance time of G, denoted by
A (G), is the minimum number of rounds required for all agents to become
acquainted with one another. The parameter A (G) was introduced
 2 by
 Ben-
n log log n
jamini, Shinkar and Tsur [70], who showed that A (G) = O log n for an
n vertex graph. The log log n factor was removed by Kinnersley, Mitsche  and
log n
Pralat [496]. The paper [496] also showed that w.h.p. A (Gn,p ) = O p for
(1+ε) log n
n ≤ p ≤ 1 − ε. The lower bound herewas 
relaxed to np − log n → ∞ in
log n
Dudek and Pralat [264]. A lower bound, Ω p for Gn,p and p ≥ n−1/2+ε
was proved in [496].

H-free process
In an early attempt to estimate the Ramsey number R(3,t), Erdős, Suen and
Winkler [287] considered the following process for generating a triangle free
graph. Let e1 , e2 , . . . , eN , N = n2 be a random ordering of the complete graph

Kn . Let P be a graph property e.g. being triangle free. We generate a sequence

of random graphs Γ0 , Γ1 , . . . , ΓN where Γi+1 = Γi + ei+1 if adding ei+1 does not
destroy P, otherwise Γi+1 = Γi . In this way we can generate a random graph
that is guaranteed to have property P.
For P is “bipartite” they show in [287] that ΓN has expected size greater than
(n2 − n)/4. When P is “triangle free” they show that w.h.p. that ΓN has size
Ω(n3/2 ) w.h.p. Bollobás and Riordan [149] studied the general H-free process.
More recently, Bohman [111] showed in the case of the triangle free process,
that w.h.p. ΓN has size Θ(n3/2 (log n)1/2  ).2 This
 provides an alternative proof
t
to that of Kim [492] that R(3,t) = Ω logt . He made use of a careful use
of the differential equations method, see Chapter 23. Bohman and Keevash
[120] and Fiz Pontiveros, Griffiths and Morris [306] have improved this result
408 Brief notes on uncovered topics

and shown that w.h.p. ΓN has size asymptotically equal to 2√1 2 n3/2 (log n)1/2 .
They also show that the independence number of Γ
N is bounded by (1 +
o(1))(2n log n)1/2 . This shows that R(3,t) > 14 − o(1) t 2 / logt.
Bohman, Mubayi and Picolleli [122] considered an r-uniform hypergraph ver-
sion. In particular they studied the T (r) -free process, where T (r) generalises a
triangle in a graph. It consists of S ∪ {ai } , i = 1, 2, . . . , r where |S| = r − 1 and
a further edge {a1 , a2 , . . . , ar }. Here hyperedges are randomly added one by
one until one is forced to create a copy of T r . They show that w.h.p. the final
1/r
hypergraph produced rhas  independence number O((n log n) ). This proves a
s (r) (r)
lower bound of Ω log s for the Ramsey number R(T , Ks ). The analysis is
based on a paper on the random greedy hypergraph independent set process by
Bennett and Bohman [75].
There has also been work on the related triangle removal process. Here we start
with Kn and repeatedly remove a random triangle until the graph is triangle
free. The main question is as to how many edges are there in the final triangle
free graph. A proof of a bound of O(n7/4+o(1) ) was outlined by Grable [388].
A simple proof of O(n7/4+o(1) ) was proved in Bohman, Frieze and Lubetzky
[116]. Furthermore, Bohman, Frieze and Lubetzky [117] have proved a tight
result of n3/2+o(1) for the number of edges left. This is close to the Θ(n3/2 )
bound conjectured by Bollobás and Erdős in 1990.
An earlier paper by Ruciński and Wormald [679] consider the d-process. Edges
were now rejected if they raised the degree of some vertex above d. Answering
a question of Erdős, they proved that the resulting graph was w.h.p. d-regular.

Logic and Random Graphs

The first order theory of graphs is a language in which one can describe some,
but certainly not all, properties of graphs. It can describe G has a triangle, but
not G is connected. Fagin [290] and Glebskii, Kogan, Liagonkii and Talanov
[378] proved that for any property A that can be described by a first order
sentence, limn→∞ P(Gn,1/2 ∈ A ) ∈ {0, 1}. We say that p = 1/2 obeys a 0-1
law. One does not need to restrict oneself to Gn,1/2 . Shelah and Spencer [702]
proved that if α is irrational then p = n−α also obeys a 0-1 law, while if α is
rational, then there are first order sentences such that limn→∞ P(Gn,1/2 ∈ A )
does not exist. See the book by Spencer [711] for much more on this subject.
 Brief notes on uncovered topics 409

Planarity
We have said very little about random planar graphs. This is partially because
there is no simple way of generating a random planar graph. The study begins
with the seminal work of Tutte [733], [734] on counting planar maps. The num-
ber of rooted maps on surfaces was found by Bender and Canfield [73]. The
size of the largest components were studied by Banderier, Flajolet, Schaeffer
and Soria [56].
When it comes to random labeled planar graphs, McDiarmid, Steger and Welsh
[580] showed that if pl(n) denotes the number of labeled planar graphs with
n vertices, then (pl(n)/n!)1/n tends to a limit γ as n → ∞. Osthus, Prömel and
Taraz [627] found an upper bound for γ, Bender, Gao and Wormald [74] found
a lower bound for γ. Finally, Giménez and Noy [375] proved that pl(n) ≈
cn−7/2 γ n n! for explicit values of c, γ.
Next let pl(n, m) denote the number of labelled planar graphs with n vertices
and m edges. Gerke, Schlatter, Steger and Taraz [371] proved that if 0 ≤ a ≤ 3
then (pl(n, an)/n!)1/n tends to a limit γa as n → ∞. Giménez and Noy [375]
showed that if 1 < a < 3 then pl(n, an) ≈ ca n−4 γan n!. Kang and Łuczak [472]
proved the existence of two critical ranges for the sizes of complex compo-
nents.

Planted Cliques, Cuts and Hamilton cycles

The question here is the following: Suppose that we plant an unusual object
into a random graph. Can someone else find it? One motivation being that if
finding the planted object is hard for someone who does not know where it
is planted, then this modified graph can be used as a signature. To make this
more precise, consider starting with Gn,1/2 , choosing an s-subset S of [n] and
then making S into a clique. Let the modified graph be denoted by Γ. Here
we assume that s  log n so that S should stand out. Can we find S, if we are
given Γ, but we are not told S. Kucera [528] proved that if s ≥ C(n log n)1/2
for a sufficiently large C then w.h.p. one can find S by looking at vertex de-
grees. Alon, Krivelevich and Sudakov [28] improved this to s = Ω(n1/2 ). They
show that the second eigenvector of the adjacency matrix of Γ contains enough
information so that w.h.p. S can be found. Frieze and Kannan [340] related
this to a problem involving optimisation of a tensor product. Recently, Feld-
man, Grigorescu, Reyzin, Vempala and Xiao [293] showed that a large class of
algorithms will fail w.h.p. if s ≤ n1/2−δ for some positive constant δ .
There has also been a considerable amount of research on planted cuts. Begin-
410 Brief notes on uncovered topics

ning with the paper of Bui, Chaudhuri, Leighton and Sipser [166] there have
been many papers that deal with the problem of finding a cut in a random
graph of unusual size. By this we mean that starting with Gn,p , someone se-
lects a partition of the vertex set into k ≥ 2 sets of large size and then alters
the edges between the subsets of the partition so that it is larger or smaller than
can be usually found in Gn,p . See Coja–Oghlan [191] for a recent paper with
many pertinent references.
As a final note on this subject of planted objects. Suppose that we start with
a Hamilton cycle C and then add a copy of Gn,p where p = nc to create Γ.
Broder, Frieze and Shamir [163] showed that if c is sufficiently large then
w.h.p. one can in polynomial time find a Hamilton cycle H in Γ. While H
may not necessarily be C, this rules out a simple use of Hamilton cycles for a
signature scheme.

Random Lifts
For a graph K, an n-lift G of K has vertex set V (K) × [n] where for each vertex
v ∈ V (K), {v} × [n] is called the fiber above v and will be denoted by Πv . The
edge set of a an n-lift G consists of a perfect matching between fibers Πu and
Πw for each edge {u, w} ∈ E(K). The set of n-lifts will be denoted Λn (K). In
a random n-lift, the matchings between fibers are chosen independently and
uniformly at random.
Lifts of graphs were introduced by Amit and Linial in [33] where they proved
that if K is a connected, simple graph with minimum degree δ ≥ 3, and G is a
random n-lift of K then G is δ (G)-connected w.h.p., where the asymptotics are
for n → ∞. They continued the study of random lifts in [34] where they proved
expansion properties of lifts. Together with Matoušek, they gave bounds on the
independence number and chromatic number of random lifts in [35]. Linial and
Rozenman [542] give a tight analysis for when a random n-lift has a perfect
matching. Greenhill, Janson and Ruciński [390] consider the number of perfect
matchings in a random lift.
Łuczak, Witkowski and Witkowski [561] proved that a random lift of H is
Hamiltonian w.h.p. if H has minimum degree at least 5 and contains two dis-
joint Hamiltonian cycles whose union is not a bipartite graph. Chebolu and
Frieze [177] considered a directed version of lifts and showed that a random
lift of the complete digraph K~ h is Hamiltonian w.h.p. provided h is sufficiently
large.
 Brief notes on uncovered topics 411

Random Simplicial Complexes

Linial and Meshulam [541] pioneered the extension of the analysis of Gn,p to
higher dimensional complexes. We are at the beginning of research in this area
and can look forward to exciting connections with Algebraic Topology. For
more details see the survey of Kahle [468].

Random Subgraphs of the n-cube

While most work on random graphs has been on random subgraphs of Kn , it
is true to say that has also been a good deal of work on random subgraphs of
the n-cube, Qn . This has vertex set Vn = {0, 1}n and an edge between x, y ∈ Vn
iff they differ in exactly one coordinate. To obtain a subgraph, we can either
randomly delete vertices with probability 1 − pv or edges with probability 1 −
pe or both. If only edges are deleted then the connectivity threshold is around
pe = 1/2, see Burtin [169] or Saposhenko [689], Erdős and Spencer [286]. If
only vertices are deleted then the connectivity threshold is around pv = 1/2,
see Saposhenko [690] or Weber [744]. If both edges and vertices and vertices
are deleted then the connectivity threshold is around pe pv = 1/2, see Dyer,
Frieze and Foulds [269].
Ajtai, Komlós and Szemerédi [8] showed that if pe = (1 + ε)/n then w.h.p.
there will be a unique giant component of order 2n . Their results were tight-
ened in Bollobás, Kohayakawa and Łuczak [146] where the case ε = o(1)
was considered. In further analysis, Bollobás, Kohayakawa and Łuczak [147]
proved the existence of giant components in the case pv = (1 + ε)/n.
The threshold for the existence of a prefect matching at around pe = 1/2 was
established by Bollobás [135]. The threshold for the existence of a Hamilton
cycle remains an open question.

Random Walks on Random Graphs

For a random walk, two of the most interesting parameters, are the mixing time
and the cover time.

Mixing Time
Generally speaking, the probability that a random walk is at a particular ver-
tex tends to a steady state probability deg(v)
2m . The mixing time is the time taken
412 Brief notes on uncovered topics

for the distribution k-step distribution to get to within variation distance 1/4,
say, of the steady state. Above the threshold for connectivity, the mixing time
of Gn,p is certainly O(log n) w.h.p. For sparser graphs, the accent has been
on finding the mixing time for a random walk on the giant component. Foun-
toulakis and Reed [318] and Benjamini, Kozma and Wormald [69] show that
w.h.p. the mixing time of a random walk on the giant component of Gn,p , p =
c/n, c > 1 is O((log n)2 ). Nachmias and Peres [614] showed that the mixing
time of the largest component of Gn,p , p = 1/n is in [εn, (1 − ε)n] with proba-
bility 1− p(ε) where p(ε) → 0 as ε → 0. Ding, Lubetzky and Peres [250] show
that mixing time for the emerging giant at p = (1+ε)/n where λ = ε 3 n → ∞ is
of order (n/λ )(log λ )2 . For random regular graphs, the mixing time is O(log n)
and Lubetzky and Sly [545] proved that the mixing time exhibits a cut-off
phenomenon i.e. the variation distance goes from near one to near zero very
rapidly.

Cover Time
The covertime CG of a graph G is the maximum over starting vertex of the ex-
pected time for a random walk to visit every vertex of G. For G = Gn,p with p =
c log n
n where c > 1, Jonasson [459] showed that w.h.p. CG = Θ(n log n).
Cooper
c
and Frieze [216] proved that CG ≈ A(c)n log n where A(c) = c log c−1 . Then
in [215] they showed that the cover time of a random r-regular graph is w.h.p.
r−1
asymptotic to r−2 n log n, for r ≥ 3. Then in a series of papers they established
the asymptotic cover time for preferential attachment graphs [216]; the giant
component of Gn,p , p = c/n, where c > 1 is constant [217]; random geomet-
ric graphs of dimension d ≥ 3, [218]; random directed graphs [219]; random
graphs with a fixed degree sequence [1], [224]; random hypergraphs [228]. The
asymptotic covertime of random geometric graphs for d = 2 is still unknown.
Avin and Ercal [42] prove that w.h.p. it is Θ(n log n). The paper [220] deals
with the structure of the subgraph Ht induced by the un-visited vertices in a
random walk on a random graph after t steps. It gives tight results on a phase
transition i.e. a point where H breaks up into small components. Cerny and
Teixeira [174] refined the result of [220] near the phase transition.

Stable Matching
In the stable matching problem we have a complete bipartite graph on vertex
sets A, B where A = {a1 , a2 , . . . , an } , B = {b1 , b2 , . . . , bn }. If we think of A as
a set of women and B as a set of men, then we refer to this as the stable
 Brief notes on uncovered topics 413

marriage problem. Each a ∈ A has a total ordering pa of B and each b ∈ B has

a total ordering pb of B. The problem is to find a perfect matching (ai , bi ), i =
1, 2, . . . , n such that there does not exist a pair i, j such that b j > bi in the order
pai and ai > b j in the order pb j . The existence of i, j leads to an unstable
matching. Gale and Shapley [363] proved that there is always a stable matching
and gave an algorithm for finding one. We focus on the case where pa , pb are
uniformly random for all a ∈ A, b ∈ B. Wilson [747] showed that the expected
number of proposals in a sequential version of the Gale-Shapley algorithm is
asymptotically equal to n log n. Knuth, Motwani and Pittel [499] studied the
likely number of stable husbands for an element of A ∪ B. I.e. they show that
w.h.p. there are constants c < C such that for a fixed a ∈ A there are between
c log n and C log n choices b ∈ B such that a and b are matched together in
some stable matching. The question of how many distinct stable matchings
there are likely to be was raised in Pittel [642] who showed that w.h.p. there
are at least n1/2−o(1) . More recently, Lennon and Pittel [535] show that there
are at least n log n with probability at least 0.45. Thus the precise growth rate
of the number of stable matchings is not clear at the moment. Pittel, Shepp
and Veklerov [646] considered the number Zn,m of a ∈ A that have exactly m
E(Z ) 1
choices of stable husband. They show that limn→∞ (log n)n,m m+1 = (m−1)! .

Universal graphs
A graph G is universal for a class of graphs H if G contains a copy of every
H ∈ H . In particular, let H (n, d) denote the set of graphs with vertex set [n]
and maximum degree at most d. One question that has concerned researchers,
is to find the threshold for Gn,p being universal for H (n, d). A counting ar-
gument shows that any H (n, d) universal graph has Ω(n2−2/d ) edges. For
random graphs this can be improved 2−2/(d+1) (log n)O(1) ). This is be-
 n  to Ω(n
cause to contain the union of d+1 disjoint copies of Kd+1 , all but at most
d vertices must lie in a copy of Kd+1 . This problem was first considered in
Alon, Capalbo, Kohayakawa, Rödl, Ruciński and Szemerédi [23]. Currently
the best upper bound on the value of p needed to make Gn,m H (n, d) univer-
sal is O(n2−1/d (log n)1/d ) in Dellamonica, Kohayakawa, Rödl, and Ruciński
[235]. Ferber, Nenadov and Peter [300] prove that if p  ∆8 n−1/2 log n then
Gn,p is universal for the set of trees with maximum degree ∆.
 P ART FOUR

TOOLS AND METHODS

 21
Moments

21.1 First and Second Moment Method

Lemma 21.1 (Markov’s Inequality) Let X be a non-negative random vari-

able. Then, for all t > 0,
EX
P(X ≥ t) ≤ .
t
Proof Let
(
1 if event A occurs,
IA =
0 otherwise.
Notice that
X = XI{X≥t} + XI{X<t} ≥ XI{X≥t} ≥ tI{X≥t} .
Hence,
EX ≥ tEI{X≥t} = t P(X ≥ t).

As an immediate corollary, we obtain

Lemma 21.2 (First Moment Method) Let X be a non-negative integer valued
random variable. Then
P(X > 0) ≤ EX.
Proof Put t = 1 in Markov’s inequality.
The following inequality is a simple consequence of Lemma 21.1.

Lemma 21.3 (Chebyshev Inequality) If X is a random variable with a finite

mean and variance, then, for t > 0,
Var X
P(|X − E X| ≥ t) ≤ .
t2

Proof
E(X − E X)2 Var X
P(|X − E X| ≥ t) = P((X − E X)2 ≥ t 2 ) ≤ = 2 .
t2 t

417
418 Moments

Throughout the book the following consequence of the Chebyshev inequality

plays a particularly important role.

Lemma 21.4 (Second Moment Method) If X is a non-negative integer valued

random variable then
Var X EX2
P(X = 0) ≤ = −1
(E X)2 (E X)2
Proof Set t = E X in the Chebyshev inequality. Then
Var X
P(X = 0) ≤ P(|X − E X| ≥ E X) ≤
(E X)2

Lemma 21.5 ((Strong) Second Moment Method) If X is a non-negative in-

teger valued random variable then
Var X (E X)2
P(X = 0) ≤ = 1 − .
EX2 EX2
Proof Notice that
X = X · I{X≥1} .
Then, by the Cauchy-Schwarz inequality,

2
(E X)2 = E(X · I{X≥1} ) ≤ E I{X≥1}
2
E X 2 = P(X ≥ 1) E X 2 .

The bound in Lemma 21.5 is stronger than the bound in Lemma 21.4, since
E X 2 ≥ (E X)2 . However, for many applications, these bounds are equally use-
ful since the Second Moment Method can be applied if
Var X
→ 0, (21.1)
(E X)2
or, equivalently,
EX2
→ 1, (21.2)
(E X)2
as n → ∞. In fact if (21.1) holds, then much more than P(X > 0) → 1 is true.
Note that
X 2
      2
Var X X X
= Var =E − E
(E X)2 EX EX EX
 21.1 First and Second Moment Method 419
 2
X
=E −1
EX

Hence
 2
X Var X
E −1 → 0 if → 0.
EX (E X)2
It simply means that
X L2
−→ 1. (21.3)
EX
In particular, it implies (as does the Chebyshev inequality) that
X P
−→ 1, (21.4)
EX
i.e., for every ε > 0,

P((1 − ε) E X < X < (1 + ε) E X) → 1. (21.5)

So, we can only apply the Second Moment Method, if the random variable
X has its distribution asymptotically concentrated at a single value (X can be
approximated by the non-random value E X, as stated at (21.3), (21.4) and
(21.5)).
We complete this section with another lower bound on the probability P(Xn ≥
1), when Xn is a sum of (asymptotically) negatively correlated indicators. No-
tice that in this case we do not need to compute the second moment of Xn .

Lemma 21.6 Let Xn = I1 + I2 + · · · + In , where {Ii }ni=1 be a collection of 0 − 1

random variables, such that

P(Ii = I j = 1) ≤ (1 + εn ) P(Ii = 1) P(I j = 1)

for i 6= j = 1, 2, . . . , n. Here εn → 0 as n → ∞. Then

1
P(Xn ≥ 1) ≥ .
1 + εn + 1/ E Xn
Proof By the (strong) second moment method (see Lemma 21.5)

(E Xn )2
P(Xn ≥ 1) ≥ .
E Xn2
Now
n n
E Xn2 = ∑ ∑ E(Ii I j )
i=1 j=1
420 Moments

≤ E Xn + (1 + εn ) ∑ E Ii E I j
i6= j
!
 n 2 n
2
= E Xn + (1 + εn ) ∑ E Ii − ∑ (E Ii )
i=1 i=1
2
≤ E Xn + (1 + εn )(E Xn ) .

21.2 Convergence of Moments

Let X be a random variable such that E |X|k < ∞, k ≥ 1, i.e., all k-th moments
E X k exist and are finite. Let the distribution of X be completely determined
by its moments. It means that all random variables with the same moments as
X have the same distribution as X. In particular, this is true when X has the
Normal or the Poisson distribution.
The method of moments provides a tool to prove the convergence in distribu-
tion of a sequence of random variables with finite moments (see Durrett [267]
for details).

Lemma 21.7 (Method of Moments) Let X be a random variable with prob-

ability distribution completely determined by its moments. If X1 , X2 , . . . Xn , . . .
are random variables with finite moments such that E Xnk → E X k as n → ∞, for
every integer k ≥ 1, then the sequence of random variables {Xn } converges in
D
distribution to random variable X, denoted as Xn → X.

The next result, which can be deduced from Theorem 21.7, provides a tool to
prove asymptotic Normality.

Corollary 21.8 Let X1 , X2 , . . . , Xn , . . . be a sequence of random variables with

finite moments and let a1 , a2 , . . . , an , . . . be a sequence of positive numbers, such
that
( (2m)!
m ak + o(akn ), when k = 2m, m ≥ 1,
E(Xn − E Xn )k = 2 m!k n
o(an ), when k = 2m − 1, m ≥ 2,

as n → ∞. Then
Xn − E Xn D Xn − E Xn D
→ Z, and X̃n = √ → Z,
an Var Xn
where Z is a random variable with the standard Normal distribution N(0, 1).
 21.2 Convergence of Moments 421

A similar result for convergence to the Poisson distribution can also be deduced
from Theorem 21.7. Instead, we will show how to derive it directly from the
Inclusion-Exclusion Principle.

The following lemma sometimes simplifies the proof of some probabilistic in-
equalities: A boolean function f of events A1 , A2 , . . . , An ⊆ Ω


is a random vari-
able where f (A1 , A2 , . . . , An ) = S∈S ( i∈S Ai ) ∩ i6∈S Aci for some collec-
S T T

tion Si of subsest of [r] = {1, 2, . . . , r}.

Lemma 21.9 (Rényi’s Lemma) Suppose that A1 , A2 , . . . , Ar are events in

some probability space Ω, f1 , f2 , . . . , fs are boolean functions of A1 , A2 , . . . , Ar ,
and α1 , α2 , . . . , αs are reals. Then, if
s
∑ αi P( fi (A1 , A2 , . . . , Ar )) ≥ 0, (21.6)
i=1

whenever P(Ai ) = 0 or 1, then (21.6) holds in general.

Proof Write
! !!
[ \ \
fi = Ai ∩ Aci ,
S∈Si i∈S i6∈S

for some collection Si of subsets of [r] = {1, 2, . . . , r}.

Then,
! !!
\ \
P( fi ) = ∑ P Ai ∩ Aci ,
S∈Si i∈S i6∈S

and then the left hand side of (21.6) becomes

! !!
\ \
∑ βS P Ai ∩ Aci ,
S⊆[r] i∈S i6∈S

for some real βS . If (21.6) holds, then βS ≥ 0 for every S, since we can choose
Ai = Ω if i ∈ S, and Ai = 0/ for i 6∈ S.

For J ⊆ [r] let AJ = i∈J Ai , and let S = | j : A j occurs | denote the number
T

of events that occur. Then let

 
S
Bk = ∑ P(AJ ) = E .
J:|J|=k
k

Let E j = |S|= j ( i∈S Ai ∩ c

S T T
/ Ai ) be the event that exactly
i∈S j among the events
A1 , A2 , . . . , Ar occur.
422 Moments

The expression in Lemma 21.10 can be motivated as follows. For ω ∈ Ω, we

let θω,i ∈ {0, 1} satisfy θω,i = 1 if and only if ω ∈ Ai . Then

Pr(E j ) = ∑ Pr(ω) ∑ ∏ θω,i · ∏(1 − θω,i ),

ω∈Ω |I|= j i∈I i∈I
/

= ∑ Pr(ω) ∑ ∑ (−1)|J\I| θω,J , where θω,J = ∏ θω,i ,

ω∈Ω |I|= j J⊇I i∈J
\
J\I
= ∑ ∑ (−1) Pr(AJ ), where AJ = Ai
|J|≥ j I⊆J i∈J
|I|= j
r  
k− j k
= ∑ (−1) ∑ Pr(AJ )
k= j j |J|=k
r  
k
= ∑ (−1)k− j j
Bk .
k= j

More precisely,

Lemma 21.10
 s
≤ ∑ (−1)k− j kj Bk


s − j even.




k= j



 s

k− j k B


P(E j ) ≥ ∑ (−1) j k s − j odd

 k= j
 s

k− j k B


= ∑ (−1) s = r.

j k


k= j

Proof It follows from Lemma 21.9 that we only need to check the truth of
the statement for

P(Ai ) = 1 1 ≤ i ≤ `,
P(Ai ) = 0 ` < i ≤ r.

where 0 ≤ ` ≤ r is arbitrary.
Now
(
1 if j = `,
P(S = j) =
0 if j 6= `,

and
 
`
Bk = .
k
 21.2 Convergence of Moments 423

So,
s   s   
k k `
∑ (−1)k− j j
Bk = ∑ (−1)k− j
j k
k= j k= j
  s  
` k− j ` − j
= ∑ (−1) . (21.7)
j k= j k− j

If ` < j then P(E j ) = 0 and the sum in (21.7) reduces to zero. If ` = j then
P(E j ) = 1 and the sum in (21.7) reduces to one. Thus in this case, the sum is
exact for all s. Assume then that r ≥ ` > j. Then P(E j ) = 0 and
s   s− j    
k− j ` − j t `− j s− j ` − j − 1
∑ (−1) k − j = ∑ (−1) t = (−1) s− j
.
k= j t=0

This explains the alternating signs of the theorem. Finally, observe that `−r−j−1


j
= 0, as required.

Now we are ready to state the main tool for proving convergence to the Poisson
distribution.
Theorem 21.11 Let Sn = ∑i≥1 Ii be a sequence of random variables, n ≥ 1
(n)
and let Bk = E Skn . Suppose that there exists λ ≥ 0, such that for every fixed

k ≥ 1,
(n) λk
lim Bk = .
n→∞ k!
Then, for every j ≥ 0,
λj
lim P(Sn = j) = e−λ ,
n→∞ j!
i.e., Sn converges in distribution to the Poisson distributed random variable
D
with expectation λ (Sn → Po(λ )).
Proof By Lemma 21.10, for l ≥ 0,
j+2l+1   j+2l  
k− j k (n) k− j k (n)
∑ (−1) B ≤ P(Sn = j) ≤ ∑ (−1) B .
k= j j k k= j j k
So, as n grows to ∞,
j+2l+1  
k− j k (n)
∑ (−1) Bk ≤ lim inf P(Sn = j)
k= j j n→∞

j+2l  
k− j k (n)
≤ lim sup P(Sn = j) ≤ ∑ (−1) B .
n→∞ k= j j k
424 Moments

But,
j+m
λj m
  k
k λ λt λ j −λ
∑ (−1)k− j j k!
= ∑
j! t=0
(−1)t
t!
→
j!
e ,
k= j

as m → ∞.
Notice that the falling factorial
(Sn )k = Sn (Sn − 1) · · · (Sn − k + 1)
counts number of ordered k-tuples of events with Ii = 1. Hence the binomial
moments of Sn can be replaced in Theorem 21.11 by the factorial moments,
defined as
E(Sn )k = E[Sn (Sn − 1) · · · (Sn − k + 1)],
and one has to check whether, for every k ≥ 1,
lim E(Sn )k = λ k .
n→∞

21.3 Stein–Chen Method

Stein in [715] introduced a powerful technique for obtaining estimates of the
rate of convergence to the standard normal distribution. His approach was sub-
sequently extended to cover convergence to the Poisson distribution by Chen
[179], while Barbour [58] ingeniously adapted both methods to random graphs.
The Stein–Chen approach has some advantages over the method of moments.
The principal advantage is that a rate of convergence is automatically obtained.
Also the computations are often easier and fewer moment assumptions are
required. Moreover, it frequently leads to conditions for convergence weaker
than those obtainable by the method of moments.
Consider a sequence of random variables (Xn )∞ ∞
n=1 and let (λn )n=1 be a se-
quence of positive integers, and let Po (λ ) denote, as before, the Poisson dis-
tribution with expectation λ . We say that Xn is Poisson convergent if the total
variation distance between the distribution L (Xn ) of Xn and Po(λn ), λn =
E Xn , distribution, tends to zero as n tends to infinity. So, we ask for
 λk 
dTV (L (Xn ), Po(λn )) = sup  P(Xn ∈ A) − ∑ n e−λn  → 0, (21.8)
 
A⊆Z +
k∈A k!

as n → ∞, where Z+ = {0, 1, . . .}.

Notice, that if Xn is Poisson convergent and λn → λ , then Xn converges in
distribution to the Po (λ ) distributed random variable. Furthermore, if λn →
 21.3 Stein–Chen Method 425

0, then Xn converges to a random variable with distribution degenerated at

0. More importantly, if λn → ∞, then the central limit theorem for Poisson
√
distributed random variables implies, that X̃n = (Xn − λn )/ λn converges in
distribution to a random variable with the standard normal random distribution
N(0, 1).
The basic feature and advantage of the Stein–Chen approach is that it gives
computationally tractable bounds for the distance dTV , when the random vari-
ables in question are sums of indicators with a fairly general dependence struc-
ture.
Let {Ia }a∈Γ , be a family of indicator random variables, where Γ is some index
set. To describe the relationship between these random variables we define a
dependency graph L = (V (L), E(L)), where V (L) = Γ. Graph L has the prop-
erty that whenever there are no edges between A and B, A, B ⊆ Γ, then {Ia }a∈A
and {Ib }b∈B are mutually independent families of random variables. The fol-
lowing general bound on the total variation distance was proved by Barbour,
Holst and Janson [59] via the Stein–Chen method.

Theorem 21.12 Let X = ∑a∈Γ Ia where the Ia are indicator random variables
with a dependency graph L. Then, with πa = E Ia and
λ = E X = ∑a∈Γ πa ,

dTV (L (X), Po(λ )) ≤

!
−1
min(λ , 1) ∑ πa2 + ∑ {E(Ia Ib ) + πa πb } ,
a∈V (L) ab∈E(L)

where ∑ab∈E(L) means summing over all ordered pairs (a, b), such that {a, b} ∈
E(L).

Finally, let us briefly mention, that the original Stein method investigates the
convergence to the normal distribution in the following metric
Z Z 
dS (L (Xn ), N(0, 1)) = sup ||h||−1  h(x)dFn (x) − h(x)dΦ(x), (21.9)
 
h

where the supremum is taken over all bounded test functions h with bounded
derivative, ||h|| = sup |h(x)| + sup |h0 (x)|.
Here Fn is the distribution function of Xn , while Φ denotes the distribution
function of the standard normal distribution. So, if dS (L (Xn ), N(0, 1)) → 0 as
n → ∞, then X̃n converges in distribution to N(0, 1) distributed random vari-
able.
Barbour, Karoński and Ruciński [61] obtained an effective upper bound on
426 Moments

dS (L (Xn ), N(0, 1)) if S belongs to a general class of decomposable random

variables. This bound involves the first three moments of S only.
For a detailed and comprehensive account of the Stein–Chen method the reader
is referred to the book by Barbour, Holst and Janson [59], or to Chapter 6
of the book by Janson, Łuczak and Ruciński [441], where other interesting
approaches to study asymptotic distributions of random graph characteristics
are also discussed. For some applications of the Stein–Chen method in random
graphs, one can look at a survey by Karoński [476].
 22
Inequalities

22.1 Binomial Coefficient Approximation

We state some important inequalities. The proofs of all but (g) are left as exer-
cises:
Lemma 22.1 (a)
1 + x ≤ ex , ∀x.
(b)
1 − x ≥ e−x/(1−x) , 0 ≤ x < 1.
(c)
   
n ne k
≤ , ∀n, k.
k k
(d)
nk k k−1
   
n
≤ 1− , ∀n, k.
k k! 2n
(e)
nk nk
   
k(k − 1) n
1− ≤ ≤ e−k(k−1)/(2n) , ∀n, k.
k! 2n k k!
(f)
nk
 
n
≈ , i f k2 = o(n).
k k!
(g) If a ≥ b then
n−a

t − b b n − t − a + b a−b  t b  n − t a−b
   
t−b
≤ n
≤ .
n−b n−a t
n n−b
Proof (g)
n−a

t−b (n − a)!t!(n − t)!
n
=
t
n!(t − b)!(n − t − a + b)!
t(t − 1) · · · (t − b + 1) (n − t)(n − t − 1) · · · (n − t − a + b + 1)
= ×
n(n − 1) · · · (n − b + 1) (n − b)(n − b − 1) · · · (n − a + 1)

427
428 Inequalities
 t b  a−b
n−t
≤ × .
n n−b
The lower bound follows similarly.
We will need also the following estimate for binomial coefficients. It is a little
more precise than those given in Lemma 22.1.

Lemma 22.2 Let k = o(n3/4 ). Then

 2
nk k3
  
n k
≈ exp − − 2 .
k k! 2n 6n
Proof

nk k−1
   
n i
= 1 −
k k! ∏
i=0 n
( )
k−1
nk

i
= exp ∑ log 1 −
k! i=0 n
(  4 )
k−1 
nk i2

i k
= exp − ∑ + 2 +O 3
k! i=0 n 2n n
k
 2 3

n k k
= (1 + o(1)) exp − − 2 .
k! 2n 6n

22.2 Balls in Boxes

Suppose that we have M boxes and we independently place N distinguishable
balls into them. Let us assume that a ball goes into box i with probability pi
where p1 + · · · + pM = 1. Let Wi denote the number of balls that are placed in
box i and for S ⊆ [M], let WS = ∑i∈S Wi . The following looks obvious and is
extremely useful.

Theorem 22.3 Let S, T be disjoint subsets of [M] and let s,t be non-negative
integers. Then

P(WS ≤ s | WT ≤ t) ≤ P(WS ≤ s). (22.1)

P(WS ≥ s | WT ≤ t) ≥ P(WS ≥ s). (22.2)
P(WS ≥ s | WT ≥ t) ≤ P(WS ≥ s). (22.3)
P(WS ≤ s | WT ≥ t) ≥ P(WS ≤ s). (22.4)
 22.3 FKG Inequality 429

Proof Equation (22.2) follows immediately from (22.1). Also, equation (22.4)
follows immediately from (22.3). The proof of (22.3) is very similar to that of
(22.1) and so we will only prove (22.1).
Let
πi = P(WS ≤ s | WT = i).
Given WT = i, we are looking at throwing N − i balls into M − 1 boxes. It is
clear therefore that πi is monotone increasing in i. Now, let qi = P(WT = i).
Then,
N
P(WS ≤ s) = ∑ πi qi .
i=0
t
qi
P(WS ≤ s | WT ≤ t) = ∑ πi .
i=0 q0 + · · · + qt
So, (22.1) reduces to
t N
(q0 + · · · + qN ) ∑ πi qi ≤ (q0 + · · · + qt ) ∑ πi qi ,
i=0 i=0
or
t N
(qt+1 + · · · + qN ) ∑ πi qi ≤ (q0 + · · · + qt ) ∑ πi q i ,
i=0 i=t+1
or
N t t N
∑ ∑ qi q j πi ≤ ∑ ∑ qi q j πi .
j=t+1 i=0 j=0 i=t+1

The result now follows from the monotonicity of πi .

The following is an immediate corollary:
Corollary 22.4 Let S1 , S2 , . . . , Sk be disjoint subsets of [M] and let s1 , s2 , . . . , sk
be non-negative integers. Then
!
k
\ k
P {WSi ≤ si } ≤ ∏ P({WSi ≤ si }).
i=1 i=1
!
k
\ k
P {WSi ≥ si } ≤ ∏ P({WSi ≥ si }).
i=1 i=1

22.3 FKG Inequality

A function f : CN = {0, 1}[N] → R is said to be monotone increasing if when-
ever x = (x1 , x2 , . . . , xN ), y = (y1 , y2 , . . . , yN ) ∈ CN and x ≤ y ∈ CN (i.e. x j ≤
430 Inequalities

y j , j = 1, 2, . . . , N) then f (x) ≤ f (y). Similarly, f is said to be monotone de-

creasing if − f is monotone increasing.
An important example for us is the case where f is the indicator function of
some subset A of 2[N] . Then
(
1 x∈A
f (x) = .
emptyset x ∈ /A

A typical example for us would be N = n2 and then each G ∈ 2[N] corresponds

to a graph with vertex set [n]. Then A will be a set of graphs i.e. a graph prop-
erty. Suppose that f is the indicator function for A . Then f is monotone in-
creasing, if whenever G ∈ A and e ∈ / E(G) we have G + e ∈ A i.e. adding an
edge does not destroy the property. We will say that the set/property is mono-
tone increasing. For example if H is the set of Hamiltonian graphs then H
is monotone increasing. If P is the set of planar graphs then P is monotone
decreasing. In other words a property is monotone increasing iff its indicator
function is monotone increasing.
Suppose next that we turn CN into a probability space by choosing some p1 , p2 ,
. . . , pN ∈ [0, 1] and then for x = (x1 , x2 , . . . , xN ) ∈ CN letting
P(x) = ∏ pj ∏ (1 − p j ). (22.5)
j:x j =1 j:x j =0

If N = n2 and p j = p, j = 1, 2, . . . , N then this model corresponds to Gn,p .

The following is a special case of the FKG inequality, Harris [404] and Fortuin,
Kasteleyn and Ginibre [313]:

Theorem 22.5 If f , g are monotone increasing functions on CN then E( f g) ≥

E( f ) E(g).
Proof We will prove this by induction on N. If N = 0 then E( f ) = a, E(g) = b
and E( f g) = ab for some constants a, b.
So assume the truth for N − 1. Suppose that E( f | xN = 0) = a and E(g | xN =
0) = b then
E(( f − a)(g − b)) − E( f − a) E(g − b) = E( f g) − E( f ) E(g).
By replacing f by f − a and g by g − b we may therefore assume that E( f |
xN = 0) = E(g | xN = 0) = 0. By monotonicity, we see that E( f | xN = 1), E(g |
xN = 1) ≥ 0.
We observe that by the induction hypothesis that
E( f g | xN = 0) ≥ E( f | xN = 0) E(g | xN = 0) = 0
 22.4 Sums of Independent Bounded Random Variables 431

E( f g | xN = 1) ≥ E( f | xN = 1) E(g | xN = 1) ≥ 0

Now, by the above inequalities,

E( f g) = E( f g | xN = 0)(1 − pN ) + E( f g | xN = 1)pN
≥ E( f | xN = 1) E(g | xN = 1)pN . (22.6)

Furthermore,

E( f ) E(g) =
(E( f | xN = 0)(1 − pN ) + E( f | xN = 1)pN )×
(E(g | xN = 0)(1 − pN ) + E(g | xN = 1)pN )
= E( f | xN = 1) E(g | xN = 1)p2N . (22.7)

The result follows by comparing (22.6) and (22.7) and using the fact that E( f |
xN = 1), E(g | xN = 1) ≥ 0 and 0 ≤ pN ≤ 1.
In terms of monotone increasing sets A , B and the same probability (22.5) we
can express the FKG inequality as

P(A | B) ≥ P(A ). (22.8)

22.4 Sums of Independent Bounded Random Variables

Suppose that S is a random variable and t > 0 is a real number. We will be
concerned here with bounds on the upper and lower tail of the distribution of
S, i.e., on P(S ≥ µ + t) and P(S ≤ µ − t), respectively, where µ = E S.
The basic observation which leads to the construction of such bounds is due to
Bernstein [83]. Let λ ≥ 0, then

P(S ≥ µ + t) = P(eλ S ≥ eλ (µ+t) ) ≤ e−λ (µ+t) E(eλ S ), (22.9)

by Markov’s inequality (see Lemma 21.1). Similarly for λ ≤ 0,

P(S ≤ µ − t) ≤ e−λ (µ−t) E(eλ S ). (22.10)

Combining (22.9) and (22.10) one can obtain a bound for P(|S − µ| ≥ t).
Now let Sn = X1 + X2 + · · · + Xn where Xi , i = 1, . . . , n are independent random
variables. Assume that 0 ≤ Xi ≤ 1 and E Xi = µi for i = 1, 2, . . . , n. Let µ =
µ1 + µ2 + · · · + µn . Then for λ ≥ 0
n
P(Sn ≥ µ + t) ≤ e−λ (µ+t) ∏ E(eλ Xi ) (22.11)
i=1
432 Inequalities

and for λ ≤ 0
n
P(Sn ≤ µ − t) ≤ e−λ (µ−t) ∏ E(eλ Xi ). (22.12)
i=1

Note that E(eλ Xi ) in (22.11) and (22.12), likewise E(eλ S ) in (22.9) and (22.10)
are the moment generating functions of the Xi ’s and S, respectively. So finding
bounds boils down to the estimation of these functions.
Now the convexity of ex and 0 ≤ Xi ≤ 1 implies that

eλ Xi ≤ 1 − Xi + Xi eλ .

Taking expectations we get

E(eλ Xi ) ≤ 1 − µi + µi eλ .

Equation (22.11) becomes, for λ ≥ 0,

n
P(Sn ≥ µ + t) ≤ e−λ (µ+t) ∏(1 − µi + µi eλ )
i=1
!n
−λ (µ+t) n − µ + µeλ
≤e . (22.13)
n

The second inequality follows from the fact that the geometric mean is at
most the arithmetic mean i.e. (x1 x2 · · · xn )1/n ≤ (x1 + x2 + · · · + xn )/n for non-
negative x1 , x2 , . . . , xn . This in turn follows from Jensen’s inequlaity and the
concavity of log x.
The right hand side of (22.13) attains its minimum, as a function of λ , at
(µ + t)(n − µ)
eλ = . (22.14)
(n − µ − t)µ
Hence, by (22.13) and(22.14), assuming that µ + t < n,
 µ µ+t  n − µ n−µ−t
P(Sn ≥ µ + t) ≤ , (22.15)
µ +t n− µ −t
while for t > n − µ this probability is zero.
Now let
ϕ(x) = (1 + x) log(1 + x) − x, x ≥ −1,

and let ϕ(x) = ∞ for x < −1. Now, for 0 ≤ t < n − µ, we can rewrite the bound
(22.15) as
    −t 
t
P(Sn ≥ µ + t) ≤ exp −µϕ − (n − µ)ϕ . (22.16)
µ n−µ
 22.4 Sums of Independent Bounded Random Variables 433

Since ϕ(x) ≥ 0 for every x, we get

P(Sn ≥ µ + t) ≤ e−µϕ(t/µ) . (22.17)
Similarly, putting n − Sn for Sn , or by an analogous argument, using (22.12),
we get for 0 ≤ t ≤ µ,
  −t   t 
P(Sn ≤ µ − t) ≤ exp −µϕ − (n − µ)ϕ . (22.18)
µ n−µ
Hence,
P(Sn ≤ µ − t) ≤ e−µϕ(−t/µ) . (22.19)
We can simplify the expressions (22.17) and (22.19) by observing that
x2
ϕ(x) ≥ . (22.20)
2(1 + x/3)
To see this observe that for |x| ≤ 1 we have
x2 ∞  
1 1
ϕ(x) − = ∑ (−1)k − xk .
2(1 + x/3) k=2 k(k − 1) 2 · 3k−2
1
Equation (22.20) for |x| ≤ 1 follows from k(k−1) − 2·31k−2 ≥ 0 for k ≥ 2. We
leave it as an exercise to check that (22.20) remains true for x > 1.
Taking this into account we arrive at the following theorem, see Hoeffding
[415].

Theorem 22.6 (Chernoff/Hoeffding inequality) Suppose that

Sn = X1 + X2 + · · · + Xn where (i) 0 ≤ Xi ≤ 1 and E Xi = µi for i = 1, 2, . . . , n,
(ii) X1 , X2 , . . . , Xn are independent. Let µ = µ1 + µ2 + · · · + µn . Then for t ≥ 0,
t2
 
P(Sn ≥ µ + t) ≤ exp − (22.21)
2(µ + t/3)
and for t ≤ µ,
t2
 
P(Sn ≤ µ − t) ≤ exp − . (22.22)
2(µ − t/3)

Putting t = ε µ, for 0 < ε < 1, one can immediately obtain the following
bounds.

Corollary 22.7 Let 0 < ε < 1, then

µε 2
 
 eε µ
P(Sn ≥ (1 + ε)µ) ≤ ≤ exp − , (22.23)
(1 + ε)1+ε 3
434 Inequalities

while
µε 2
 
P(Sn ≤ (1 − ε)µ) ≤ exp − (22.24)
2

Proof The formula (22.24) follows directly from (22.22) and (22.23) follows
from (22.16).
One can “tailor” Chernoff bounds with respect to specific needs. For exam-
ple, for small ratios t/µ, the exponent in (22.21) is close to t 2 /2µ, and the
following bound holds.

Corollary 22.8
 2
t3

t
P(Sn ≥ µ + t) ≤ exp − + 2 (22.25)
2µ 6µ
 2
t
≤ exp − for t ≤ µ. (22.26)
3µ

Proof Use (22.21) and note that

(µ + t/3)−1 ≥ (µ − t/3)/µ 2 .

For large deviations we have the following result.

Corollary 22.9 If c > 1 then
n e ocµ
P(Sn ≥ cµ) ≤ . (22.27)
ce1/c

Proof Put t = (c − 1)µ into (22.17).

We also have the simple easiest to use version:

Corollary 22.10 Suppose that X1 , X2 , . . . , Xn are independent random vari-
able and that ai ≤ Xi ≤ bi for i = 1, 2, . . . , n. Let X1 + X2 + · · · + Xn and µi =
E(Xi ), i = 1, 2, . . . , n and µ = E(S). Then for t > 0 and ci = bi −ai , i = 1, 2, . . . , n,
we have
2t 2
 
P(S ≥ µ + t) ≤ exp − 2 . (22.28)
c1 + c22 + · · · + c2n
2t 2
 
P(S ≤ µ − t) ≤ exp − 2 . (22.29)
c1 + c22 + · · · + c2n
 22.4 Sums of Independent Bounded Random Variables 435

Proof We can assume without loss of generality that ai = 0, i = 1, 2, . . . , n.

We just subtract A = ∑ni=1 ai from S. We proceed as before.
n
P(S ≥ µ + t) = P(eλ S ≥ eλ (µ+t) ) ≤ e−λ (µ+t) E(eλ S ) = e−λt ∏ E(eλ (Xi −µi ) ).
i=1

Note that eλ x is a convex function of x, and since 0 ≤ Xi ≤ ci , we have

 
λ (Xi −µi ) −λ µi Xi Xi λ ci
e ≤e 1− + e
ci ci
and so
 
µi µi
E(eλ Xi ) ≤ e−λ µi 1 − + eλ ci
ci ci
 
−θi pi
=e 1 − pi + pi eθi , (22.30)

where θi = λ ci and pi = µi /ci .

Then, taking the logarithm of the RHS of (22.30), we have
 
f (θi ) = −θi pi + log 1 − pi + pi eθi .
pi
f 0 (θi ) = −pi + .
1 − p i + p i e θi
pi (1 − pi )e−θi
f 00 (θi ) = .
((1 − pi )e−θi + pi )2

Now αβ
(α+β )2
≤ 1/4 and so f 00 (θi ) ≤ 1/4 anf therefore

1 λ 2 c2i
f (θi ) ≤ f (0) + f 0 (0)θi + θi2 = .
8 8
It follows then that
( )
n
−λt λ 2 c2i
P(S ≥ µ + t) ≤ e exp ∑ .
i=1 8
4
We obtain (22.28) by puttng λ = and (22.29) is proved in a similar
∑ni=1 c2i
manner.
Our next bound incorporates the variance of the Xi ’s.
Theorem 22.11 (Bernstein’s Theorem) Suppose that Sn = X1 + X2 + · · · +
Xn where (i) |Xi | ≤ 1 and E Xi = 0 and Var Xi = σi2 for i = 1, 2, . . . , n, (ii)
X1 , X2 , . . . , Xn are independent. Let σ 2 = σ12 + σ22 + · · · + σn2 . Then for t ≥ 0,
t2
 
P(Sn ≥ t) ≤ exp − (22.31)
2(σ 2 + t/3)
436 Inequalities

and
t2
 
P(Sn ≤ −t) ≤ exp − . (22.32)
2(σ 2 + t/3)
Proof The strategy is once again to bound the moment generating function.
Let
∞
λ r−2 E Xir ∞
λ r−2 σi2 eλ − 1 − λ
Fi = ∑ ≤ ∑ = .
r=2 r!σi2 r=2 r!σi
2 λ2

Here E Xir ≤ σi2 , since |Xi | ≤ 1.

We then observe that
∞
λ r E Xir
E(eλ Xi ) = 1 + ∑
r=2 r!
= 1 + λ 2 σi2 Fi
2σ 2F
≤ eλ i i
n o
≤ exp (eλ − λ − 1)σi2 .

So,
n n o
P(Sn ≥ t) ≤ e−λt ∏ exp (eλ − λ − 1)σi2
i=1
σ 2 (eλ −λ −1)−λt
=e
n  t o
= exp −σ 2 ϕ .
σ2
after assigning
 t 
λ = log 1 + 2 .
σ
To obtain (22.31) we use (22.20). To obtain (22.32) we apply (22.31) to Yi =
−Xi , i = 1, 2, . . . , n.

22.5 Sampling Without Replacement

Let a multi-set A = {a1 , a2 , . . . , aN } ⊆ R be given. We consider two random
variables. For the first let X = ai where i is chosen uniformly at random from
[N]. Let

1 N 2 1 N
µ = EX = ∑
N i=1
ai and σ = Var X = ∑ (ai − µ)2 .
N i=1
 22.5 Sampling Without Replacement 437

Now let Sn = X1 + X2 + · · · + Xn be the sum of n independent copies of X.

Next let Wn = ∑i∈X ai where X is a uniformly random n-subset of [N]. We
have E Sn = EWn = nµ but as shown in Hoeffding [415], Wn is more tightly
concentrated around its mean than Sn . This will follow from the following:

Lemma 22.12 Let f : R → R be continuous and convex. Then

E f (Wn ) ≤ E f (Sn ).

Proof We write, where (A)n denotes the set of sequences of n distinct mem-
bers of A and (N)n = N(N − 1) · · · (N − n + 1) = |(A)n |,

1
E f (Sn ) = ∑n f (y1 + · · · + yn ) =
N n y∈A
1
∑ g(x1 , x2 , . . . , xn ) = E g(X), (22.33)
(N)n x∈(A)
n

where g is a symmetric function of x and

g(x1 , x2 , . . . , xn ) = ∑ ψ(k, i, r) f (ri1 xi1 + · · · + rik xik ).

k,i,r

Here i ranges over sequences of k distinct values i1 , i2 , . . . , ik ∈ [n] and ri1 +

· · · + rik = n. The factors ψ(k, i, r) are independent of the function f .
Putting f = 1 we see that ∑k,i,r ψ(k, i, r) = 1. Putting f (x) = x we see that g is
a linear symmetric function and so

∑ ψ(k, i, r)(ri1 xi1 + · · · + rik xik ) = K(x1 + · · · + xn ),

k,i,r

for some K. Equation (22.33) implies that K = 1.

Applying Jensen’s inequality we see that

g(x) ≥ f (x1 + · · · + xn ).

It follows that
E g(X) ≥ E f (Wn )

and the Lemma follows from (22.33).

As a consequence we have that (i) VarWn ≤ Var Sn and (ii) E eλWn ≤ E eλ Sn for
any λ ∈ R.
Thus all the inequalities developed in Section 22.4 can a fortiori be applied to
Wn in place of Sn . Of particular importance in this context, is the hypergeomet-
ric distribution: Here we are given a set of S ⊆ [N], |S| = m and we choose a
438 Inequalities

random set X of size k from [N]. Let Z = |X ∩ S|. Then

m
N−m

t k−t
P(Z = t) = N

, for 0 ≤ t ≤ k.
k

22.6 Janson’s Inequality

In Section 22.4 we found bounds for the upper and lower tails of the distri-
bution of a random variable Sn composed of n independent summands. In the
previous section we allowed some dependence between the summands. We
consider another case where the random variables in question are not necessar-
ily independent. In this section we prove an inequality of Janson [427]. This
generalised an earlier inequality of Janson, Łuczak and Ruciński [440], see
Corollary 22.14.
Fix a family of n subsets Di , i ∈ [n]. Let R be a random subset of [N] such
that for s ∈ [N] we have 0 < P(s ∈ R) = qs < 1. The elements of R are chosen
independently of each other and the sets Di , i = 1, 2, . . . , n. Let Ai be the event
that Di is a subset of R. Moreover, let Ii be the indicator of the event Ai . Note
that, Ii and I j are independent iff Di ∩ D j = 0.
/ One can easily see that the Ii ’s
are increasing.
We let
Sn = I1 + I2 + · · · + In ,

and
n
µ = ESn = ∑ E(Ii ).
i=1

We write i ∼ j if Di ∩ D j 6= 0.
/ Then, let

∆= ∑ E(Ii I j ) = µ + ∆ (22.34)
{i, j}:i∼ j

where
∆= ∑ E(Ii I j ). (22.35)
{i, j}:i∼ j
i6= j

As before, let ϕ(x) = (1 + x) log(1 + x) − x. Now, with Sn , ∆, ϕ given above one

can establish the following upper bound on the lower tail of the distribution of
Sn .
 22.6 Janson’s Inequality 439

Theorem 22.13 (Janson’s Inequality) For any real t, 0 ≤ t ≤ µ,

ϕ(−t/µ)µ 2
   2
t
P(Sn ≤ µ − t) ≤ exp − ≤ exp − . (22.36)
∆ 2∆

Proof We begin as we did in Section 22.4. Put ψ(λ ) = E(e−λ Sn ), λ ≥ 0. By

Markov’s inequality we have
P(Sn ≤ µ − t) ≤ eλ (µ−t) Ee−λ Sn .
Therefore,
log P(Sn ≤ µ − t) ≤ log ψ(λ ) + λ (µ − t). (22.37)
Now let us estimate log ψ(λ ) and minimise the right-hand-side of (22.37) with
respect to λ .
Note that
n
−ψ 0 (λ ) = E(Sn e−λ Sn ) = ∑ E(Ii e−λ Sn ). (22.38)
i=1

Now for every i ∈ [n], split Sn into Yi and Zi , where

Yi = ∑ Ij, Zi = ∑ Ij, Sn = Yi + Zi .
j: j∼i j: j6∼i

Then by the FKG inequality (applied to the random set R and conditioned on
Ii = 1) we get, setting pi = E(Ii ) = ∏s∈Di qs ,

E(Ii e−λ Sn ) = pi E(e−λYi e−λ Zi  Ii = 1) ≥



pi E(e−λYi  Ii = 1)E(e−λ Zi  Ii = 1).

 

Since Zi and Ii are independent we get

E(Ii e−λ Sn ) ≥ pi E(e−λYi  Ii = 1)E(e−λ Zi )



≥ pi E(e−λYi  Ii = 1)ψ(λ ). (22.39)



From (22.38) and (22.39), applying Jensen’s inequality to get (22.40) and re-
membering that µ = ESn = ∑ni=1 pi , we get
ψ 0 (λ )
− (log ψ(λ ))0 = −
ψ(λ )
n
≥ ∑ pi E(e−λYi  Ii = 1)

i=1
n
pi  
≥µ∑ exp − E(λYi  Ii = 1)
i=1 µ
440 Inequalities
( )
1 n 
≥ µ exp − ∑ pi E(λYi  Ii = 1) (22.40)
µ i=1
( )
λ n
= µ exp − ∑ E(Yi Ii )
µ i=1

= µe−λ ∆/µ .
So
−(log ψ(λ ))0 ≥ µe−λ ∆/µ (22.41)
which implies that
µ2
Z λ
− log ψ(λ ) ≥ µe−z∆/µ dz = (1 − e−λ ∆/µ ). (22.42)
0 ∆
Hence by (22.42) and (22.37)
µ2
log P(Sn ≤ µ − t) ≤ − (1 − e−λ ∆/µ ) + λ (µ − t), (22.43)
∆
which is minimized by choosing λ = − log(1 − t/µ)µ/∆. It yields the first
bound in (22.36), while the final bound in (22.36) follows from the fact that
ϕ(x) ≥ x2 /2 for x ≤ 0.
The following Corollary is very useful:
Corollary 22.14 (Janson, Łuczak, Ruciński Inequality)
P(Sn = 0) ≤ e−µ+∆ .
2
n o
Proof We put t = µ into (22.36) giving P(Sn = 0) ≤ exp − ϕ(−1)µ
∆ . Now
µ2 µ2
note that ϕ(−1) = 1 and ∆ ≥ µ+∆ ≥ µ − ∆.

22.7 Martingales. Azuma-Hoeffding Bounds

Before we present the basic results of this chapter we have to briefly introduce
martingales and concentration inequalities for martingales. Historically, mar-
tingales were applied to random graphs for the first time in the context of the
chromatic number of Gn,p .
Let (Ω, F , P) be a probability space. If the sample space Ω is finite, then F is
the algebra of all subsets of Ω. For simplicity, let us assume that we deal with
this case.
Recall that if D = {D1 , D2 , . . . , Dm } is a partition of Ω, i.e., m
S
i=1 Di = Ω and
Di ∩ D j = 0/ if i 6= j, then it generates an algebra of subsets A (D) of Ω. The
 22.7 Martingales. Azuma-Hoeffding Bounds 441

algebra generated by the partition D and denoted by A (D) is the family of all
unions of the events (sets) from D, with 0/ obtained by taking an empty union.
Let D = {D1 , D2 , . . . , Dm } be a partition of Ω and A be any event, A ⊂ Ω and
let P(A|D) be the random variable defined by
m
P(A|D)(ω) = ∑ P(A|Di )IDi (ω)
i=1
= P(A|Di(ω) ) where ω ∈ Di(ω) .
Note that if D a trivial partition, i.e., D = D0 = {Ω} then P(A|D0 ) = P(A),
while, in general,
P(A) = E P(A|D). (22.44)
Suppose that X is a discrete random variable taking values
{x1 , x2 , . . . , xl } and write X as
l
X= ∑ x j IA j , (22.45)
j=1

where A j = {ω : X(ω) = x j }. Notice that the random variable X generates a

partition DX = {A1 , A2 , . . . , Al }.
Now the conditional expectation of X with respect to a partition D of Ω is
given as
l
E(X|D) = ∑ x j P(A j |D). (22.46)
j=1

Hence, E(X|D)(ω
 1 ) is the expected value of X conditional on the event
ω ∈ Di(ω1 ) .
Suppose that D and D 0 are two partitions of Ω. We say that D 0 is finer than D
if A (D) ⊆ A (D 0 ) and denote this as D ≺ D 0 .
If D is a partition of Ω and Y is a discrete random variable defined on Ω, then
Y is D-measurable if DY ≺ D, i.e., if the partition D is finer than the partition
induced by Y . It simply means that Y takes constant values yi on the atoms Di
of D, so Y can be written as Y = ∑m i=1 yi IDi , where some yi may be equal. Note
that a random variable Y is D0 -measurable if Y has a degenerate distribution,
i.e., it takes a constant value on all ω ∈ Ω. Also, trivially, the random variable
Y is DY -measurable.
Note that if D 0 is finer than D then
E(E(X | D 0 ) | D) = E(X | D). (22.47)
Indeed, if ω ∈ Ω then
E(E(X | D 0 ) | D)(ω) =
442 Inequalities
 
P(ω 00 ) 0
 P(ω )
= ∑  ∑0 X(ω 00 )
0
ω ∈Di(ω) 00
ω ∈Di(ω 0 )
P(D0i(ω 0 ) ) P(Di(ω) )

P(ω 0 )
= ∑ X(ω 00 ) P(ω 00 ) ∑0
00
ω ∈Di(w) 0
ω ∈Di(ω 00 )
P(D0i(ω 0 ) ) P(Di(ω) )

P(ω 0 )
= ∑ X(ω 00 ) P(ω 00 ) ∑
ω 00 ∈D ω 0 ∈D0i(ω 00 )
P(D0i(ω 00 ) ) P(Di(ω) )
i(w)

P(ω 00 )
= ∑ X(ω 00 )
ω 00 ∈Di(w)
P(Di(ω) )

= E(X | D)(ω).
Note that despite all the algebra, (22.47) just boils down to saying that the
properly weighted average of averages is just the average.
Finally, suppose a partition D of Ω is induced by a sequence of random vari-
ables {Y1 ,Y2 , . . . ,Yn }. We denote such partition as DY1 ,Y2 ,...,Yn . Then the atoms
of this partition are defined as
Dy1 ,y2 ,...,yn = {ω : Y1 (ω) = y1 ,Y2 (ω) = y2 , . . . ,Yn (ω) = yn },
where the yi range over all possible values of the Yi ’s. DY1 ,Y2 ,...,Yn is then the
coarsest partition such that Y1 ,Y2 , . . . ,Yn are all constant over the atoms of the
partition. For convenience, we simply write
E(X|Y1 ,Y2 , . . . ,Yn ), instead of E(X|DY1 ,Y2 ,...,Yn ).

Now we are ready to introduce an important class of dependent random vari-

ables called martingales.
Let (Ω, F , P) be a finite probability space and D0 ≺ D1 ≺ D2 ≺ . . . ≺ Dn = D ∗
be a nested sequence of partitions of Ω (a filtration of Ω), where D0 is a trivial
partition, while D ∗ stands for the discrete partition (i.e., A (D0 ) = {0,/ Ω},
while A (D ∗ ) = 2Ω = F ).
A sequence of random variables X0 , X1 , . . . , Xn is called (a) a martingale, (b)
a super-martingale, (c) a sub-martingale, with respect to the partition D0 ≺
D1 ≺ D2 ≺ . . . ≺ Dn = D ∗ if
Xk is Dk -measurable
and

(a) E(Xk+1 | Dk ) = Xk k = 0, 1, . . . , n − 1.
(b) E(Xk+1 | Dk ) ≤ Xk k = 0, 1, . . . , n − 1.
 22.7 Martingales. Azuma-Hoeffding Bounds 443

(c) E(Xk+1 | Dk ) ≥ Xk k = 0, 1, . . . , n − 1.

If the partition D of Ω is generated by a sequence of random variables Y1 , . . . ,Yn

then the sequence X1 , . . . , Xn is called a martingale with respect to the sequence
Y1 , . . . ,Yn . In particular, when Y1 = X1 , . . . ,Yn = Xn , i.e., when Dk = DX1 ,...,Xk ,
then we simply say that X is a martingale with respect to itself. Observe also
that E Xk = E X1 = X0 , for every k. Analogous statements hold for super- and
sub-martingales.
Martingales are ubiquitous, we can obtain a martingale from essentially any
random variable. Let Z = Z(Y1 ,Y2 , . . . ,Yn ) be a random variable defined on the
random variables Y1 ,Y2 , . . . ,Yn . The sequence of random variables

Xk = E(Z | Y1 ,Y2 , . . . ,Yk ), k = 0, 1, . . . , n

is called the Doob Martingale of Z.

Theorem 22.15 We have (i) X0 = E Z, (ii) Xn = Z and (iii) the sequence

X0 , X1 , . . . , Xn is a martingale with respect to (the partition defined by) Y1 ,Y2 , . . . ,Yn .

Proof Only (iii) needs to be explicitly checked.

E(Xk | Y1 , . . . ,Yk−1 ) = E(E(Z | Y1 , . . . ,Yk ) | Y1 , . . . ,Yk−1 )

= E(Z | Y1 , . . . ,Yk−1 )
= Xk−1 .

Here the second equality comes from (22.47).

We next show how one can define the so called, vertex and edge exposure
martingales, on the space of random graphs. Consider the binomial random
graph Gn,p . Let us view Gn,p as a vector of random variables (I1 , I2 , . . . , I(n) ),
2
where Ii is the indicator of the event that the ith edge is present, with P(Ii =
n

1) = p and P(Ii = 0) = 1 − p for i = 1, 2, . . . , 2 . These random variables
are independent of each
other. Hence, in this case, Ω consists of all (0, 1)-
sequences of length n2 .
Now given any graph invariant (a random variable) X : Ω → R, (for example,
the chromatic number, the number of vertices of given degree, the size of the
largest clique, etc.), we will define a martingale generated by X and certain
sequences of partitions of Ω.
Let the random variables I1 , I2 , . . . , I(n) be listed in a lexicographic order. De-
2
fine D0 ≺ D1 ≺ D2 ≺ . . . ≺ Dn = D ∗ in the following way: Dk is the partition
of Ω induced by the sequence of random variables I1 , . . . , I(k) , and D0 is the
2
444 Inequalities

trivial partition. Finally, for k = 1, . . . , n,

Xk = E(X | Dk ) = E(X | DI1 ,I2 ,...,I k ).

(2)

Hence, Xk is the conditional expectation of X, given that we “uncovered” the

set of edges induced by the first k vertices of our random graph Gn,p . A martin-
gale determined through such a sequence of nested partitions is called a vertex
exposure martingale.
An edge exposure martingale is defined in a similar way. The martingale se-
quence is defined as follows

Xk = E(X | Dk ) = E(X | DI1 ,I2 ,...,Ik ),

where k = 1, 2, . . . , n2 , i.e., we uncover the edges of Gn,p one by one.

We next give upper bounds for both the lower and upper tails of the probability
distributions of certain classes of martingales.

Theorem 22.16 (Azuma-Hoeffding bound) Let {Xk }n0 be a sequence of ran-

dom variables such that |Xk − Xk−1 | ≤ ck , k = 1, . . . , n and X0 is constant.

(a) If {Xk }n0 is a super-martingale then for all t > 0 we have

t2
 
P(Xn ≥ X0 + t) ≤ exp − n 2 .
2 ∑i=1 ci

(b) If {Xk }n0 is a sub-martingale then for all t > 0 we have

t2
 
P(Xn ≤ X0 − t) ≤ exp − n 2 .
2 ∑i=1 ci

(c) If {Xk }n0 is a martingale then for all t > 0 we have

t2
 
P(|Xn − X0 | ≥ t) ≤ 2 exp − n 2 .
2 ∑i=1 ci
Proof We only need to prove (a), since (b), (c) will then follow easily, since
{Xk }n0 is a sub-martingale iff −{Xk }n0 is a super-martingale and {Xk }n0 is a
martingale iff it is a super-martingale and a sub-martingale.
Define the martingale difference sequence by Y1 = 0 and

Yk = Xk − Xk−1 , k = 1, . . . , n.

Then
n
∑ Yk = Xn − X0 ,
k=1
 22.7 Martingales. Azuma-Hoeffding Bounds 445

and
E(Yk+1 | Y0 ,Y1 , . . . ,Yk ) ≤ 0. (22.48)

Let λ > 0. Then

( ) !
n
λt
P (Xn − X0 ≥ t) = P exp λ ∑ Yi ≥e
i=1
( )!
n
−λt
≤e E exp λ ∑ Yi ,
i=1

by Markov’s inequality.
Note that eλ x is a convex function of x, and since −ci ≤ Yi ≤ ci , we have
1 −Yi /ci −λ ci 1 +Yi /ci λ ci
eλYi ≤ e + e
2 2
Yi
= cosh(λ ci ) + sinh(λ ci ).
ci
It follows from (22.48) that

E(eλYn | Y0 ,Y1 , . . . ,Yn−1 ) ≤ cosh(λ cn ). (22.49)

We then see that

( )!
n
E exp λ ∑ Yi
i=1
( )!
n−1
= E E(eλYn | Y0 ,Y1 , . . . ,Yn−1 ) × exp λ ∑ Yi
i=1
( )!
n−1 n
≤ cosh(λ cn ) E exp λ ∑ Yi ≤ ∏ cosh(λ ci ).
i=1 i=1

The expectation in the middle term is over Y0 ,Y1 , . . . ,Yn−1 and the last inequal-
ity follows by induction on n.
By the above equality and the Taylor expansion, we get
n n
(λ ci )2m
∞
eλt P (Xn − X0 ≥ t) ≤ ∏ cosh(λ ci ) = ∏ ∑
i=1 i=1 m=0 (2m)!
( )
n ∞
(λ ci )2m 1 2 n 2
≤∏ ∑ m = exp λ ∑ ci .
i=1 m=0 2 m! 2 i=1

Putting λ = t/ ∑ni=1 c2i we arrive at the theorem.

We end by describing a simple situation where we can apply these inequalities.
446 Inequalities

Lemma 22.17 (McDiarmid’s Inequality) Let Z = Z(W1 ,W2 , . . . ,Wn ) be a ran-

dom variable that depends on n independent random variables W1 ,W2 , . . . ,Wn .
Suppose that
|Z(W1 , . . . ,Wi , . . . ,Wn ) − Z(W1 , . . . ,Wi0 , . . . ,Wn )| ≤ ci
for all i = 1, 2, . . . , n and W1 ,W2 , . . . ,Wn ,Wi0 . Then for all t > 0 we have
t2
 
P(Z ≥ E Z + t) ≤ exp − n 2 ,
2 ∑i=1 ci
and
t2
 
P(Z ≤ E Z − t) ≤ exp − .
2 ∑ni=1 c2i
Proof We consider the martingale
Xk = Xk (W1 ,W2 , . . . ,Wk ) = E(Z | W1 ,W2 , . . . ,Wk ).
Then
X0 = E Z and Xn = Z.
We only have to show that the martingale differences Yk = Xk − Xk−1 are
bounded. But,
|Xk (W1 , . . . ,Wk ) − Xk−1 (W1 , . . . ,Wk−1 )|
≤ ∑ |Z(W1 , . . . ,Wk , . . . ,Wn ) − Z(W1 , . . . ,Wk0 , . . . ,Wn ))|
Wk0 ,Wk+1 ,...,Wn
n
× P(Wk0 ) ∏ P(Wi )
i=k+1
n
≤ ∑ ck P(Wk0 ) ∏ P(Wi )
Wk0 ,Wk+1 ,...,Wn i=k+1

= ck .

22.8 Talagrand’s Inequality

In this section we describe a concentration inequality that is due to Talagrand
[725] that has proved to be very useful. It can often overcome the following
problem with using Theorems 22.16, 22.17: If E Xn = O(n1/2 ) then the bounds
they give are weak. Our treatment is a re-arrangement of the treatment in Alon
and Spencer [30].
 22.8 Talagrand’s Inequality 447

Let Ω = ∏ni=1 Ωi , where each Ωi is a probability space and Ω has the product
measure. Let A ⊆ Ω and let x = (x1 , x2 , . . . , xn ) ∈ Ω.
For α = (α1 , α2 , . . . , αn ) we let
dα (A, x) = inf ∑ αi .
y∈A
i:yi 6=xi

Then we define
ρ(A, x) = sup dα (A, x),
|α|=1

where |α| denotes the Euclidean norm, (α12 + · · · + αn2 )1/2 .

We then define, for t ≥ 0,
At = {x ∈ Ω : ρ(A, x) ≤ t} .
The following theorem is due to Talagrand [725]:

Theorem 22.18
2 /4
P(At )(1 − P(At )) ≤ e−t .

Theorem 22.18 follows from

Lemma 22.19  
1 2 1
Z
exp ρ (A, x) dx ≤ .
Ω 4 P(A)
Proof Indeed, fix A and consider X = ρ(A, x). Then,
2 /4 2 /4 2 /4 2 /4
1 − P(At ) = P(X > t) = P(eX > et ) ≤ E(eX )e−t .
The lemma states that E(e X 2 /4 )≤ 1
P(A) .

The following alternative description of ρ is important. Let

U(A, x) = {s ∈ {0, 1}n : ∃y ∈ A s.t. si = 0 implies xi = yi }
and let V (A, x) be the convex hull of U(A, x). Then
Lemma 22.20
ρ(A, x) = min |v|.
v∈V (A,x)

Here |v| denotes the Euclidean norm of v. We leave the proof of this lemma as
a simple exercise in convex analysis.
We now give the proof of Lemma 22.19.
448 Inequalities

Proof We use induction on the dimension n. For n = 1, ρ(A, x) = 1x∈A

/ so that
 
1 2 1
Z
exp ρ (A, x) = P(A) + (1 − P(A))e1/4 ≤
Ω 4 P(A)
which follows from u + (1 − u)e1/4 ≤ u−1 for 0 < u ≤ 1.
Assume the result for n. Write Ψ = ∏ni=1 Ωi so that Ω = Ψ × Ωn+1 . Any z ∈ Ω
can be written uniquely as z = (x, ω) where x ∈ Ψ and ω ∈ Ωn+1 . Set
B = {x ∈ Ψ : (x, ω) ∈ A for some ω ∈ Ωn+1 }
and for ω ∈ Ωn+1 set
Aω = {x ∈ Ψ : (x, ω) ∈ A} .
Then
s ∈ U(B, x) =⇒ (s, 1) ∈ U(A, (x, ω)).
t ∈ U(Aω , x) =⇒ (t, 0) ∈ U(A, (x, ω)).
If s ∈ V (B, x) and t ∈ V (Aω , x) then (s, 1) and (t, 0) are both in V (A, (x, ω))
and hence for any λ ∈ [0, 1],
((1 − λ )s + λ t, 1 − λ ) ∈ V (A, (x, ω)).
Then,
ρ 2 (A, (x, ω)) ≤ (1 − λ )2 + |(1 − λ )s + λ t|2 ≤ (1 − λ )2 + (1 − λ )|s|2 + λ |t|2 ,
where the second inequality uses the convexity of | · |2 .
Selecting, s, t with minimal norms yields the critical inequality,
ρ 2 (A, (x, ω)) ≤ (1 − λ )2 + λ ρ 2 (Aω , x) + (1 − λ )ρ 2 (B, x).
Now fix ω and bound,
 
1 2
Z
exp ρ (A, (x, ω)) ≤
x∈Ψ 4
 λ  1−λ
1 2 1 2
Z
(1−λ )2 /4
e exp ρ (Aω , x)) exp ρ (B, x)) .
x∈Ψ 4 4
By Hölder’s inequality this is at most
Z  λ Z  1−λ
(1−λ )2 /4 1 2 1 2
e exp ρ (Aω , x)) exp ρ (B, x)) ,
x∈Ψ 4 x∈Ψ 4
which by induction is at most
2 /4 1 1 1 (1−λ )2 /4 −λ
e(1−λ ) · = e r
P(Aω )λ P(B)1−λ P(B)
 22.8 Talagrand’s Inequality 449

where r = P(Aω )/ P(B) ≤ 1.

2
Using calculus, we minimise e(1−λ ) /4 r−λ by choosing λ = 1 + 2 log r for
2
e−1/2 ≤ r ≤ 1, λ = 0 otherwise. Further calculation shows that e(1−λ ) /4 r−λ ≤
2 − r for this value of λ . Thus,
   
1 2 1 P(Aω )
Z
exp ρ (A, (x, ω)) ≤ 2− .
x∈Ψ 4 P(B) P(B)
We integrate over ω to give
 
1 2
Z Z
exp ρ (A, (x, ω)) ≤
ω∈Ωn+1 x∈Ψ 4
 
1 P(A) 1
2− = x(2 − x),
P(B) P(B) P(A)
where x = P(A)/ P(B) ≤ 1. But x(2 − x) ≤ 1, completing the induction and
hence the theorem.
We call h : Ω → R Lipschitz if |h(x) − h(y)| ≤ 1 whenever x, y differ in at most
one coordinate.
Definition 22.21 Let f : N → N. h is f -certifiable if whenever h(x) ≥ s then
there exists I ⊆ [n] with |I| ≤ f (s) so that if y ∈ Ω agrees with x on coordinates
I then h(y) ≥ s.

Theorem 22.22 Suppose that h is Lipschitz and f -certifiable. Then if X =

h(x) for x ∈ Ω, then for all b and for all t ≥ 0,
2
P(X ≤ b − t f (b)) P(X ≥ b) ≤ e−t /4 .
p
n p o
Proof Set A = x : h(x) < b − t f (b) . Now suppose that h(y) ≥ b. We
claim that y ∈ / At . Let I be a set of indices of size at most f (b) that certifies
h(y) ≥ b as given above. Define αi = 0 when i ∈ / I and αi = |I|−1/2 when
i ∈ I. Using Lemma 22.20 we see p that if y ∈ At then there exists a z ∈ A that
differs from y in at most t|I|1/2 ≤ t f (b) coordinates of I, though at arbitrary
coordinates outside I. Let y0 agree with y on I and agree p with z outside I. By
0 0
the certification h(y ) ≥ b. Now y , z differ in at most t f (b) coordinates and
so, by Lipschitz,
h(z) ≥ h(y0 ) − t f (b) ≥ b − t f (b),
p p

but then z ∈
/ A, a contradiction. So, P(X ≥ b) ≤ 1 − P(At ) and from Theorem
22.18,
2
P(X < b − t f (b)) P(X ≥ b) ≤ e−t /4 .
p
450 Inequalities

As the RHS is continuous in t, we may replace “<” by “≤” giving Theorem

22.22.
Next let m denote the median of X so that P(X ≥ m) ≥ 1/2 and P(X ≤ m) ≥
1/2.

Corollary 22.23
2
f (m)) ≤ 2e−t /4 .
p
(a) P(X ≤ m − t
2
(b) Suppose that b − t f (b) ≥ m, then P(X ≥ b) ≤ 2e−t /4 .
p

22.9 Dominance
We say that a random variable X stochastically dominates a random variable Y
if
P(X ≥ t) ≥ P(Y ≥ t) for all real t.

There are many cases when we want to use our inequalities to bound the upper
tail of some random variable Y and (i) Y does not satisfy the necessary condi-
tions to apply the relevant inequality, but (ii) Y is dominated by some random
variable X that does. Clearly, we can use X as a surrogate for Y .
The following case arises quite often. Suppose that Y = Y1 +Y2 +· · ·+Yn where
Y1 ,Y2 , . . . ,Yn are not independent, but instead we have that for all t in the range
[Ai , Bi ] of Yi ,
P(Yi ≥ t | Y1 ,Y2 , . . . ,Yi−1 ) ≤ Φ(t)

where Φ(t) decreases monotonically from 1 to 0 in [Ai , Bi ].

Let Xi be a random variable taking values in the same range as Yi and such that
P(Xi ≥ t) = Φ(t). Let X = X1 + · · · + Xn where X1 , X2 , . . . , Xn are independent
of each other and Y1 ,Y2 , . . . ,Yn . Then we have

Lemma 22.24 X stochastically dominates Y .

Proof Let X (i) = X1 + · · · + Xi and Y (i) = Y1 + · · · + Yi for i = 1, 2, . . . , n. We

will show by induction that X (i) dominates Y (i) for i = 1, 2, . . . , n. This is triv-
ially true for i = 1 and for i > 1 we have

P(Y (i) ≥ t | Y1 . . . ,Yi−1 ) = P(Yi ≥ t − (Y1 + · · · +Yi−1 ) | Y1 . . . ,Yi−1 )

≤ P(Xi ≥ t − (Y1 + · · · +Yi−1 ) | Y1 . . . ,Yi−1 ).

Removing the conditioning we have

 22.9 Dominance 451

P(Y (i) ≥ t) ≤ P(Y (i−1) ≥ t − Xi ) ≤ P(X (i−1) ≥ t − Xi ) = P(X (i) ≥ t),

where the seond inequality follows by induction.
 23
Differential Equations Method

Let D ⊆ R2 be open and bounded and connected. Consider a general random

process
X(0), X(1), . . . , X(t), . . . , X(n) ∈ Z.
 
where X(0) is fixed and 0, X(0)
n ∈ D.
Let Ht denote the history X(0), X(1), . . . , X(t) of the process to time t. Let TD
be the stopping time which is the minimum t such that (t/n, X(t)/n) ∈ / D. We
further assume
(P1) |X(t)| ≤ C0 n, ∀t < TD , where C0 is a constant.
(P2) |X(t + 1) − X(t)| ≤ β = β (n) ≥ 1, ∀t < TD .
(P3) | E(X(t + 1) − X(t)|Ht , E ) − f (t/n, X(t)/n)| ≤ λ , ∀t < TD . Here E
is some likely event that holds with probability at least
1 − γ.
(P4) f (t, x) is continuous and satisfies a Lipschitz condition
| f (t, x) − f (t 0 , x0 )| ≤ Lk(t, x) − (t 0 , x)0 k∞
for (t, x), (t 0 , x0 ) ∈ D ∩ {(t, x) : t ≥ 0}

Theorem 23.1 Suppose that

nλ 3
λ = o(1) and α =  1.
β3

σ = inf{τ : (τ, z(τ)) 6∈ D0 = {(t, z) ∈ D : l ∞ distance of (t, z) from

the boundary of D ≥ 2λ }}
Let z(τ), 0 ≤ τ ≤ σ be the unique solution to the differential equation
z0 (τ) = f (τ, z(τ)) (23.1)
X(0)
z(0) = (23.2)
n
Then,
X(t) = nz(t/n) + O(λ n), (23.3)
uniformly in 0 ≤ t ≤ σ n, with probability 1 − O(γ + β e−α /λ ).

452
 Differential Equations Method 453

Proof The γ in the probability of success will be handled by conditioning on

E . Now let  
nλ
ω= .
β
We study the difference X(t + ω) − X(t). Assume that (t/n, X(t)/n) ∈ D0 . For
0 ≤ k ≤ ω we have from (P2) that
 
 X(t + k) X(t)  kβ ωβ
 − ≤ ≤ ,
 n n  n n
so    
t + k X(t + k) t X(t)
, − , ≤ 2λ ,
n n n n
∞
 
t+k X(t+k)
and so n , n is in D.

Therefore, using (P3),

E(X(t + k + 1) − X(t + k)|Ht+k , E ) =
 
t + k X(t + k)
f , + θk =
n n
 
t X(t)
f , + θk + ψk =
n n
 
t X(t)
f , + ρ,
n n

where |ρ| ≤ (2L + 1)λ , since |θk | ≤ λ (by (P3)) and |ψk | ≤ Lβn k (by (P4)).
Now, given Ht , let
  
X(t + k) − X(t) − k f t , X(t) − (2L + 1)kλ E
n n
Zk = .
emptyset ¬E

Then
E(Zk+1 − Zk |Z0 , Z1 , . . . , Zk ) ≤ 0,
i.e., Z0 , Z1 , . . . , Zω is a super-martingale.
Also   
 t X(t) 
|Zk+1 − Zk | ≤ β +  f ,  + (2L + 1)λ ≤ K0 β ,
n n
 
where K0 = O(1), since f nt , X(t) n = O(1) by continuity and boundedness of
D. So, using Theorem 22.16 we see that conditional on Ht , E ,
454 Differential Equations Method
 √ 
P X(t + ω) − X(t) − ω f (t/n, X(t)/n) ≥ (2L + 1)ωλ + K0 β 2αω
2K02 β 2 αω
 
≤ exp − = e−α . (23.4)
2ωK02 β 2
Similarly,
 √ 
P X(t + ω) − X(t) − ω f (t/n, X(t)/n) ≤ −(2L + 1)ωλ − K0 β 2αω
≤ e−α . (23.5)

Thus
 √ 
P |X(t + ω) − X(t) − ω f (t/n, X(t)/n)| ≥ (2L + 1)ωλ + K0 β 2αω
≤ 2e−α .
√
We have that ωλ and β 2αω are both Θ(nλ 2 /β ) giving
√ nλ 2
(2L + 1)ωλ + K0 β 2αω ≤ K1 .
β
Now let ki = iω for i = 0, 1, . . . , i0 = bσ n/ωc. We will show by induction that

P (∃ j ≤ i : |X(k j ) − z(k j /n)n| ≥ B j ) ≤ 2ie−α , (23.6)

where
!
Lω j+1
 
nλ
Bj = B 1+ −1 (23.7)
n L

and where B is another constant.

X(0)
The induction begins with z(0) = n and B0 = 0. Note that

Beσ L λ
Bi0 ≤ n = O(λ n). (23.8)
L
Now write

|X(ki+1 ) − z(ki+1 /n)n| = |A1 + A2 + A3 + A4 |,

A1 = X(ki ) − z(ki /n)n,
A2 = X(ki+1 ) − X(ki ) − ω f (ki /n, X(ki )/n),
A3 = ωz0 (ki /n) + z(ki /n)n − z(ki+1 /n)n,
A4 = ω f (ki /n, X(ki )/n) − ωz0 (ki /n).

We now bound each of these terms individually.

 Differential Equations Method 455

z(ki+1 /n)

A3

A4

slope z0 (ki /n)

z(ki /n)

slope f (ki .n, X(ki )/n)

A1

slope f (ki .n, X(ki )/n) A2

X(ki )/n)

X(ki+1 )/n

Figure 23.1 Error terms

Our induction gives that with probability at most 2ie−α ,

|A1 | ≤ Bi .
Equations (23.4) and (23.5) give
nλ 2
|A2 | ≤ K1 ,
β
with probability 1 − 2e−α .
A3 = ωz0 (ki /n) + z(ki /n)n − z(ki+1 /n)n
456 Differential Equations Method

Now
ω 0
z(ki+1 /n) − z(ki /n) =
z (ki /n + ω̂/n)
n
for some 0 ≤ ω̂ ≤ ω and so (P4) implies that
ω2 nλ 2
|A3 | = ω|z0 (ki /n + ω/n) − z0 (ki /n + ω̂/n)| ≤ L ≤ 2L 2 .
n β
Finally, (P4) gives
ωL|A1 | ωL
|A4 | ≤ ≤ Bi .
n n
Thus for some B > 0,
Bi+1 ≤ |A1 | + |A2 | + |A3 | + |A4 |
λ2
 
ωL
≤ 1+ Bi + Bn .
n β
A little bit of algebra verifies (23.6) and (23.7).
Finally consider ki ≤ t < ki+1 . From “time” ki to t the change in X and nz is at
most ωβ = O(nλ ).
Remark 23.2 The above proof generalises easily to the case where
X(t) is replaced by X1 (t), X2 (t), . . . , Xa (t) where a = O(1).
The earliest mention of differential equations with respect to random graphs
was in the paper by Karp and Sipser [486]. The paper by Ruciński and Wormald
[679] was also influential. See Wormald [752] for an extensive survey on the
differential equations method.
 24
Branching Processes

In the Galton-Watson branching process, we start with a single particle com-

prising generation 0. In general, the nth generation consists of Zn particles and
each member x of this generation independently gives rise to a random num-
ber X of descendants in generation n + 1. In the book we need the following
theorem about the probability that the process continues indefinitely: Let
pk = P(X = k), k = 0, 1, 2, . . . .
Let
∞
G(z) = ∑ pk zk
k=0

be the probability generating function (p.g..f.) of X. Let µ = E X. Let

!
[
η =P {Zn = 0} (24.1)
n≥0

be the probability of ultimate extinction of the process.

Theorem 24.1 η is the smallest non-negative root to the equation G(s) = s.
Here η = 1 if µ < 1.
Proof If Gn (z) is the p.g.f. of Zn , then Gn (z) = G(Gn−1 (z)). This follows from
the fact that Zn is the sum of Zn−1 independent copies of G. Let ηn = P(Zn = 0).
Then
ηn = Gn (0) = G(Gn−1 (0)) = G(ηn−1 ).
It follows from (24.1) that ηn % η. Let ψ be any other non-negative solution
to G(s) = s. We have
η1 = G(0) ≤ G(ψ) = ψ.
Now assume inductively that ηn ≤ ψ for some n ≥ 1. Then
ηn+1 = G(ηn ) ≤ G(ψ) = ψ.

457
 25
Entropy

25.1 Basic Notions

Entropy is a useful tool in many areas. The entropy we talk about here was
introduced by Shannon in [700]. We need some results on entropy in Chapter
14. We collect them here for convenience. For more on the subject we refer
the reader to Cover and Thomas [231], or Gray [389] or Martin and England
[584].
Let X be a random variable taking values in a finite set RX . Let p(x) = P(X = x)
for x ∈ RX . Then the entropy of X is given by

h(X) = − ∑ p(x) log p(x).

x∈RX

We have a choice for the base of the logarithm here. We use the natural loga-
rithm, for use in Chapter 14.
Note that if X is chosen uniformly from RX , i.e. P(X = x) = 1/|RX | for all
x ∈ RX then then
log |RX |
h(X) = ∑ = log |RX |.
x∈RX |RX |

We will see later that the uniform distribution maximises entropy.

If Y is another random variable with a finite range then we define the condi-
tional entropy

p(x, y)
h(X | Y ) = ∑ p(y)h(Xy ) = − ∑ p(x, y) log , (25.1)
y∈RY x,y p(y)

where Xy is the random variable with P(Xy = x) = P(X = x | Y = y). Here

p(y) = P(Y = y). The summation is over y such that p(y) > 0. We will use
notation like this from now on, without comment.
Chain Rule:

Lemma 25.1
m
h(X1 , X2 , . . . , Xm ) = ∑ h(Xi | X1 , X2 , . . . , Xi−1 ). (25.2)
i=1

458
 25.1 Basic Notions 459

Proof This follows by induction on m, once we have verified it for m = 2.

For then
h(X1 , X2 , . . . , Xm ) = h(X1 , X2 , . . . , Xm−1 ) + h(Xm | X1 , X2 , . . . , Xm−1 ).
Now,
p(x1 , x2 )
h(X2 | X1 ) = − ∑ p(x1 , x2 ) log
x1 ,x2 p(x1 )
=− ∑ p(x1 , x2 ) log p(x1 , x2 ) + ∑ p(x1 , x2 ) log p(x1 )
x1 ,x2 x1 ,x2

= h(X1 , X2 ) + ∑ p(x1 ) log p(x1 )

x1

= h(X1 , X2 )−h(X1 ).

Inequalities:
Entropy is a measure of uncertainty and so we should not be surprised to learn
that h(X | Y ) ≤ h(X) for all random variables X,Y – here conditioning on Y
represents providing information. Our goal is to prove this and a little more.
Let p, q be probability measures on the finite set X. We define the
Kullback-Liebler distance
p(x)
D(p||q) = ∑ p(x) log
x∈A q(x)
where A = {x : p(x) > 0}.
Lemma 25.2
D(p||q) ≥ 0
with equality iff p = q.
Proof Let
q(x)
−D(p||q) = ∑ p(x) log p(x)
x∈A
q(x)
≤ log ∑ p(x) (25.3)
x∈A p(x)
= log 1
= 0.
Inequality (25.3) follows from Jensen’s inequality and the fact that log is a
concave function. Because log is strictly concave, will have equality in (25.3)
iff p = q.
460 Entropy

It follows from this that

h(X) ≤ log |RX |. (25.4)

Indeed, let u denote the uniform distribution over RX i.e. u(x) = 1/|RX |. Then

0 ≤ D(p||u) = ∑ p(x)(log p(x) + log |RX |) = −h(X) + log |RX |.

x

We can now show that conditioning does not increase entropy.

Lemma 25.3 For random variables X,Y, Z,

h(X | Y, Z) ≤ h(X | Y ).

Taking Y to be a constant e.g. Y = 1 with probability one, we see

h(X | Z) ≤ h(X).

Proof

h(X | Y ) − h(X | Y, Z)
p(x, y) p(x, y, z)
= − ∑ p(x, y) log + ∑ p(x, y, z) log
x,y p(y) x,y,z p(y, z)
p(x, y) p(x, y, z)
= − ∑ p(x, y, z) log + ∑ p(x, y, z) log
x,y,z p(y) x,y,z p(y, z)
p(x, y, z)p(y)
= ∑ p(x, y, z) log p(x, y)p(y, z)
x,y,z

= D(px,y,z ||p(x, y)p(y, z)/p(y))

≥ 0.

Note that ∑x,y,z p(x, y)p(y, z)/p(y) = 1.

Working through the above proof we see that h(X) = h(X | Z) iff p(x, z) =
p(x)p(z) for all x, z, i.e. iff X, Z are independent.

25.2 Shearer’s Lemma

The original proof is from Chung, Frankl, Graham and Shearer [182]. The
following proof is from Radakrishnan [657].
 25.2 Shearer’s Lemma 461

Lemma 25.4 Let X = (X1 , X2 , . . . , XN ) be a (vector) random variable and

A = {Ai : i ∈ I} be a collection of subsets of a set B, where |B| = N, such
that each element of B appears in at least k members of A . For A ⊆ B, let
XA = (X j : j ∈ A). Then,
1
h(X) ≤ h(XAi ).
k∑i∈I

Proof We have, from Lemma 25.1 that

h(X) = ∑ h(X j | X1 , X2 , . . . , X j−1 ) (25.5)
j∈B

and
h(XAi ) = ∑ h(X j | X` , ` ∈ Ai , ` < j). (25.6)
j∈Ai

We sum (25.6) for all i ∈ I. Then

∑ h(XAi ) = ∑ ∑ h(X j | X` , ` ∈ Ai , ` < j)

i∈I i∈I j∈Ai

= ∑ ∑ h(X j | X` , ` ∈ Ai , ` < j) (25.7)

j∈B Ai 3 j

≥ ∑ ∑ h(X j | X1 , X2 , . . . , X j−1 ) (25.8)

j∈B Ai 3 j

≥ k ∑ h(X j | X1 , X2 , . . . , X j−1 ) (25.9)

j∈B

= kh(X). (25.10)
Here we obtain (25.8) from (25.7) by applying Lemma 25.3. We obtain (25.9)
from (25.8) and the fact that each j ∈ B appears in at least k Ai ’s. We then
obtain (25.10) by using (25.5).
 References

[1] M. Abdullah, C.Cooper and A.M. Frieze, Cover time of a random graph with
given degree sequence, Discrete Mathematics 312 (2012) 3146-3163.
[2] D. Achlioptas, J-H. Kim, M. Krivelevich and P. Tetali, Two-Coloring Random
Hypergraphs, Random Structures and Algorithms 20 (2002) 249-259.
[3] D. Achlioptas and C. Moore, Almost all Graphs with Average Degree 4 are 3-
colorable, Journal of Computer and System Sciences 67 (2003) 441-471.
[4] D. Achlioptas and C. Moore, The Chromatic Number of Random Regular
Graphs, Proceedings of RANDOM 2004.
[5] D. Achlioptas and A. Naor, The two possible values of the chromatic number of
a random graph, Annals of Mathematics 162 (2005) 1335-1351.
[6] R. Adamczak and P. Wolff, Concentration inequalities for non-Lipschitz func-
tions with bounded derivatives of higher order, Probability Theory and Related
Fields 161 (2014) 1-56.
[7] W. Aiello, A. Bonato, C. Cooper, J. Janssen and P. Pralat, A spatial web graph
model with local influence regions, Internet Mathematics 5 (2009) 175-196.
[8] M. Ajtai, J. Komlós and E. Szemerédi, Largest random component of a k-cube,
Combinatorica 2 (1982) 1-7.
[9] M. Ajtai, J. Komlós and E. Szemerédi, Topological complete subgraphs in ran-
dom graphs, Studia Sci. Math. Hungar. 14 (1979) 293-297.
[10] M. Ajtai, J. Komlós and E. Szemerédi, The longest path in a random graph,
Combinatorica 1 (1981) 1-12.
[11] M. Ajtai, J. Komlós and E. Szemerédi. The first occurrence of Hamilton cycles
in random graphs, Annals of Discrete Mathematics 27 (1985) 173-178.
[12] D. Aldous, Exchangeability and related topics,Lecture Notes in Mathematics,
1117, Springer Verlag, New York (1985)
[13] D. Aldous, Asymptotics in the random assignment problem, Probability Theory
and Related Fields 93 (1992) 507-534.
[14] D. Aldous, Asymptotic fringe distributions for general families of random trees.
Ann. Appl. Probab. 1 (1991) 228-266.
[15] D. Aldous, Brownian excursions, critical random graphs and the multiplicative
coalescent, Annals of Probability 25 (1997) 812-854.
[16] D. Aldous, The ζ (2) limit in the random assignment problem, Random Struc-
tures and Algorithms 4 (2001) 381-418.
[17] D. Aldous, G. Miermont and J. Pitman, Brownian bridge asymptotics for random
p-mappings, Electronic J. Probability 9 (2004) 37-56.
[18] D. Aldous, G. Miermont and J. Pitman, Weak convergence of random p-
mappings and the exploration process of inhomogeneous continuum random
trees, Probab. Th. Related Fields 133 (2005) 1-17.

462
 References 463

[19] P. Allen, J. Böttcher, Y. Kohayakawa and Y. Person, Tight Hamilton cycles in

random hypergraphs, see arxiv.org.
[20] N. Alon, Choice numbers of graphs; a probabilistic approach, Combinatorics,
Probability and Computing 1 (1992) 107-114.
[21] N. Alon, Restricted colorings of graphs, in Surveys of Combinatorics 1993, Lon-
don Mathematical Society Lecture Notes Series 187 (K. Walker, ed.), Cambridge
University Press 1993, 1-33.
[22] N.Alon, A note on network reliability, In Discrete Probability and Algorithms
(Minneapolis, MN, 1993), IMA Vol. Math. Appl., 72, Springer, New York (1995)
11-14.
[23] N. Alon, M. Capalbo, Y. Kohayakawa, V. Rödl, A. Ruciński and E. Szemerédi,
Universality and tolerance (extended abstract), In 41st Annual Symposium on
Foundations of Computer Science (Redondo Beach, CA, 2000), IEEE Computer
Society Press Los Alamitos, CA, 2000, pp. 14-21.
[24] N. Alon, P. Pralat, N. Wormald, Cleaning regular graphs with brushes, SIAM
Journal on Discrete Mathematics 23 (2008) 233-250.
[25] N. Alon and Z. Füredi, Spanning subgraphs of random graphs, Graphs and Com-
binatorics 8 (1992) 91-94.
[26] N. Alon and M. Krivelevich, The concentration of the chromatic number of a
random graph, Combinatorica 17 (1997) 303-313.
[27] N. Alon, M. Krivelevich and B. Sudakov, List coloring of random and pseudo
random-graphs, Combinatorica 19 (1999) 453-472.
[28] N. Alon, M. Krivelevich and B. Sudakov, Finding a large hidden clique in a
random graph, Random Structures and Algorithm 13 (1998) 457-466.
[29] N. Alon, M. Krivelevich and V. H. Vu, On the concentration of eigenvalues of
random symmetric matrices, Israel Journal of Mathematics 131 (2002) 259-267.
[30] N. Alon and J. Spencer, The probabilistic method, Third Edition, John Wiley and
Sons (2008).
[31] N. Alon and B. Sudakov, Increasing the chromatic number of a random graph,
Journal of Combinatorics 1 (2010) 345–356.
[32] N. Alon and R. Yuster, Threshold functions for H-factors, Combinatorics, Prob-
ability and Computing 2 (1993) 137-144.
[33] A. Amit and N. Linial, Random Graph Coverings I: General Theory and Graph
Connectivity, Combinatorica 22 (2002), 1-18.
[34] A. Amit and N. Linial, Random Lifts of Graphs II: Edge Expansion, Combina-
torics Probability and Computing 15 (2006) 317-332.
[35] A. Amit, N. Linial and J. Matoušek, Random Lifts of Graphs III: Independence
and Chromatic Number, Random Structures and Algorithms 20 (2002) 1-22.
[36] O. Angel, A. Flaxman and D. Wilson, A sharp threshold for minimum bounded-
depth and bounded-diameter spanning trees and steiner trees in random net-
works, Combinatorica (2012) 1-33.
[37] S. Anoulova, J. Bennies, J. Lenhard, D. Metzler, Y. Sung and A. Weber (1999)
The American Mathematical Monthly 106 (199) 345-351.
[38] T. Apostol, Mathematical Analysis, Second Edition, Addison-Wesley, 1973.
[39] J. Aronson, A.M. Frieze and B. Pittel, Maximum matchings in sparse random
graphs: Karp-Sipser revisited, Random Structures and Algorithms 12 (1998)
111-178.
464 References

[40] R. Arratia, A.D. Barbour and S. Tavaré, A tale of three couplings: Poisson-
Dirichlet and GEM approximations for random permutations, Combinatorics,
Probability and Computing 15 (2006) 31-62.
[41] R. Arratia and S. Tavaré, The cycle structure of random permutations, The An-
nals of Probability 20 (1992) 1567-1591.
[42] C. Avin and G. Ercal, On the Cover Time of Random Geometric Graphs, Au-
tomata, Languages and Programming, Lecture Notes in Computer Science 3580
(2005) 677-689.
[43] L.Babai, P. Erdős, S.M. Selkow, Random graph isomorphism, SIAM Journal on
Computing 9 (1980) 628-635.
[44] L. Babai, M. Simonovits and J. Spencer, Extremal Subgraphs of Random
Graphs, Journal of Graph Theory 14 (1990) 599-622.
[45] A. Backhaus and T.F. Móri, Local degree distribution in scale free random
graphs, Electronic Journal of Probability 16 (2011) 1465-1488.
[46] D. Bal, P. Bennett, A. Dudek and A.M. Frieze, The t-tone chromatic number of
random graphs, see arxiv.org.
[47] D. Bal, P. Bennett, A.M. Frieze and P. Pralat, Power of k choices and rainbow
spanning trees in random graphs, see arxiv.org.
[48] J. Balogh, T. Bohman and D. Mubayi, Erdős-Ko-Rado in Random Hypergraphs,
Combinatorics, Probability and Computing 18 (2009) 629-646.
[49] D. Bal and A.M. Frieze, Rainbow Matchings and Hamilton Cycles in Random
Graphs, see arxiv.org.
[50] J. Balogh, B. Bollobás, M. Krivelevich, T. Müeller and M. Walters, Hamilton
cycles in random geometric graphs, Annals of Applied Probability 21 (2011)
1053-1072.
[51] P. Balister and B. Bollobás, Percolation in the k-nearest neighbour graph.
[52] P. Balister, B. Bollobás, A. Sarkar and M. Walters, Connectivity of random k-
nearest neighbour graphs, Advances in Applied Probability 37 (2005) 1-24.
[53] P. Balister, B. Bollobás, A. Sarkar and M. Walters, A critical constant for the
k-nearest neighbour model, Advances in Applied Probability 41 (2009) 1-12.
[54] F. Ball, D. Mollison and G. Scalia-Tomba, Epidemics with two levels of mixing,
Ann. Appl. Prob. 7 (1997) 46-89.
[55] J. Balogh, R. Morris and W. Samotij, Independent sets in hypergraphs, see
arxiv.org.
[56] C. Banderier, P. Flajolet, G. Schaeffer, and M. Soria, Random maps, coalesc-
ing saddles, singularity analysis, and Airy phenomena, Random Structures and
Algorithms 19 (2001) 194-246.
[57] L. Barabási and R. Albert, Emergence of scaling in random networks, Science
286 (1999) 509-512.
[58] A.D Barbour, Poisson convergence and random graph, Math. Proc. Camb. Phil.
Soc. 92 (1982) 349-359.
[59] A.D. Barbour, L. Holst and S. Janson, Poisson Approximation, 1992, Oxford
University Press, Oxford.
[60] A.D. Barbour, S. Janson, M. Karoński and A. Ruciński, Small cliques in random
graphs, Random Structures and Algorithms 1 (1990) 403-434.
 References 465

[61] A.D. Barbour, M. Karoński and A. Ruciński, A central limit theorem for de-
composable random variables with applications to random graphs, J. Combin.
Theory B 47 (1989) 125-145.
[62] A.D. Barbour and G. Reinert, The shortest distance in random multi-type inter-
section graphs, Random Structures Algorithms 39 (2011) 179-209.
[63] T. Bartnicki, J. Grytczuk, H.A. Kierstead and X. Zhu, The Map-Coloring Game,
American Mathematical Monthly, 114 (2007) 793-803.
[64] M. Bayati, D. Gamarnik and P. Tetali, Combinatorial approach to the interpola-
tion method and scaling limits in sparse random graphs, Annals of Probability
41 (2013) 4080-4115.
[65] J. Beck, Combinatorial Games: Tic-Tac-Toe theory, Encyclopedia of Mathemat-
ics and its Applications 114, Cambridge University Press 2008.
[66] M. Bednarska and T. Łuczak, Biased positional games for which random strate-
gies are nearly optimal, Combinatorica 20 (2000) 477-488.
[67] M. Behrisch, Component evolution in random intersection graphs, The Elec-
tronic Journal of Combinatorics 14 (2007) #R17
[68] M. Behrisch, A. Taraz and M. Ueckerdt, Coloring random intersection graphs
and complex networks, SIAM J. Discrete Math. 23 (2009) 288-299.
[69] I. Benjamini, G. Kozma and N. Wormald, The mixing time of the giant compo-
nent of a random graph, Random Structures and Algorithms 45 (2014) 383-407.
[70] I. Benjamini, I. Shinkar, G. Tsur, Acquaintance time of a graph, see arxiv.org.
[71] E.A. Bender and E.R. Canfield, The asymptotic number of labelled graphs with
given degree sequences, Journal of Combinatorial Theory A 24 (1978) 296-307.
[72] E.A. Bender, E.R. Canfield and B.D. McKay, The asymptotic number of labelled
connected graphs, Random Structures and Algorithms 1 (1990) 127-169.
[73] E.A. Bender and E.R. Canfield, The number of rooted maps on an orientable
surface. Journal of Combinatorial Theory B 53 (1991) 293-299.
[74] A. Bender, Z. Gao, and N. C. Wormald, The number of labeled 2-connected
planar graphs, Electronic Journal of Combinatorics 9 (2002).
[75] P. Bennett and T. Bohman, A note on the random greedy independent set algo-
rithm, see arxiv.org.
[76] S. Ben-Shimon, M. Krivelevich and B. Sudakov, Local resilience and Hamil-
tonicity Maker-Breaker games in random regular graph, Combinatorics, Proba-
bility and Computing 20 (2011) 173-211.
[77] S. Ben-Shimon, M. Krivelevich and B. Sudakov, On the resilience of Hamil-
tonicity and optimal packing of Hamilton cycles in random graphs, SIAM Jour-
nal of Discrete Mathematics 25 (2011) 1176-1193.
[78] S. Ben-Shimon, A. Ferber, D. Hefetz and M. Krivelevich, Hitting time results
for Maker-Breaker games, Random Structures and Algorithms 41 (2012) 23-46.
[79] S. Berg, Random contact processes, snowball sampling and and factorial series
distributions, Journal of Applied Probability 20 (1983) 31-46.
[80] C. Berge, Graphs and Hypergraphs, North-Holland publishing com-
pany,Amsterdam 1973.
[81] F. Bergeron, P. Flajolet and B. Salvy, Varieties of increasing trees, Lecture Notes
in Computer Science 581 (1992) 24-48.
[82] J.Bertoin, G. Uribe Bravo, Super-critical percolation on large scale-free random
trees, The Annals of Applied Probability 25 (2015) 81-103.
466 References

[83] S.N. Bernstein, Theory of Probability, (Russian), Moscow, 1927.

[84] J. Bertoin, Sizes of the largest clusters for supercritical percolation on random
recursive trees, Random Structures Algorithms 44 (2014) 29-44.
[85] A. Beveridge, A. Dudek, A.M. Frieze, T.Müller and M. Stojaković, Maker-
Breaker games on random geometric graphs, see arxiv.org in Random Structures
and Algorithms.
[86] A.Beveridge, A.M. Frieze and C.McDiarmid, Random minimum length span-
ning trees in regular graphs, Combinatorica 18 (1998) 311-333.
[87] S. Bhamidi and R. van der Hofstad, Weak disorder asymptotics in the stochastic
mean-field model of distance, Annals of Applied Probability 22 (2012) 29-69.
[88] S. Bhamidi, R. van der Hofstad and G. Hooghiemstra, Universality for first pas-
sage percolation on sparse uniform and rank-1 random graphs, see arxiv.org.
[89] S. Bhamidi, S. Sen and X. Wang, Continuum limit of inhomogeneous random
graphs, see arxiv.org.
[90] Ph. Biane, J. Pitman and M. Yor, Probability laws related to the Jacobi theta and
Riemann zeta functions, and brownian excursions, Bulletin of the AMS 38 (2001)
435-465.
[91] S.R. Blackbourn and S. Gerke, Connectivity of the uniform random intersection
graph, Discrete Mathematics 309 (2009) 5130-5140.
[92] P. Blackwell, M. Edmonson-Jones and J. Jordan, Spectra of adjacency matrices
of random geometric graphs, see arxiv.org.
[93] M. Bloznelis, Degree distribution of a typical vertex in a general random inter-
section graph, Lithuanian Mathematical Journal 48 (2008) 38-45.
[94] M. Bloznelis, Component evolution in general random intersection graphs, SIAM
J. Discrete Mathematics 24 (2010) 639-654.
[95] M. Bloznelis, A random intersection digraph: indegree and outdegree distribu-
tions, Discrete Mathematics 310 (2010) 2560-2566.
[96] M. Bloznelis, The largest component in an inhomogeneous random intersec-
tion graph with clustering, The Electronic Journal of Combinatorics 17 (2010)
#R110.
[97] M. Bloznelis and J. Damarackas, Degree distribution of an inhomogeneous ran-
dom intersection graph, The Electronic Journal of Combinatorics 20(3) (2013)
#P3.
[98] M. Bloznelis, E. Godehardt, J. Jaworski, V. Kurauskas, and K. Rybarczyk, Re-
cent progress in complex network analysis - models of random intersection
graphs, In:B. Lausen, S. Krolak-Schwerdt, M. Bhmer, Eds., Data Science, Learn-
ing by Latent Structures, and Knowledge Discovery, Springer (2015) 59-68.
[99] M. Bloznelis, E. Godehardt, J. Jaworski, V. Kurauskas, and K. Rybarczyk, Re-
cent progress in complex network analysis - properties of random intersection
graphs, In:B. Lausen, S. Krolak-Schwerdt, M. Bhmer, Eds., Data Science, Learn-
ing by Latent Structures, and Knowledge Discovery, Springer (2015) 69-78.
[100] M. Bloznelis and F. Goetze, Preferred attachment model of affiliation network,
Journal of Statistical Physics 156 (2014) 800-821.
[101] M. Bloznelis, F. Göetze and J. Jaworski, Birth of a strongly connected giant in
an inhomogeneous random digraph, Journal of Applied Probability 49 (2012),
601-611.
 References 467

[102] M. Bloznelis, J. Jaworski and K. Rybarczyk, Component evolution in a secure

wireless sensor network, Networks 53 (2009) 19-26.
[103] M. Bloznelis and M. Karoński, Random intersection graph process, to appear in
Internet Mathematics.
[104] M. Bloznelis and T. Łuczak, Perfect matchings in random intersection graphs,
Acta Mathematica Hungarica, 138 (2013) 15-33.
[105] M. Bloznelis and I. Radavicius, A note on Hamiltonicity of uniform random
intersection graphs, Lithuanian Mathematical Journal, 51 (2011) 155-161.
[106] M. Bloznelis and K. Rybarczyk, k-connectivity of uniform s-intersection graphs,
Discrete Mathematics, 333 (2014) 94-100.
[107] M. Bode, N. Fountoulakis and T. Müller, The probability that the hyperbolic
random graph is connected, see arxiv.org.
[108] M. Bode, M. Fountoulakis, T. Müller, On the largest component of a hyperbolic
model of complex networks, see arxiv.org.
[109] E. Bodine-Baron, B. Hassibi and A. Wierman, Generalizing Kronecker graphs
in order to model searchable networks, in 47th Annual Allerton Conference on
Communication, Control, and Computing (2009) 194-201.
[110] E. Bodine-Baron, B. Hassibi and A. Wierman, Distance-dependent Kronecker
graphs for modeling social networks, IEEE J. of Selected Topics in Signal Pro-
cessing 4 (2010) 718-731.
[111] T. Bohman, The Triangle-Free Process, Advances in Mathematics 221 (2009)
1653-1677.
[112] T. Bohman, C. Cooper, A.M. Frieze, R. Martin and M. Ruszinko, On Ran-
domly Generated Intersecting Hypergraphs, Electronic Journal on Combina-
torics 2003) R29.
[113] T. Bohman, A.M. Frieze, R. Martin and M. Ruszinko and C. Smyth, On Ran-
domly Generated Intersecting Hypergraphs II, Random Structures and Algo-
rithms 30 (2007) 17-34.
[114] T. Bohman and A.M. Frieze, Avoiding a giant component, Random Structures
and Algorithms (2001) 75-85.
[115] T. Bohman and A.M. Frieze, Hamilton cycles in 3-out, Random Structures and
Algorithms 35 (2009) 393-417.
[116] T. Bohman, A.M. Frieze and E. Lubetzky, A note on the random greedy triangle-
packing algorithm, Journal of Combinatorics 1 (2010) 477-488.
[117] T. Bohman, A.M. Frieze and E. Lubetzky, Random triangle removal, see
arxiv.org.
[118] T. Bohman, A.M. Frieze, O. Pikhurko and C. Smyth, Anti-Ramsey Properties of
Random Graphs, Journal of Combinatorial Theory B 100 (2010) 299-312.
[119] T. Bohman, A.M. Frieze and B.Sudakov, The game chromatic number of random
graphs, Random Structures and Algorithms 32 (2008) 223-235.
[120] T. Bohman and P. Keevash, Dynamic concentration of the triangle-free process,
see arxiv.org.
[121] T. Bohman and D. Kravitz, Creating a giant component, Combinatorics, Proba-
bility and Computing 15 (2006) 489-511.
[122] T. Bohman, D. Mubayi and M. Picolleil, The independent neighborhoods pro-
cess, see arxiv.org.
468 References

[123] B. Bollobás, A probabilistic proof of an asymptotic formula for the number of

labelled graphs, European Journal on Combinatorics 1 (1980) 311-316.
[124] B. Bollobás, Random Graphs, In Combinatorics, Proceedings, Swansea (1981)
80-102.
[125] B. Bollobás, The diameter of random graphs, Tranasactions of the American
Mathematical Society 267 (1981) 41-52.
[126] B. Bollobás, Degree sequence of random graphs, Discrete Mathematics 13
(1981) 1-19.
[127] B. Bollobás, Long paths in sparse random graphs, Combinatorica 2 (1982) 223-
228.
[128] B. Bollobás, Vertices of given degree in a random graph, Journal of Graph The-
ory 6 (1982) 147-155.
[129] B. Bollobás, Distinguishing vertices of random graphs, Annals of Discrete Math-
ematics 13 (1982) 33-50.
[130] B. Bollobás, The evolution of random graphs, Transactions of the Americam
Mathematical Society 286 (1984) 257-274.
[131] B. Bollobás, The evolution of sparse graphs, Proceedings of a Cambridge Com-
binatorial Conference in honour of Paul Erős (1984) 35-57.
[132] B. Bollobás, Random Graphs, First Edition, Academic Press, London 1985, Sec-
ond Edition, Cambridge University Press, 2001.
[133] B. Bollobás, The chromatic number of random graphs, Combinatorica 8 (1988)
49-56.
[134] B. Bollobás, Modern Graph Theory, (1998) Springer, New York.
[135] B. Bollobás, Complete matchings in random subgraphs of the cube, Random
Structures and Algorithms 1 (1990) 95-104.
[136] B. Bollobás, P. Catlin and P. Erdős, Hadwigers conjecture is true for almost every
graph, European Journal of Combinatorics 1 (1980) 195-199.
[137] B. Bollobás, C. Cooper, T.I. Fenner and A.M. Frieze, On Hamilton cycles in
sparse random graphs with minimum degree at least k, Journal of Graph Theory
34 (2000) 42-59.
[138] B. Bollobás and P. Erdős, Cliques in random graphs, Mathematical Proceedings
of the Cambridge Philosophical Society 80 (1976) 419-427.
[139] B. Bollobás, T.I. Fenner and A.M. Frieze, An algorithm for finding hamilton
paths and cycles in random graphs, Combinatorica 7 (1987) 327-341.
[140] B. Bollobás, T.I. Fenner and A.M. Frieze, Hamilton cycles in random graphs
with minimal degree at least k, in A tribute to Paul Erdos, edited by A.Baker,
B.Bollobás and A.Hajnal (1990) 59-96.
[141] B. Bollobás, T.I. Fenner and A.M. Frieze, Long cycles in sparse random graphs,
In Graph Theory and Combinatorics, Proceedings, Cambridge Combinatorial
Conference in honour of Paul Erdős (B. Bollobás Ed.) (1984) 59-64.
[142] B. Bollobás and A. Frieze, On matchings and hamiltonian cycles in random
graphs, Annals of Discrete Mathematics 28 (1985) 23-46.
[143] B. Bollobás and A. Frieze, Spanning Maximal Planar Subgraphs of Random
Graphs, Random Structures and Algorithms 2 (1991) 225-231.
[144] B. Bollobás, S. Janson and O. Riordan, The phase transition in inhomogeneous
random graphs, Random Structures and Algorithms 31 (2007) 3-122.
 References 469

[145] B. Bollobás, D. Gamarnik, O. Riordan and B. Sudakov, On the Value of a Ran-

dom Minimum Length Steiner Tree, Combinatorica 24 (2004) 187-207.
[146] B. Bollobás, Y. Kohayakawa and T. Łuczak, The evolution of random subgraphs
of the cube, Random Structures and Algorithms 3 (1992) 55-90.
[147] B. Bollobás, Y. Kohayakawa and T. Łuczak, Connectivity properties of random
subgraphs of the cube, Random Structures and Algorithms 6 (1995) 221-230.
[148] B. Bollobás, R. Kozma and D. Miklós, Handbook of Large-Scale Random Net-
works, Springer and János Bolyai Mathematical Society, Budapest.
[149] B. Bollobás and O. Riordan, Constrained graph processes, Electronic Journal of
Combinatorics 7 (2000) R18.
[150] B. Bollobás and O. Riordan, The diameter of a scale free random graph, Combi-
natorica 24 (2004) 5-34.
[151] B. Bollobás, O. Riordan, J. Spencer and G. Tusnády, The degree sequence of a
scale-free random graph process, Random Structures and Algorithms 18 (2001)
279-290.
[152] B. Bollobás and A. Thomason, Hereditary and monotone properties of graphs, in
The mathematics of Paul Erdős, II, eds. R.L. Graham and J. Nešetřil, Algorithms
and Combinatorics, 14, Springer, Berlin (1997) 70-78.
[153] A. Bonato and P. Pralat, Probabilistic Graph Searching, see arxiv.org.
[154] C. Borgs, M. Brautbar, J. Chayes, S. Khanna and B. Lucier, The Power of Local
Information in Social Networks, see arxiv.org.
[155] J. Bourgain and G. Kalai, Influences of variables and threshold intervals under
group symmetries, Geometry and Functional Analysis 7 (1997) 438-461.
[156] J. Bourgain, Appendix to [322].
[157] M. Bradonjić, R. Elsässer, T. Friedrich, T. Sauerwald and A. Stauffer, Efficient
broadcast on random geometric graphs, 21st Annual ACM-SIAM Symposium on
Discrete Algorithms (2010) 1412-1421.
[158] M. Bradonjić, A. Hagberg, N.W. Hengartner and A.G. Percus, Component evo-
lution in general random intersection graphs, In Algorithms and Models for the
Web-Graph, LNCS 6516, Springer (2010) 36-43.
[159] G.R. Brightwell and M.J. Luczak Vertices of high degree in the preferential at-
tachment tree, Elect. J. Probab. 17 (2012) 43pp.
[160] G.R. Brightwell, K. Panagiotou and A. Steger, Extremal subgraphs of random
graphs, Random Structures and Algorithms 41 (2012) 147-178.
[161] T. Britton, M. Deijfen, A. Lageras and M. Lindholm, Epidemics on random
graphs with tunable clustering J. Applied Probability, 45 (2008) 743-756.
[162] T. Britton, M. Deijfen, and A. Martin-Löf, Generating simple random graphs
with prescribed degree distribution, J. Stat. Phys. 124 (2006) 1377-1397.
[163] A. Broder, A.M. Frieze and E. Shamir, Finding hidden Hamilton cycles, Random
Structures and Algorithms 5 (1994) 395-410.
[164] A. Broder, A.M. Frieze, S. Suen and E. Upfal, An efficient algorithm for the
vertex-disjoint paths problem in random graphs, Proceedings of SODA ’96, 261-
268.
[165] A. Broder, A.M. Frieze, S. Suen and E. Upfal, Optimal construction of edge-
disjoint paths in random graphs, SIAM Journal on Computing 28 (1999) 541-
574.
470 References

[166] T. Bui, S. Chaudhuri, T. Leightonand M. Sipser, Graph bisection algorithms with

good average case behavior, Combinatorica 7 (1987) 171-191.
[167] J. D. Burtin, Extremal metric characteristics of a random graph I, Teor. Veroyat-
nost. i Primenen 19 (1974) 740-754.
[168] J. D. Burtin, Extremal metric characteristics of a random graph II, Teor. Veroyat-
nost. i Primenen 20 (1975) 82-99.
[169] Y.D. Burtin, On the probability of connectedness of a random subgraph of the
n-cube, (in Russian) Problemy Peredaci Informacii 13 (1977) 90-95.
[170] Y.D. Burtin, On a simple formula for a random mappings and its applications, J.
Applied Probability 17 (1980) 403-414.
[171] D.S. Callaway, J.E. Hopcroft, J.M. Kleinberg, M.E.J. Newman and S.H. Stro-
gatz, Are randomly grown graphs really random? Phys.Rev. E 64 (2001) 041902.
[172] E. Candellero and N. Fountoulakis, Clustering and the hyperbolic geometry of
complex networks, see arxiv.org.
[173] Y. Caro, A. Lev, Y. Roditty, Z. Tuza, and R. Yuster, On rainbow connection,
Electronic Journal of Combinatorics 15 (2008).
[174] J. Cerny and A. Teixeira, Critical window for the vacant set left by random walk
on random regular graphs, see arxiv.org.
[175] D. Chakrabarti, Y. Zhan and C. Faloutsos, R-MAT: A recursive model for graph
mining, In: Proceedings of 4th International Conference on Data Mining (2004)
442-446.
[176] S. Chatterjee and A. Dembo, Nonlinear large deviations, see arxiv.org.
[177] P. Chebolu and A.M. Frieze, Hamilton cycles in random lifts of complete di-
rected graphs, SIAM Journal on Discrete Mathematics 22 (2008) 520-540.
[178] P.Chebolu, A.M. Frieze and P.Melsted, Finding a Maximum Matching in a
Sparse Random Graph in O(n) Expected Time, Journal of the ACM 57 (2010)
1-27.
[179] L.H.Y. Chen, Poisson approximation for dependent trials, Annals of Probability
3 (1975) 534-545.
[180] W-C. Chen and W-C. Ni, Heap-ordered trees, 2-partitions and continued frac-
tions, European Journal of Combinatorics 15 (1994) 513-517.
[181] W. Chen, W. Fang, G. Hu and M. Mahoney, On the hyperbolicity of small-world
and tree-like random graphs, ISAAC 2012.
[182] F. Chung, F. Frank, R. Graham and J. Shearer, Some intersection theorems for
ordered sets and graphs, Journal of Combinatorial Theory A 43 (86) 23-37.
[183] F. Chung and L. Lu, Connected components in random graphs with given ex-
pected degree sequence Annals of Combinatorics, 6 (2002) 125-145.
[184] F. Chung and L. Lu. The average distances in random graphs with given expected
degree, Internet Math. 1 (2003) 91-113.
[185] F. Chung and L. Lu, The volume of the giant component of a random graph
with given expected degrees, SIAM Journal on Discrete Mathematics 20 (2006)
395-411.
[186] F. Chung, L. Lu and V. Vu, Eigenvalues of random power law graphs, Annals of
Combinatorics 7 (2003) 21-33.
[187] F. Chung, L. Lu and V. Vu, The spectra of random graphs with expected degrees,
Proceedings of the National Academy of Sciences 100 (2003) 6313-6318.
 References 471

[188] V. Chvátal, DETERMINING THE STABILITY NUMBER OF A GRAPH,

SIAM Journal on Computing 6 (1977) 643-662.
[189] V. Chvátal, Almost all graphs with 1.44n edges are 3-colorable, Random Struc-
tures and Algorithms 2 (1991) 11-28.
[190] V. Chvátal and P. Erdős, Biased positional games, Annals of Discrete Mathemat-
ics 2 (1978) 221-228.
[191] A. Coja–Oghlan, Graph Partitioning via Adaptive Spectral Techniques, Combi-
natorics, Probability and Computing 19 (2010) 227-284.
[192] A. Coja–Oghlan, On the Laplacian eigenvalues of G(n, p), Combinatorics, Prob-
ability and Computing 16 (2007) 923-946.
[193] A. Coja–Oghlan, Upper bounding the k-colorability threshold by counting cov-
ers, Electronic Journal of Combinatorics 20 (2013) P32.
[194] A. Coja–Oghlan, K. Panagiotou and A. Steger, On the chromatic number of ran-
dom graphs, Journal of Combinatorial Theory, B 98 (2008) 980-993.
[195] A. Coja–Oghlan and D. Vilenchik, Chasing the k-colorability threshold, see
arxiv.org.
[196] N. A. Cook and A. Dembo, Large deviations of subgraph counts for sparse Erdő
s–Rényi graphs, see arxiv.org.
[197] A. Collevecchio, A. Mehrabian and N. Wormald, Longest paths in random Apol-
lonian networks and largest r-ary subtrees of random d-ary recursive trees, see
arxiv.org.
[198] P. Condon, A. Espuny Dáz, J. Kim, D. Kühn and D. Osthus, Resilient degree
sequences with respect to Hamilton cycles and matchings in random graphs, see
arxiv.org.
[199] O. Cooley, M. Kang and C. Koch, The size of the giant component in random
hypergraphs, see arxiv.org.
[200] D. Conlon, T. Gowers, Combinatorial theorems in sparse random sets, see
arxiv.org.
[201] C. Cooper, Pancyclic Hamilton cycles in random graphs, Discrete Mathematics
91 (1991) 141-148.
[202] C. Cooper, On the thickness of sparse random graphs, Combinatorics Probability
and Computing 1 (1992) 303-309.
[203] C. Cooper, 1-pancyclic Hamilton cycles in random graphs, Random Structures
and Algorithms 3 (1992) 277-287.
[204] C. Cooper, The size of the cores of a random graph with a given degree sequence,
Random Structures and Algorithms (2004) 353-375.
[205] C. Cooper, The age specific degree distribution of web-graphs, Combinatorics
Probability and Computing 15 (2006) 637 - 661.
[206] C. Cooper and A.M. Frieze, On the number of hamilton cycles in a random
graph, Journal of Graph Theory 13 (1989) 719-735.
[207] C. Cooper and A.M. Frieze, Pancyclic random graphs, in Random Graphs, (Ed:
Karonski M, Javorski J, and Rucinski A.) Wiley (1990).
[208] C. Cooper and A.M. Frieze, The limiting probability that a-in, b-out is strongly
connected, Journal of Combinatorial Theory B 48 (1990) 117-134.
[209] C. Cooper and A.M. Frieze, Multicoloured Hamilton cycles in random graphs:
an anti-Ramsey threshold, Electronic Journal of Combinatorics 2, R19.
472 References

[210] C. Cooper and A.M. Frieze, On the connectivity of random k-th nearest neigh-
bour graphs, Combinatorics, Probability and Computing 4 (196) 343-362.
[211] C. Cooper and A.M. Frieze, Hamilton cycles in random graphs and directed
graphs, Random Structures and Algorithms 16 (2000) 369-401.
[212] C. Cooper and A.M. Frieze, Multi-coloured Hamilton cycles in randomly
coloured random graphs, Combinatorics, Probability and Computing 11 (2002)
129-134.
[213] C. Cooper and A.M. Frieze, On a general model of web graphs Random Struc-
tures and Algorithms 22 (2003) 311-335.
[214] C. Cooper and A.M. Frieze, The Size of the Largest Strongly Connected Com-
ponent of a Random Digraph with a Given Degree Sequence, Combinatorics,
Probability and Computing 13 (2004) 319-337.
[215] C. Cooper and A.M. Frieze, The cover time of random regular graphs, SIAM
Journal on Discrete Mathematics 18 (2005) 728-740.
[216] C. Cooper and A.M. Frieze, The cover time of the preferential attachment graph,
Journal of Combinatorial Theory B 97 (2007) 269-290.
[217] C. Cooper and A.M. Frieze, The cover time of the giant component of a random
graph, Random Structures and Algorithms 32 (2008) 401-439.
[218] C. Cooper and A.M. Frieze, The cover time of random geometric graphs, Ran-
dom Structures and Algorithms 38 (2011) 324-349.
[219] C. Cooper and A.M. Frieze, Stationary distribution and cover time of random
walks on random digraphs, Journal of Combinatorial Theory B 102 (2012) 329-
362.
[220] C. Cooper and A.M. Frieze, Component structure induced by a random walk on
a random graph, Random structures and Algorithms 42 (2013) 135-158.
[221] C. Cooper and A.M. Frieze, Long paths in random Apollonian networks, see
arxiv.org.
[222] C. Cooper, A.M. Frieze, N. Ince, S. Janson and J. Spencer, On the length of a
random minimum spanning tree, see arxiv.org.
[223] C.Cooper, A.M. Frieze and M.Krivelevich, Hamilton cycles in random graphs
with a fixed degree sequence, SIAM Journal on Discrete Mathematics 24 (2010)
558-569.
[224] C.Cooper, A.M. Frieze and E. Lubetzky, Cover time of a random graph with
given degree sequence II: Allowing vertices of degree two, Random Structures
and Algorithms 45 (2014) 627-674.
[225] C.Cooper, A.M. Frieze and M. Molloy, Hamilton cycles in random regular di-
graphs, Combinatorics, Probability and Computing 3 (1994) 39-50.
[226] C. Cooper, A,M, Frieze and P. Pralat, Some typical properties of the Spatial
Preferred Attachment model, Internet Mathematics 10 (2014), 27-47.
[227] C. Cooper, A.M. Frieze and B. Reed, Perfect matchings in random r-regular, s-
uniform hypergraphs, Combinatorics, Probability and Computing 5 (1996) 1-15.
[228] C.Cooper, A.M. Frieze and T. Radzik, The cover time of random walks on ran-
dom uniform hypergraphs, Proceeedings of SIROCCO 2011, 210-221.
[229] C. Cooper, A.M. Frieze and B. Reed, Random regular graphs of non-constant
degree: connectivity and Hamilton cycles, Combinatorics, Probability and Com-
puting 11 (2002) 249-262.
 References 473

[230] D. Coppersmith and G. Sorkin, Constructive bounds and exact expectations for
the random assignment problem, Random Structures and Algorithms 15 (1999)
133-144.
[231] T. Cover and J. Thomas, Elements of information theory, 2nd Edition, Wiley-
Interscience, 2006.
[232] V. Dani and C. Moore, Independent sets in random graphs from the weighted
second moment method, Proceedings of RANDOM 2011 (2011) 472-482.
[233] A. Darrasse, H-K. Hwang and M. Soria, Shape measures of random increasing
k-trees, manuscript (2013).
[234] M. Deijfen and W. Kets, Random intersection graphs with tunable degree distri-
bution and clustering, Probab. Engrg. Inform. Sci. 23 (2009) 661-674.
[235] D. Dellamonica, Jr., Y. Kohayakawa, V. Rödl, and A. Ruciński, An improved
upper bound on the density of universal random graphs, Random Structures and
Algorithms, see arxiv.org.
[236] B. DeMarco and J. Kahn, Tight upper tail bounds for cliques, Random Structures
and Algorithms 41 (2012)
[237] B. DeMarco and J. Kahn, Mantel’s theorem for random graphs, see arxiv.org.
[238] B. DeMarco and J. Kahn, Turán’s theorem for random graphs, see arxiv.org.
[239] L. Devroye, Branching processes in the analysis of the heights of trees, Acta
Informatica 24 (1987) 277-298.
[240] L. Devroye, O. Fawzi and N. Fraiman, Depth properties of scaled attachment
random recursive tree, Random Structures and Algorithms 41 (2012) 66-98.
[241] L. Devroye and N. Fraiman, Connectivity in inhomogeneous random graphs,
Random Structures and Algorithms 45 (2013) 408-420.
[242] L. Devroye and H-K. Hwang, Width and mode of the profile for some random
trees of logarithmic height, The Annals of Applied Probability 16 (2006) 886-
918.
[243] L. Devroye and S. Janson, Protected nodes and fringe subtrees in some random
trees, Electronic Communications in Probability 19 (2014) 1-10.
[244] L. Devroy and J. Lu, The strong convergence of maximal degrees in uniform
random recursive trees and dags, Random Structure and Algorithms 7 (1995)
1-14.
[245] J. Dı́az, D. Mitsche and X. Pérez-Giménez, Sharp threshold for hamiltonicity
of random geometric graphs, SIAM Journal on Discrete Mathematics 21 (2007)
57-65.
[246] J. Dı́az, X. Pérez-Giménez, J. Serna and N. Wormald, Walkers on the Cycle and
the Grid, SIAM Journal on Discrete Mathematics 22 (2008) 747-775.
[247] J. Dı́az, F. Grandoni and A. Marchetti-Spaccamela, Balanced Cut Approximation
in Random Geometric Graphs, Theoretical Computer Science 410 (2009) 527-
536.
[248] J. Dı́az, M.D. Penrose, J. Petit and M.J. Serna, Approximating layout problems
on random geometric graphs, Journal of Algorithms 39 (2001) 78-116.
[249] J. Ding, J.H. Kim, E. Lubetzky and Y. Peres, Anatomy of a young giant com-
ponent in the random graph, Random Structures and Algorithms 39 (2011) 139-
178.
[250] J. Ding, E. Lubetzky and Y. Peres, Mixing time of near-critical random graphs,
Annals of Probability 40 3 (2012) 979-1008.
474 References

[251] R. DiPietro, L.V. Mancini, A. Mei, A. Panconesi and J. Radhakrishnan, Re-

doubtable sensor networks, ACM Trans. Inform. Systems Security 11 (2008) 1-
22.
[252] R.P. Dobrow and R.T. Smythe, Poisson approximations for functionals of ran-
dom trees, Random Structures and Algorithms 9 (1996) 79-92.
[253] M. Dondajewski and J. Szymański, On a generalized random recursive tree, un-
published manuscript.
[254] M. Drmota, Random Trees, Springer, Vienna, 2009.
[255] M. Drmota and H-K. Hwang, Profiles of random trees: Correlation and width of
random recursive trees and binary search trees, Adv. in Appl. Probab. 37 (2005)
321-341.
[256] A. Dudek and A.M. Frieze, Loose Hamilton Cycles in Random k-Uniform Hy-
pergraphs, Electronic Journal of Combinatorics (2011) P48.
[257] A. Dudek and A.M. Frieze, Tight Hamilton Cycles in Random Uniform Hyper-
graphs, Random structures and Algorithms 42 (2012) 374-385.
[258] A. Dudek, A.M. Frieze, P. Loh and S. Speiss Optimal divisibility conditions for
loose Hamilton cycles in random hypergraphs, Electronic Journal of Combina-
torics 19 (2012).
[259] A. Dudek, A.M. Frieze, A. Ruciński and M. Šileikis, Loose Hamilton Cycles
in Regular Hypergraphs, Combinatorics, Probability and Computing 24 (2015)
179-194.
[260] A. Dudek, A.M. Frieze, A. Ruciński and M. Šilekis, Embedding the Erdős-
Rényi Hypergraph into the Random Regular Hypergraph and Hamiltonicity, see
arxiv.org.
[261] A. Dudek, A.M. Frieze and C. Tsourakakis, Rainbow connection of random reg-
ular graphs, see arxiv.org.
[262] A. Dudek, D. Mitsche and P. Pralat, The set chromatic number of random graphs,
see arxiv.org.
[263] A. Dudek and P. Pralat, An alternative proof of the linearity of the size-Ramsey
number of paths, to appear in Combinatorics, Probability and Computing.
[264] A. Dudek, P. Pralat, Acquaintance time of random graphs near connectivity
threshold, see arxiv.org.
[265] R. Durrett, Rigoous result for CHKNS random graph model, In: Proc. Discrete
Random Walks (Paris 2003), Discrete Mathematics and Theoretical Computer
Science AC (2003) 95-104.
[266] R. Durrett, Random Graph Dynamics, Cambridge University Press, 2007.
[267] R. Durrett, Probability: Theory and Examples, 4th Edition, Cambridge Univer-
sity Press 2010.
[268] M. Dwass, The total progeny in a branching process, Journal of Applied Proba-
bility 6 (1969) 682-686.
[269] M.E.Dyer, A.M.Frieze and L.R. Foulds, On the strength of connectivity of ran-
dom subgraphs of the n-cube, in Random Graphs ’85, Annals of Discrete Math-
ematics 33 (Karoński, Palka Eds.), North-Holland (1987) 17-40.
[270] M. Dyer, A. Flaxman, A. M. Frieze and E. Vigoda, Random colouring sparse
random graphs with fewer colours than the maximum degree, Random Structures
and Algorithms 29 (2006) 450-465.
 References 475

[271] M.E.Dyer, A.M.Frieze and C. Greenhill, On the chromatic number of a random

hypergraph, see arxiv.org.
[272] M.E.Dyer, A.M.Frieze and C.J.H.McDiarmid, Linear programs with random
costs, Mathematical Programming 35 (1986) 3-16.
[273] M.E.Dyer, A.M.Frieze and B. Pittel, The average performance of the greedy
matching algorithm, The Annals of Applied Probability 3 (1993) 526-552.
[274] E. Ebrahimzadeh, L. Farczadi, P. Gao, A. Mehrabian, C. Sato, N. Wormald and J.
Zung, On the Longest Paths and the Diameter in Random Apollonian Networks.
[275] J. Edmonds, Paths, Trees and Flowers, Canadian Journal of Mathematics 17
(1965) 449-467.
[276] C. Efthymiou, MCMC sampling colourings and independent sets of G(n, d/n)
near uniqueness threshold, Proceedings of SODA 2014 305-316.
[277] G.P. Egorychev, A solution of the Van der Waerden’s permanent problem,
Preprint IFSO-L3 M Academy of Sciences SSSR, Krasnoyarks (1980).
[278] P. Erdős, Graph Theory and Probability, Canadian Journal of Mathematics 11
(1959) 34-38.
[279] P. Erdős and A. Rényi, On random graphs I, Publ. Math. Debrecen 6 (1959)
290-297.
[280] P. Erdős and A. Rényi, On the evolution of random graphs, Publ. Math. Inst.
Hungar. Acad. Sci. 5 (1960) 17-61.
[281] P. Erdős and A. Rényi, On the strength of connectedness of a random graph,
Acta. Math. Acad. Sci. Hungar., 8 (1961) 261-267.
[282] P. Erdős and A. Rényi, On random matrices, Publ. Math. Inst. Hungar. Acad.
Sci. 8 (1964) 455-461.
[283] P. Erdős and A. Rényi, On the existence of a factor of degree one of a connected
random graph, Acta. Math. Acad. Sci. Hungar. 17 (1966) 359-368.
[284] P. Erdős, M. Simonovits and V. T. Sós, Anti-Ramsey Theorems, Colloquia Math-
ematica Societatis János Bolya 10, Infinite and Finite Sets, Keszethely, 1973.
[285] P. Erdős and M. Simonovits, Supersaturated graphs and hypergraphs, Combina-
torica 3 (1983) 181-192.
[286] P. Erdős and J. Spencer, Evolution of the n-cube, Computers & Mathematics with
Applications 5 (1979) 33-39.
[287] P. Erdős, S. Suen and P. Winkler, On the size of a random maximal graph, Ran-
dom Structures and Algorithms 6 (1995) 309-318.
[288] L. Eschenauer and V.D. Gligor, A key managment scheme for distributed sensor
networks, In: Proceedings of the 9th ACM Conference on Computer and Com-
munication Security (2002) 41-47.
[289] H. van den Esker, A geometric preferential attachment model with fitness, see
arxiv.org.
[290] R. Fagin, Probabilities in Finite Models, Journal of Symbolic Logic 41 (1976)
50-58.
[291] D.I. Falikman, The proof of the Van der Waerden’s conjecture regarding to dou-
bly stochastic matrices, Mat. Zametki 29 (1981).
[292] M. Faloutsos, P. Faloutsos and C. Faloutsos, On Power-Law Relationships of the
Internet Topology, ACM SIGCOMM, Boston (1999).
[293] V. Feldman, E. Grigorescu, L. Reyzin, S. Vempala and Y. Xiao, Statistical Algo-
rithms and a lower bound for detecting planted cliques, STOC 2013.
476 References

[294] W. Feller, An Introduction to Probability Theory and its Applications, 3rd Ed.,
Wiley, New York (1968).
[295] Q. Feng, H.M. Mahmoud and A. Panholzer, Phase changes in subtree varieties
in random recursive and binary search trees, SIAM Journal Discrete Math. 22
(2008) 160-184.
[296] A. Ferber, Closing gaps in problems related to Hamilton cycles in random graphs
and hypergraphs, see arxiv.org.
[297] A. Ferber, R. Glebov, M. Krivelevich and A. Naor, Biased games on random
boards, to appear in Random Structures and Algorithms.
[298] A. Ferber, G. Kronenberg and E. Long, Packing, Covering and Counting Hamil-
ton Cycles in Random Directed Graphs, see arxiv.org.
[299] A. Ferber, G. Kronenberg, F. Mousset and C. Shikhelman, Packing a randomly
edge-colored random graph with rainbow k-outs, see arxiv.org.
[300] A. Ferber, R. Nenadov and U. Peter, Universality of random graphs and rainbow
embedding, see arxiv.org.
[301] A. Ferber, R. Nenadov, A. Noever, U. Peter and N. Škorić, Robust hamiltonicity
of random directed graphs, see arxiv.org.
[302] T.I. Fenner and A.M. Frieze, On the connectivity of random m-orientable graphs
and digraphs, Combinatorica 2 (1982) 347-359.
[303] D. Fernholz ande V. Ramachandran, The diameter of sparse random graphs, Ran-
dom Structures and Algorithms 31 (2007) 482-516.
[304] J.A. Fill, E.R. Scheinerman and K.B. Singer-Cohen, Random intersection graphs
when m = ω(n): An equivalence theorem relating the evolution the evolution of
the G(n, m, p) and G(n, p) models, Random Structures and Algorithms 16 (2000)
156-176.
[305] M. Fischer, N. Škorić, A. Steger and M. Trujić, Triangle resilience of the square
of a Hamilton cycle in random graphs, see arxiv.org.
[306] G. Fiz Pontiveros, S. Griffiths and R. Morris, The triangle-free process and
R(3, k), see arxiv.org.
[307] A. Flaxman, The Lower Tail of the Random Minimum Spanning Tree, The Elec-
tronic Journal of Combinatorics 14 (2007) N3.
[308] A. Flaxman, A.M. Frieze and T. Fenner, High degree vertices and eigenvalues in
the preferential attachment graph, Internet Mathematics 2 (2005) 1-20.
[309] A.Flaxman, A.M. Frieze and J. Vera, A Geometric Preferential Attachment
Model of Networks, Internet Mathematics 3 (2007) 187-205.
[310] A.Flaxman, A.M. Frieze and J.Vera, A Geometric Preferential Attachment
Model of Networks, Internet Mathematics 4 (2007) 87-112.
[311] A. Flaxman, D. Gamarnik and G.B. Sorkin, Embracing the giant component,
Random Structures and Algorithms 27 (2005) 277-289.
[312] J.E. Folkert, The distribution of the number of components of a random mapping
function, PhD Dissertation, Michigan State University, (1955).
[313] C.M. Fortuin, P.W. Kasteleyn and J. Ginibre, Correlation inequalities on some
partially ordered sets, Communications in Mathematical Physics 22 (1971) 89-
103.
[314] N. Fountoulakis, On the evolution of random graphs on spaces of negative cur-
vature, see arxiv.org.
 References 477

[315] N. Fountalakis and K. Panagiotou, Sharp load thresholds for cuckoo hashing,
Random Structures and Algorithms 41 (2012) 306-333.
[316] N. Fountalakis, M. Khosla and K. Panagiotou, The Multiple-orientability
Thresholds for Random Hypergraphs, see arxiv.org.
[317] N. Fountoulakis, D. Kühn and D. Osthus, The order of the largest complete mi-
nor in a random graph, Random Structures and Algorithms 33 (2008) 127-141.
[318] N. Fountoulakis and B. A. Reed, The evolution of the mixing rate of a simple
random walk on the giant component of a random graph, Random Structures
Algorithms 33 (2008) 68-86.
[319] K. Frankston, J. Kahn, B. Narayanan and J. Park, Thresholds versus fractional
expectation thresholds, see arxiv.org.
[320] E. Friedgut, Hunting for Sharp Thresholds, Random Structures Algorithms 26
(2005) 37-51.
[321] E. Friedgut and G. Kalai, Every Monotone Graph Property Has a Sharp Thresh-
old, Proceedings of the American Mathematical Society 124 (1996) 2993-3002.
[322] E. Friedgut, Sharp Thresholds of Graph Proprties, and the k-sat Problem, Journal
of the American Mathematical Society 12 (1999) 1017-1054.
[323] J. Friedman, A proof of Alon’s second eigenvalue conjecture and related prob-
lems, Memoirs of the American Mathematical Society, 2008.
[324] T. Friedrich and A. Krohmer, Cliques in hyperbolic random graphs, IEEE IN-
FOCOM (2015).
[325] T. Friedrich, T. Sauerwald and A. Stauffer, Diameter and Broadcast Time of
Random Geometric Graphs in Arbitrary Dimensions, Algorithmica 67 (2013)
65-88.
[326] A.M. Frieze, On the value of a random minimum spanning tree problem, Dis-
crete Applied Mathematics 10 (1985) 47-56.
[327] A.M. Frieze, On large matchings and cycles in sparse random graphs, Discrete
Mathematics 59 (1986) 243-256.
[328] A.M. Frieze, Maximum matchings in a class of random graphs, Journal of Com-
binatorial Theory B 40 (1986) 196-212.
[329] A.M. Frieze, An algorithm for finding hamilton cycles in random digraphs, Jour-
nal of Algorithms 9 (1988) 181-204.
[330] A.M. Frieze, On random regular graphs with non-constant degree,
http://www.math.cmu.edu/ af1p/Texfiles/REGNONCONSTANT.pdf.
[331] A.M. Frieze, On the independence number of random graphs, Discrete Mathe-
matics 81 (1990) 171-176.
[332] A.M. Frieze, Perfect matchings in random bipartite graphs with minimal degree
at least 2, Random Structures and Algorithms 26 (2005) 319-358.
[333] A.M. Frieze, Loose Hamilton Cycles in Random 3-Uniform Hypergraphs, Elec-
tronic Journal of Combinatorics 17 (2010) N28.
[334] A.M. Frieze, On a Greedy 2-Matching Algorithm and Hamilton Cycles in Ran-
dom Graphs with Minimum Degree at Least Three, Random Structures and Al-
gorithms 45 (2014) 443-497.
[335] A.M. Frieze, N. Goyal, L. Rademacher and S. Vempala, Expanders via Random
Spanning Trees, SIAM Journal on Computing 43 (2014) 497-513.
478 References

[336] A.M. Frieze and S. Haber, An almost linear time algorithm for finding Hamilton
cycles in sparse random graphs with minimum degree at least three, to appear in
Random Structures and Algorithms.
[337] A.M. Frieze, S. Haber and M. M. Lavrov, On the game chromatic number of
sparse random graphs, SIAM Journal of Discrete Mathematics 27 (2013) 768-
790.
[338] A.M. Frieze, M.R.Jerrum, M.Molloy, R.Robinson and N.C.Wormald, Generat-
ing and counting Hamilton cycles in random regular graphs, Journal of Algo-
rithms 21 (1996) 176-198.
[339] A.M. Frieze and T. Johansson, On random k-out sub-graphs of large graphs, see
arxiv.org.
[340] A.M. Frieze and R. Kannan, A new approach to the planted clique problem,
Proceedings of Foundations of Software Technology and Theoretical Computer
Science, Bangalore, India (2008).
[341] A.M. Frieze and M. Krivelevich, On two Hamilton cycle problems in random
graphs, Israel Journal of Mathematics 166 (2008) 221-234.
[342] A.M. Frieze, M. Krivelevich and C. Smyth, On the chromatic number of random
graphs with a fixed degree sequence, Combinatorics, Probability and Computing
16 (2007) 733-746.
[343] A.M. Frieze, S. Haber and M. Lavrov, On the game chromatic number of sparse
random graphs, SIAM Journal of Discrete Mathematics 27 (2013) 768-790.
[344] A.M. Frieze and P. Loh, Rainbow Hamilton cycles in random graphs, Random
Structures and Algorithms 44 (2014) 328-354.
[345] A.M. Frieze and T. Łuczak, Hamiltonian cycles in a class of random graphs: one
step further, in em Proceedings of Random Graphs ’87, Edited by M.Karonski,
J.Jaworski and A.Rucinski, John Wiley and Sons (1990) 53-59.
[346] A.M. Frieze and T. Łuczak, Edge disjoint trees in random graphs, Periodica
Mathematica Hungarica 21 (1990) 28-30.
[347] A.M. Frieze and T. Łuczak, On the independence and chromatic numbers of
random regular graphs, Journal of Combinatorial Theory 54 (1992) 123-132.
[348] A.M. Frieze and C.J.H. McDiarmid, On random minimum length spanning trees,
Combinatorica 9 (1989) 363 - 374.
[349] A.M. Frieze and B.D. McKay, Multicoloured trees in random graphs, Random
Structures and Algorithms 5 (1994) 45-56.
[350] A.M. Frieze and P. Melsted, Maximum Matchings in Random Bipartite Graphs
and the Space Utilization of Cuckoo Hashtables, Random Structures and Algo-
rithms 41 (2012) 334-364.
[351] A.M. Frieze, D. Mitsche, X. Pérez-Giménez and P. Pralat, On-line list colouring
of random graphs, see arxiv.org.
[352] A.M. Frieze and W. Pegden, Between 2- and 3-colorability, see arxiv.org.
[353] A.M. Frieze and W. Pegden, Looking for vertex number one, see arxiv.org.
[354] A.M. Frieze and B. Pittel, Perfect matchings in random graphs with prescribed
minimal degree, Trends in Mathematics, Birkhauser Verlag, Basel (2004) 95-
132.
[355] A.M. Frieze, M.Ruszinko and L.Thoma, A note on random minimum length
spanning trees, Electronic Journal of Combinatorics 7 (2000) R41.
 References 479

[356] A.M. Frieze and G. Sorkin, Efficient algorithms for three-dimensional axial
and planar random assignment problems, Random Structures and Algorithms
46 (2015) 160-196.
[357] A.M. Frieze and C. Tsourakakis, High Degree Vertices, Eigenvalues and Diam-
eter of Random Apollonian Networks, WAW2012 (2012).
[358] A.M. Frieze and C. E. Tsourakakis, Rainbow connectivity of sparse random
graphs, Electronic Journal of Combinatorics 19 (2012).
[359] A.M. Frieze and L. Zhao, Optimal Construction of Edge-Disjoint Paths in Ran-
dom Regular Graphs, Combinatorics, Probability and Computing 9 (2000) 241-
264.
[360] M. Fuchs, The subtree size profile of plane-oriented recursive trees, manuscript
[361] M. Fuchs, H-K. Hwang, R. Neininger, Profiles of random trees: Limit theorems
for random recursive trees and binary search trees,Algorithmica 46 (2006) 367-
407.
[362] Z. Füredi and J. Komlós, The eigenvalues of random symmetric matrics, Com-
binatorica 1 (1981) 233-241.
[363] D. Gale and L. S. Shapley, College Admissions and the Stability of Marriage,
American Mathematical Monthly 69 (1962) 9-14.
[364] P. Gao and M. Molloy, The stripping process can be slow, see arxiv.org.
[365] P. Gao, X. Pérez-Giménez and C. M. Sato, Arboricity and spanning-tree packing
in random graphs with an application to load balancing, see arxiv.org.
[366] P. Gao and N. Wormald, Orientability thresholds for random hypergraphs, see
arxiv.org.
[367] J. Gastwirth, A probability model of a pyramid scheme, American Statistician
31 (1977) 79-82.
[368] H. Gebauer and T. Szabó, Asymptotic random graph intuition for the biased
connectivity game, Random Structures and Algorithms 35 (2009) 431-443.
[369] S. Gerke, H-J. Prömel, T. Schickinger, A. Steger and A. Taraz, K4 -free subgraphs
of random graphs revisited, Combinatorica 27 (2007) 329-365.
[370] S. Gerke, H-J. Prömel, T. Schickinger, K5 -free subgfraphs of random graphs,
Random Structures and Algorithms 24 (2004) 194-232.
[371] S. Gerke, D. Schlatter, A. Steger and A. Taraz, The random planar graphs pro-
cess, Random Structures and Algorithms 32 (2008) 236-261.
[372] I.B. Gertsbakh, Epidemic processes on random graphs: Some preliminary re-
sults, J. Appl. Prob. 14 (1977) 427-438.
[373] I. Gessel and R.P. Stanley, Stirling polynomials, Journal of Combinatorial The-
ory A 24 (1978) 24-33.
[374] E.N. Gilbert, Random graphs, Annals of Mathematical Statistics 30 (1959) 1141-
1144.
[375] O. Giménez and M. Noy, Asymptotic enumeration and limit laws of planar
graphs, Journal of the American Mathematical Society 22 (2009) 309-329.
[376] R. Glebov and M. Krivelevich, On the number of Hamilton cycles in sparse
random graphs, SIAM Journal on Discrete Mathematics 27 (2013) 27-42.
[377] R. Glebov, H. Naves and B. Sudakov, The threshold probability for cycles, see
arxiv.org.
480 References

[378] Y.V. Glebskii, D.I. Kogan, M.I. Liagonkii and V.A. Talanov, Range and degree
of realizability of formulas in the restricted predicate calculus, Cybernetics 5
(1969) 142-154.
[379] E. Godehardt and J.Jaworski, On the connectivity of a random interval graph,
Random Structures and Algorithms 9 (1996) 137-161.
[380] E. Godehardt and J. Jaworski, Two models of random intersection graphs for
classification, in: Data Analysis and Knowledge Organization, Springer, Berlin,
22 (2003) 67-82.
[381] E. Godehardt, J. Jaworski and K. Rybarczyk, On the connectivity of random in-
tersection graphs, In: Studies in Classification, Data Analysis and Knowledge
Organization 33, Lenz H. J., Decker R., (Eds), Springer Verlag,Heidelberg -
Berlin (2007) 67-74.
[382] A. Goel, S. Rai and B. Krishnamachari, Monotone properties of Random geo-
metric graphs have sharp thresholds, Annals of Applied Probability 15 (2005)
2535-2552.
[383] W. Goh, E. Schmutz, Limit distribution for the maximum degree of a random
recursive tree, Journal of Computing and Applied Mathematics 142 (2002) 61-
82.
[384] Ch. Goldschmidt and J.B. Martin, Random recursive trees and the Bolthausen-
Sznitman coalescent, Electronic Journal of Probability 10 (2005) 718-745.
[385] S.W. Golomb, Random permutations, Bulletin of the American Mathematical
Society 70 (1964) 747.
[386] V.L. Goncharov, Sur la distribution des cycles dans les permutations C.R. Acad.
Sci USSR (N.S.) 35, (1942) 267-269.
[387] V.L. Goncharov, Some facts in combinatorics, Izv. Acad. Nauk SSSR, Ser. Mat. 8
(1944) 3-48 (in Russian). [See also: On the field of combinatory analysis, Trans.
Amer. Math. Soc. 19 (1962) 1-46.]
[388] D. A. Grable, On random greedy triangle packing, Electronic Journal on Com-
binatorics 4 (1997) R11.
[389] R. Gray, Entropy and Information Theory, Springer, 2011.
[390] C. Greenhill, S. Janson and A. Ruciński, On the number of perfect matchings in
random lifts, Combinatorics, Probability and Computing 19 (2010) 791-817.
[391] G.R. Grimmett, Random labelled trees and their branching networks, Journal of
the Australian Mathematical Society 30 (1980) 229-237.
[392] G. Grimmett and C. McDiarmid, On colouring random graphs, Mathematical
Proceedings of the Cambridge Philosophical Society 77 (1975) 313-324.
[393] C. Groër, B.D. Sullivan and S. Poole, A mathematical analysis of the R-MAT
random graph generator, Networks,58 (2011) 159-170.
[394] L. Gugelmann, K. Panagiotou and U. Peter, Random hyperbolic graphs: Degree
sequence and clustering, Proceedings of the 39th International Colloquium Con-
ference on Automata, Languages, and Programming ( ICALP) (2012) 573-585.
[395] P. Gupta and P.R. Kumar, Critical Power for Asymptotic Connectivity in Wire-
less Networks, in Stochastic Analysis, Control, Optimization and Applications:
A Volume in Honor of W. H. Fleming, W. M. McEneany, G. Yin, and Q. Zhang,
Eds. Boston, MA: Birkhauser, (1998) 547-566.
[396] Y. Gurevich and S. Shelah, Expected computation time for Hamiltonian path
problem, SIAM Journal on Computing 16 (1987) 486-502.
 References 481

[397] E. Győri, B. Rorthchild and A. Ruciński, Every graph is contained in a sparest

possible balanced graph, Mathematical Proceedings of the Cambridge Philo-
sophical Society 98 (1985) 397-401.
[398] S. Haber and M. Krivelevich, The logic of random regular graphs, Journal of
Combinatorics 1 (2010) 389-440.
[399] On resilience of connectivity in the evolution of random graphs, see arxiv.org.
[400] A. Hamm and J. Kahn, On Erdős-Ko-Rado for random hypergraphs I, see
arxiv.org.
[401] A. Hamm and J. Kahn, On Erdős-Ko-Rado for random hypergraphs II, see
arxiv.org.
[402] J.C. Hansen and J. Jaworski, Random mappings with exchangeable in-degrees,
Random Structures and Algorithms 33 (2008) 105-126.
[403] J.C. Hansen and J. Jaworski, Predecessors and successors in random mappings
with exchangeable in-degrees, J. Applied Probability 50 (2013) 721-740.
[404] T. Harris, A lower bound for the critical probability in a certain percolation,
Proceedings of the Cambridge Philosophical Society 56 (1960) 13-20.
[405] H. Hatami, Random cubic graphs are not homomorphic to the cycle of size 7,
Journal of Combinatorial Theory B 93 (2005) 319-325.
[406] H. Hatami and M. Molloy, The scaling window for a random graph with a given
degree sequence, Random Structures and Algorithms 41 (2012) 99-123.
[407] P. Haxell, Y. Kohayakawa and T. Łuczak, Turáns extremal problem in random
graphs: forbidding even cycles, Journal of Combinatorial Theory B 64 (1995)
273-287.
[408] P. Haxell, Y. Kohayakawa and T. Łuczak, Turáns extremal problem in random
graphs: forbidding odd cycles, Combinatorica 16 (1995) 107-122.
[409] J. He and H. Liang, On rainbow-k-connectivity of random graphs, Information
Processing Letters 112 (2012) 406-410.
[410] A. Heckel and O. Riordan, The hitting time of rainbow connection number two,
Electronic Journal on Combinatorics 19 (2012).
[411] D. Hefetz, M. Krivelevich, M. Stojaković and T. Szabó, Positional Games (Ober-
wolfach Seminars), Birkhauser 2014.
[412] D. Hefetz, M. Krivelevich and T. Szabo, Sharp threshold for the appearance of
certain spanning trees in random graphs, Random Structures and Algorithms 41
(2012) 391-412.
[413] D. Hefetz, A. Steger and B. Sudakov, Random directed graphs are robustly
Hamiltonian, Random Structures and Algorithms 49 (2016) 345-362.
[414] D. Hefetz, A. Steger and B. Sudakov, Random directed graphs are robustly
hamiltonian, see arxiv.org.
[415] W. Hoeffding, Probability inequalities for sums of bounded random variables,
Journal of the American Statistical Association 58 (1963) 13-30.
[416] R. van der Hofstad, Random graphs and complex networks: Volume 1, see
arxiv.org.
[417] R. van der Hofstad, Critical behaviour in inhomogeneous random graphs, Ran-
dom Structures and Algorithms 42 (2013) 480-508.
[418] R. van der Hofstad, G. Hooghiemstra and P. Van Mieghem, On the covariance of
the level sizes in random recursive trees, Random Structures and Algorithms 20
(2002) 519-539.
482 References

[419] R. van der Hofstad, S. Kliem and J. van Leeuwaarden, Cluster tails for critical
power-law inhomogeneous random graphs, see arxiv.org
[420] C. Holmgren and S. Janson, Limit laws for functions of fringe trees for binary
search trees and recursive trees, see arxiv.org
[421] P. Horn and M. Radcliffe, Giant components in Kronecker graphs, Random
Structures and Algorithms 40 (2012) 385-397.
[422] H-K. Hwang, Profiles of random trees: plane-oriented recursive trees, Random
Structures and Algorithms 30 (2007) 380-413.
[423] G.I. Ivchenko, On the asymptotic behaviour of the degrees of vertices in a ran-
dom graph, Theory of Probability and Applications 18 (1973) 188-195.
[424] S. Janson, Poisson convergence and Poisson processes with applications to ran-
dom graphs, Stochastic Process. Appl. 26 (1987) 1-30.
[425] S. Janson, Normal convergence by higher semi-invariants with applications to
random graphs, Ann. Probab. 16 (1988) 305-312.
[426] S. Janson, A functional limit theorem for random graphs with applications to
subgraph count statistics Random Structures and Algorithms 1 (1990) 15-37.
[427] S. Janson, Poisson approximation for large deviations, Random Structures and
Algorithms 1 (1990) 221-230.
[428] S. Janson, Multicyclic components in a random graph process, Random Struc-
tures and Algorithms 4 (1993) 71-84.
[429] S. Janson, The numbers of spanning trees, Hamilton cycles and perfect match-
ings in a random graph, Combin. Probab. Comput. 3 (1994) 97-126.
[430] S. Janson, The minimal spanning tree in a complete graph and a functional limit
theorem for trees in a random graph, Random Structures Algorithms 7 (1995)
337-355.
[431] S. Janson, Random regular graphs: Asymptotic distributions and contiguity,
Combinatorics, Probability and Computing 4 (1995) 369-405.
[432] S. Janson, One, two and three times log n/n for paths in a complete graph with
random weights, Combinatorics, Probability and Computing 8 (1999) 347-361.
[433] S. Janson, Asymptotic degree distribution in random recursive trees, Random
Structures and Algorithms, 26 (2005) 69-83.
[434] S. Janson, Monotonicity, asymptotic normality and vertex degrees in random
graphs, Bernoulli 13 (2007) 952-965.
[435] S. Janson, Plane recursive trees, Stirling permutations and an urn model,Fifth
Colloquium on Mathematics and Computer Science, Discrete Math. Theor. Com-
put. Sci. Proc., AI, Assoc. Discrete Math. Theor. Comput. Sci., Nancy, (2008)
541-547.
[436] S. Janson, Asymptotic equivalence and contiguity of some random graphs, Ran-
dom StructuresAlgorithms 36 (2010) 26-45.
[437] S. Janson, M. Kuba and A. Panholzer, Generalized Stirling permutations, fami-
lies of increasing trees and urn models, Journal of Combinatorial Theory A 118
(2010) 94-114.
[438] S. Janson and M.J. Luczak, A simple solution to the k-core problem, Random
Structures Algorithms 30 (2007) 50-62.
[439] S. Janson, K. Oleszkiewicz and A. Ruciński, Upper tails for subgraph counts in
random graphs, Israel Journal of Mathematics 142 (2004) 61-92.
 References 483

[440] S. Janson, T. Łuczak, A. Ruciński, An exponential bound for the probability of

nonexistence of a specified subgraph in a random graph, M. Karoński (ed.) J.
Jaworski (ed.) A. Ruciski (ed.) , Random Graphs ’87 , Wiley (1990) 73-87.
[441] S. Janson, T. Łuczak and A. Ruciński, Random Graphs, John Wiley and Sons,
New York, 2000.
[442] S. Janson and K. Nowicki, The asymptotic distributions ofgeneralized U-
statistics with applications to random graphs, Probability Theory and Related
Fields 90 (1991) 341-375.
[443] S. Janson and O. Riordan, Susceptibility in inhomogeneous random graphs, The
Electr. J. Comb. 19 (2012) #P31
[444] S. Janson and N. Wormald, Rainbow Hamilton cycles in random regular graphs,
Random Structures Algorithms 30 (2007) 35-49.
[445] S. Janson, D.E. Knuth, T. Łuczak and B.G. Pittel, The birth of the giant compo-
nent, Random Structures and Algorithms 4 (1993) 233-358.
[446] J. Janssen, P. Pralat, and R. Wilson, Geometric Graph Properties of the Spatial
Preferred Attachment model, Advances in Applied Mathematics 50 (2013) 243-
267.
[447] J. Janssen, P. Pralat, and R. Wilson, Asymmetric Distribution of Nodes in the
Spatial Preferred Attachment Model, Proceedings of the 10th Workshop on Algo-
rithms and Models for the Web Graph (WAW 2013), Lecture Notes in Computer
Science 8305, Springer (2013) 1-13.
[448] J. Jaworski, On a random mapping (T, Pj ), J. Applied Probability 21 (1984) 186-
191.
[449] J. Jaworski, Random mappings with independent choices of images, In: Random
Graphs’87, Wiley, New York (1990) 89-101.
[450] J. Jaworski, Predecessors in a random mapping, Random Structures and Algo-
rithms 13 (1998) 501-519.
[451] J. Jaworski, Epidemic processes on digraphs of random mappings, J. Applied
Probabability 36 (1999) 780-798.
[452] J. Jaworski and T. Łuczak, Cycles in a uniform graph process, Combinatorics,
Probability and Computing 1 (1992) 223-239.
[453] J. Jaworski and Z. Palka, Remarks on a general model of a random digraph,Ars
Combinatoria 65 (2002) 135-144.
[454] J. Jaworski and I.H. Smit, On a random digraph, Annals of Discrete Math. 33
(1987) 111-127.
[455] J. Jaworski, M. Karoński and D. Stark, The degree of a typical vertex in gener-
alized random intersection graph model, Discrete Mathematics 30 (2006) 2152-
2165.
[456] J. Jaworski and D. Stark, The vertex degree distribution of passive random inter-
section graph models, Combin. Prob Comput. 17 (2008) 549-558.
[457] M.R. Jerrum, Large Cliques Elude the Metropolis Process, Random Structures
and Algorithms 3 (1992) 347-359.
[458] J. Jiang, M. Mitzenmacher and J. Thaler, Parallel Peeling Algorithms, see
arxiv.org.
[459] J. Jonasson, On the cover time of random walks on random graphs, Combina-
torics, Probability and Computing 7 (1998), 265-279.
484 References

[460] A. Johansson, J. Kahn and V. Vu, Factors in Random Graphs, Random Structures
and Algorithms 33 (2008) 1-28.
[461] J. Jordan, Geometric preferential attachment in non-uniform metric spaces, Elec-
tronic Journal of Probability 18 (2013) 1-15.
[462] J. Jordan and A. Wade, Phase transitions for random geometric preferential at-
tachment graphs, Advances in Applied Probability 47 (2015).
[463] F. Juhász, On the spectrum of a random graph, in Algebraic methods in graph
theory, (Lovász et al., eds) Coll. Math. Soc. J. Bolyai, North Holland 25 (1981)
313-316.
[464] J. Kahn, E. Lubetzky and N. Wormald, The threshold for combs in random
graphs, see arxiv.org.
[465] J. Kahn, E. Lubetzky and N. Wormald, Cycle factors and renewal theory, see
arxiv.org.
[466] The threshold for the square of a Hamilton cycle, see arxiv.org.
[467] J. Kahn and E. Szemerédi, On the second eigenvalue in random regular graphs -
Section 2, Proceedings of the 21’st Annual ACM Symposium on Theory of Com-
puting (1989) 587-598.
[468] M. Kahle, Topology of random simplicial complexes: a survey, AMS Contem-
porary Volumes in Mathematics, To appear in AMS Contemporary Volumes in
Mathematics. Nov 2014. arXiv:1301.7165.
[469] S. Kalikow and B. Weiss, When are random graphs connected? Israel J. Math.
62 (1988) 257-268.
[470] N. Kamčev, M. Krivelevich and B. Sudakov, Some Remarks on Rainbow Con-
nectivity, see arxiv.org.
[471] M. Kang, M. Karoński, C. Koch and T. Makai, Properties of stochastic Kro-
necker graphs, Journal of Combinatorics, to appear.
[472] M. Kang and T. Łuczak, Two critical periods in the evolution of random planar
graphs, Transactions of the American Mathematical Society 364 (2012) 4239-
4265.
[473] M. Kang, W. Perkins and J. Spencer, The Bohman-Frieze Process Near Critical-
ity, see arxiv.org.
[474] D. Karger and C. Stein, A new approach to the minimum cut problem, Journal
of the ACM 43 (1996) 601-640.
[475] M. Karoński, A review of random graphs, Journal of Graph Theory 6 (1982)
349-389.
[476] M. Karoński, Random graphs, In: Handbook of Combinatorics, T. Graham, M.
Grötschel and L. Lovás, Eds., Elsevier science B.V. (1995) 351-380.
[477] M. Karoński and T. Łuczak, The phase transition in a random hypergraph, Jour-
nal of Computational and Applied Mathematics 1 (2002) 125-135.
[478] M. Karoński and B. Pittel, Existence of a perfect matching in a random (1 +
ε −1 )–out bipartite graph, Journal of Combinatorial Theory B (2003) 1-16.
[479] M. Karoński and A. Ruciński, On the number of strictly balanced subgraphs of
a random graph, In: Graph Theory, Proc. Łagów, 1981, Lecture Notes in Mathe-
matics 1018, Springer, Berlin (1983) 79-83.
[480] M. Karoński and A. Ruciński, Problem 4, In Graphs and other Combinatorial
Topics, Proceedings, Third Czech. Symposium on Graph Theory, Prague (1983).
 References 485

[481] M. Karoński and A. Ruciński, Poisson convergence and semi-induced properties

of random graphs, Math. Proc. Camb. Phil. Soc. 101 (1987) 291-300.
[482] M. Karoński, E. Scheinerman and K. Singer-Cohen, On Random Intersection
Graphs: The Subgraph Problem, Combinatorics, Probability and Computing 8
(1999) 131-159.
[483] R.M. Karp, An upper bound on the expected cost of an optimal assignment, Dis-
crete Algorithms and Complexity: Proceedings of the Japan-US Joint Seminar
(D. Johnson et al., eds.), Academic Press, New York, 1987, 1-4.
[484] R.M. Karp, The transitive closure of a random digraph, Random Structures and
Algorithms 1 (1990) 73-93.
[485] R.M. Karp, A. Rinnooy-Kan and R.V. Vohra, Average Case Analysis of a Heuris-
tic for the Assignment Problem, Mathematics of Operations Research 19 (1994)
513-522.
[486] R.M. Karp and M. Sipser, Maximum matchings in sparse random graphs, Pro-
ceedings of the 22nd IEEE Symposium on the Foundations of Computer Science
(1981) 364-375.
[487] Zs. Katona, Width of scale-free tree, J. of Appl. Prob. 42 (2005) 839-850.
[488] Zs. Katona, Levels of scale-free tree, manuscript
[489] L. Katz, Probability of indecomposability of a random mapping function, Ann.
Math. Statist. 26 (1955) 512-517.
[490] G. Kemkes, X. Perez-Gimenez and N. Wormald, On the chromatic number of
random d-regular graphs, Advances in Mathematics 223 (2010) 300-328.
[491] R. Keusch and A. Steger, The game chromatic number of dense random graphs,
Electronic Journal of Combinatorics 21 (2014).
[492] J-H. Kim, The Ramsey number R(3,t) has order of magnitude t 2 / logt, Random
Structures and Algorithms (1995) 173-207.
[493] J-H. Kim and V. Vu, Sandwiching random graphs: universality between random
graph models, Advances in Mathematics 188 (2004) 444-469.
[494] J.H. Kim and N.C. Wormald, Random matchings which induce Hamilton cycles,
and hamiltonian decompositions of random regular graphs, Journal of Combina-
torial Theory B 81 (2001) 20-44.
[495] J.F.C. Kingman, The population structure associated with the Ewens sampling
formula, Theoretical Population Biology 11 (1977) 274-283.
[496] W. Kinnersley, D. Mitsche, P. Pralat, A note on the acquaintance time of random
graphs, Electronic Journal of Combinatorics 20(3) (2013) #P52.
[497] M. Kiwi and D. Mitsche, A bound for the diameter of random hyperbolic graphs,
see arxiv.org.
[498] J. Kleinberg, The small-world phenomenon: An algorithmic perspective, Pro-
ceedings of the 32nd ACM Symposium on Theory of Computing (2000) 163-170.
[499] D. E. Knuth, R. Motwani and B. Pittel, Stable husbands, Random Structures and
Algorithms 1 (1990) 1-14.
[500] F. Knox, D. Kühn and D. Osthus, Edge-disjoint Hamilton cycles in random
graphs, to appear in Random Structures and Algorithms.
[501] L.M. Koganov, Universal bijection between Gessel-Stanley permutations and di-
agrams of connections of corresponding ranks, Mathematical Surveys 51 (1996)
333-335.
486 References

[502] Y. Kohayakawa, B. Kreuter and A. Steger, An extremal problem for random

graphs and the number of graphs with large even girth, Combinatorica 18 (1998)
101-120.
[503] Y. Kohayakawa, T. Łuczak and V. Rödl, On K4 -free subgraphs of random graphs,
Combinatorica 17 (1997) 173-213.
[504] J. Komlós and E. Szemerédi, Limit distributions for the existence of Hamilton
circuits in a random graph, Discrete Mathematics 43 (1983) 55-63.
[505] V.F. Kolchin, A problem of the allocation of particles in cells and cycles of ran-
dom permutations, Theory Probab. Appl. 16 (1971) 74-90.
[506] V.F. Kolchin, Branching processes, random trees, and a generalized scheme of
arrangements of particles, Mathematical Notes 21 (1977) 386-394.
[507] V.F. Kolchin, Random Mappings, Optimization Software, Inc., New York
(1986).
[508] V.F. Kolchin, Random Graphs, Cambridge University Press, Cambridge (1999).
[509] W. Kordecki, On the connectedness of random hypergraphs, Commentaciones
Mathematicae 25 (1985) 265-283.
[510] W. Kordecki, Normal approximation and isolated vertices in random graphs, In:
Random Graphs’87, Eds. M. Karoński, J. Jaworski and A. Ruciński, John Wiley
and Sons, Chichester (1990) 131-139.
[511] I.N. Kovalenko, Theory of random graphs, Cybernetics and Systems Analysis 7
(1971) 575-579.
[512] D. Krioukov, F. Papadopoulos, M. Kitsak, A. Vahdat, and M. Boguñá, Hyper-
bolic geometry of complex networks, Physics Review E (2010) 82:036106.
[513] M. Krivelevich, The choice number of dense random graphs, Combinatorics,
Probability and Computing 9 (2000) 19-26.
[514] M. Krivelevich, C. Lee and B. Sudakov, Resilient pancyclicity of random and
pseudo-random graphs, SIAM Journal on Discrete Mathematics 24 (2010) 1-16.
[515] M. Krivelevich, C. Lee and B. Sudakov, Long paths and cycles in random sub-
graphs of graphs with large minimum degree, see arxiv.org.
[516] M. Krivelevich, P.-S. Loh and B. Sudakov, Avoiding small subgraphs in Achliop-
tas processes, Random Structures and Algorithms 34 (2009) 165-195.
[517] M. Krivelevich, E. Lubetzky and B. Sudakov, Hamiltonicity thresholds in
Achlioptas processes, Random Structures and Algorithms 37 (2010) 1-24.
[518] M. Krivelevich, A. Lubetzky and B. Sudakov, Longest cycles in sparse random
digraphs, Random Structures and Algorithms 43 (2013) 1-15.
[519] M. Krivelevich, E. Lubetzky and B. Sudakov, Cores of random graphs are born
Hamiltonian, Proceedings of the London Mathematical Society 109 (2014) 161-
188.
[520] M. Krivelevich and W. Samotij, Optimal packings of Hamilton cycles in sparse
random graphs, SIAM Journal on Discrete Mathematics 26 (2012) 964-982.
[521] M. Krivelevich and B. Sudakov, The chromatic numbers of random hypergraphs,
Random Structures and Algorithms 12 (1998) 381-403.
[522] M. Krivelevich and B. Sudakov, Minors in expanding graphs, Geometric and
Functional Analysis 19 (2009) 294-331.
[523] M. Krivelevich and B. Sudakov, The phase transition in random graphs - a simple
proof, Random Structures and Algorithms 43 (2013) 131-138.
 References 487

[524] M. Krivelevich, B. Sudakov, V. Vu and N. Wormald, Random regular graphs of

high degree, Random Structures and Algorithms 18 (2001) 346-363.
[525] M. Krivelevich, B. Sudakov, V.H. Vu and N.C. Wormald, On the probability of
independent sets in random graphs, Random Structures & Algorithms 22 (2003)
1-14.
[526] M. Krivelevich, V.H. Vu, Choosability in random hypergraphs, Journal of Com-
binatorial Theory B 83 (2001) 241-257.
[527] M. Kuba and A. Panholzer, On the degree distribution of the nodes in increasing
trees, J. Combinatorial Theory (A) 114 (2007) 597-618.
[528] L. Kucera, Expected complexity of graph partitioning problems, Discrete Ap-
plied Mathematics 57 (1995) 193-212.
[529] D. Kühn and D. Osthus, On Pósa’s conjecture for random graphs, SIAM Journal
on Discrete Mathematics 26 (2012) 1440-1457.
[530] V. Kurauskas, On small subgraphs in random intersection digraphs, Discrete
Mathematics 313 (2013) 872-885.
[531] V. Kurauskas and M. Bloznelis, Large cliques in sparse random intersection
graphs, see arxiv.org.
[532] A.N. Lageras and M. Lindholm, A note on the component structure in random
intersection graphs with tunable clustering, The Electronic Journal of Combina-
torics 15 (2008) #N10.
[533] C. Lee and B. Sudakov, Dirac’s theorem for random graphs, Random Structures
and Algorithms 41 (2012) 293-305.
[534] M. Lelarge, A new approach to the orientation of random hypergraphs, see
arxiv.org.
[535] C. Lennon and B. Pittel, On a likely number of solutions for a stable marriage
problem, Combinatorics, Probability and Computing (2009) 371-421.
[536] J. Leskovec, Dynamics of Large Networks, PhD in Computer Science, Carnegie
Mellon University, 2008.
[537] J. Leskovec, D. Chakrabarti, J. Kleinberg and C. Faloutsos, Realistic, mathemat-
ically tractable graph generation and evolution, using Kronecker multiplication,
in:PKDD ’05: Proceedings of the 9th European Conference on Principles and
Practice of Knowledge Discovery in Databases (2005) 133-145.
[538] J. Leskovec, J. Kleinberg, and C. Faloutsos, Graphs over time:densification laws,
shrinking diameters and possible explanations, in Proc. of ACM SIGKDD Conf.
on Knowledge Discovery in Data Mining (2005) 177187.
[539] J. Leskovec, D. Chakrabarti, J. Kleinberg, C. Faloutsos and Z. Ghahramani, Kro-
necker Graphs: An Approach to Modeling Networks, Journal of Machine Learn-
ing Research 11 (2010) 985-1042.
[540] K. Lin and G. Reinert, Joint vertex degrees in the inhomogeneous random graph
model G (n, {pi j }), Adv. in Appl. Prob. 44 (2012) 139-165.
[541] N. Linial and R. Meshulam, Homological connectivity of random 2-complexes,
Combinatorica 26 (2006) 475-487.
[542] N. Linial, and E. Rozenman, Random Lifts of Graphs: Perfect Matchings, Com-
binatorica 25 (2005) 407-424.
[543] S. Linusson and J. Wästlund, A proof of Parisi’s conjecture on the random as-
signment problem, Probability Theory and Related Fields 128 (2004) 419-440.
488 References

[544] J. Lu, and Q. Feng, (1998). Strong consistency of the number of vertices of given
degrees in nonuniform random recursive trees, Yokohama Math. J 45 (1998) 61-
69.
[545] E. Lubetzky and A. Sly, Cutoff phenomena for random walks on random regular
graphs, Duke Mathematical Journal 153 (2010) 475-510.
[546] E. Lubetzky and Y. Zhao, On the variational problem for upper tails in sparse
random graphs, see arxiv.org.
[547] T. Łuczak, On the equivalence of two basic models of random graphs, Proceed-
ings of Random Graphs’87, (Michal Karonski, Jerzy Jaworski, Andrzej Rucinski,
eds.), Wiley, Chichester (1990) 151-158.
[548] T. Łuczak, On the number of sparse connected graphs, Random Structures and
Algorithms 1 (1990) 171-174.
[549] T. Łuczak, Component behaviour near the critical point, Random Structures and
Algorithms 1 (1990) 287-310.
[550] T. Łuczak, Cycles in a random graph near the critical point, Random Structures
and Algorithms 2 (1991) 421-440.
[551] T. Łuczak, Size and connectivity of the k-core of a random graph, Discrete Math-
ematics 91 (1991) 61-68.
[552] T. Łuczak, The chromatic number of random graphs, Combinatorica 11 (1991)
45-54.
[553] T. Łuczak, A note on the sharp concentration of the chromatic number of random
graphs, Combinatorica 11 (1991) 295-297.
[554] T. Łuczak, The phase transition in a random graph, Combinatorics, Paul Erdős
is Eighty, Vol. 2, eds. D. Miklós, V.T. Sós, and T. Szőnyi, Bolyai Soc. Math. Stud.
2, J. Bolyai Math. Soc., Budapest (1996) 399-422.
[555] T. Łuczak, Random trees and random graphs, Random Structures and Algorithms
13 (1998) 485-500.
[556] T. Łuczak, On triangle free random graphs, Random Structures and Algorithms
16 (2000) 260-276.
[557] T. Łuczak, B. Pittel and J. C. Wierman, The structure of a random graph at the
point of the phase transition, Transactions of the American Mathematical Society
341 (1994) 721-748
[558] T. Łuczak and P. Pralat, Chasing robbers on random graphs: zigzag theorem,
Random Structures and Algorithms 37 (2010) 516-524.
[559] T. Łuczak and A. Ruciński, Tree-matchings in graph processes SIAM Journal on
Discrete Mathematics 4 (1991) 107-120.
[560] T. Łuczak, A. Ruciński and B. Voigt, Ramsey properties of random graphs, Jour-
nal of Combinatorial Theory B 56 (1992) 55-68.
[561] T. Łuczak, Ł. Witkowski and M. Witkowski, Hamilton Cycles in Random Lifts
of Graphs, see arxiv.org.
[562] M. Mahdian and Y. Xu. Stochastic Kronecker graphs, Random Structures and
Algorithms 38 (2011) 453-466.
[563] H. Mahmoud, Evolution of Random Search Trees, John Wiley & Sons, Inc., New
York, 1992.
[564] H. Mahmoud, The power of choice in the construction of recursive trees,
Methodol. Comput. Appl. Probab. 12 (2010) 763-773.
 References 489

[565] H.M. Mahmoud and R.T. Smythe, A survey of recursive trees, Theory Probabil-
ity Math. Statist. 51 (1995) 1-27.
[566] H.M. Mahmoud, R.T. Smythe and J. Szymański, On the structure of random
plane-oriented recursive trees and their branches, Random Structures and Algo-
rithms, 4 (1993) 151-176.
[567] E. Marczewski, Sur deux propriétés des classes d’ensembles, Fundamenta Math-
ematicae 33 (1945) 303-307.
[568] D. Matula, The largest clique size in a random graph, Technical Report, Depart-
ment of Computer Science, Southern methodist University, (1976).
[569] D. Matula, Expose-and-merge exploration and the chromatic number of a ran-
dom graph, Combinatorica 7 (1987) 275-284.
[570] R. Mauldin, The Scottish Book: mathematics from the Scottish Cafe, Birkhäuser
Boston, 1981.
[571] G. McColm, First order zero-one laws for random graphs on the circle, Random
Structures and Algorithms 14 (1999) 239-266.
[572] G. McColm, Threshold functions for random graphs on a line segment, Combi-
natorics, Probability and Computing 13 (2004) 373-387.
[573] C. McDiarmid, Determining the chromatic number of a graph, SIAM Journal on
Computing 8 (1979) 1-14.
[574] C. McDiarmid, Clutter percolation and random graphs, Mathematical Program-
ming Studies 13 (1980) 17-25.
[575] C. McDiarmid, On the method of bounded differences, in Surveys in Combina-
torics, ed. J. Siemons, London Mathematical Society Lecture Notes Series 141,
Cambridge University Press, 1989.
[576] C. McDiarmid, Expected numbers at hitting times, Journal of Graph Theory 15
(1991) 637-648.
[577] C. McDiarmid, Random channel assignment in the plane, Random Structures
and Algorithms 22 (2003) 187-212.
[578] C. McDiarmid and T. Müller, On the chromatic number of random geometric
graphs, Combinatorica 31 (2011) 423-488.
[579] C. McDiarmid and B. Reed, Colouring proximity graphs in the plane, Discrete
Mathematics 199 (1999) 123-137.
[580] C. McDiarmid, A. Steger, and D. Welsh, Random planar graphs, Journal of Com-
binatorial Theory B 93 (2005) 187-205.
[581] B.D. McKay, Asymptotics for symmetric 0-1 matrices with prescribed row sums,
Ars Combinatoria 19 (1985) 15-25.
[582] B.D. McKay and N.C. Wormald, Asymptotic enumeration by degree sequence
of graphs with degrees o(n1/2 ), Combinatorica 11 (1991) 369-382.
[583] B.D. McKay and N.C. Wormald, The degree sequence of a random graph I, The
models, Random Structures and Algorithms 11 (1997) 97-117.
[584] N. Martin and J. England, Mathematical Theory of Entropy. Cambridge Univer-
sity Press, 2011.
[585] A. Mehrabian, Justifying the small-world phenomenon via random recursive
trees, mauscript (2014).
[586] A. Meir and J.W. Moon, The distance between points in random trees, J. Com-
binatorial Theory 8 (1970) 99-103.
490 References

[587] A. Meir and J.W. Moon, Cutting down random trees, J. Austral. Math. Soc. 11
(1970) 313-324.
[588] A. Meir and J.W. Moon, Cutting down recursive trees, Mathematical Biosciences
21 (1974) 173-181.
[589] A. Meir and J.W. Moon, On the altitude of nodes in random trees, Canadian J.
Math. 30 (1978) 997-1015.
[590] S.Micali and V.Vazirani, An O(|V |1/2 |E|) algorithm for finding maximum
matching in general graphs, Proceedings of the 21st IEEE Symposium on Foun-
dations of Computing (1980) 17-27.
[591] M. Mihail and C. Papadimitriou, On the eigenvalue power law, Randomization
and Approximation Techniques, 6th International Workshop (2002) 254-262.
[592] S. Milgram, The Small World Problem, Psychology Today 2 (1967) 60-67.
[593] I. M. Milne-Thomson, The Calculus of Finite Differences. Macmillian, London
(1951).
[594] M. Molloy, Cores in random hypergraphs and Boolean formulas, Random Struc-
tures and Algorithms 27 (2005) 124-135.
[595] M. Molloy and B. Reed, The Size of the Largest Component of a Random
Graph on a Fixed Degree Sequence, Combinatorics, Probability and Comput-
ing 7 (1998) 295-306.
[596] M. Molloy, H. Robalewska, R. W. Robinson and N. C. Wormald, 1- factorisa-
tions of random regular graphs, Random Structures and Algorithms 10 (1997)
305-321.
[597] R. Montgomery, Embedding bounded degree spanning trees in random graphs,
see arxiv.org.
[598] R. Montgomery, Sharp threshold for embedding combs and other spanning trees
in random graphs, see arxiv.org.
[599] R. Montgomery, Hamiltonicity in random graphs is born resilient, see arxiv.org.
[600] R. Montgomery, Spanning trees in random graphs, Advances in Mathematics
(2019) .
[601] R. Montgomery, Hamiltonicity in random graphs is born resilient, see arxiv.org.
[602] R. Montgomery, Topics in random graphs, Lecture notes, University of Birm-
ingham.
[603] J.W. Moon, The distance between nodes in recursive trees, London Math. Soc.
Lecture Notes Ser. 13 (1974) 125-132.
[604] T.F. Móri, On random trees, Studia Sci. Math. Hungar. 39 (2003) 143-155.
[605] T.F. Móri The maximum degree of the Barabási-Albert random tree, Com-
bin.Probab. Comput. 14 (2005) 339-348.
[606] K. Morris, A. Panholzer and H. Prodinger, On some parameters in heap ordered
trees, Combinatorics, Probability and Computing 13 (2004) 677-696.
[607] E. Mossel and A. Sly, Gibbs rapidly samples colorings of Gn,d/n , Journal of
Probability Theory and Related Fields 148 (2010) 37-69.
[608] T. Müller, Two point concentration in random geometric graphs, Combinatorica
28 (2008) 529-545.
[609] T. Müller, Private communication.
[610] T. Müller, X. Pérez and N. Wormald, Disjoint Hamilton cycles in the random
geometric graph, Journal of Graph Theory 68 (2011) 299-322.
 References 491

[611] T. Müller and M. Stokajović, A threshold for the Maker-Breaker clique game,
Random Structurs and Algorithms 45 (2014) 318-341.
[612] L. Mutavchiev, A limit distribution related to a random mappings and its appli-
cation to an epidemic process, Serdica 8 (1982) 197-203.
[613] H. S. Na and A. Rapoport, Distribution of nodes of a tree by degree, Mathemat-
ical Biosciences 6 (1970) 313-329.
[614] A. Nachmias and Y. Peres, Critical random graphs: diameter and mixing time,
Annals of Probability 36 (2008) 1267-1286.
[615] A. Nachmias and Y. Peres, The critical random graph, with martingales, Israel
Journal of Mathematics 176 (2010) 29-41.
[616] C. Nair, B. Prabhakar and M. Sharma, Proofs of the Parisi and Coppersmith-
Sorkin random assignment conjectures, Random Structures and Algorithms 27
(2005) 413-444.
[617] D. Najock and C.C. Heyde (1982). On the number of terminal vertices in certain
random trees with an application to stemma construction in philology, Journal
of Applied Probability 19 (1982) 675-680.
[618] C. St.J. Nash-Williams, Edge-disjoint spanning trees of finite graphs, Journal of
the London Mathematical Society (1961) 445-450.
[619] R. Nenadov, Y. Person, N. Škoric and A. Steger, An algorithmic framework for
obtaining lower bounds for random Ramsey problems, see arxiv.org.
[620] R. Nenadov and A. Steger, A Short proof of the Random Ramsey Theorem, to
appear in Combinatorics, Probability and Computing.
[621] R. Nenadov and N. Škorić, Powers of Hamilton cycles in random graphs and
tight Hamilton cycles in random hypergraphs, see arxiv.org.
[622] R. Nenadov, A. Steger and Stojacović, On the threshold for the Maker-Breaker
H-game, see arxiv.org.
[623] R. Nenadov, A. Steger and M. Trujić, Resilience of Perfect matchings and
Hamiltonicity in Random Graph Processes, see arxiv.org.
[624] S. Nicoletseas, C. Raptopoulos and P.G. Spirakis, The existence and efficient
construction of large independent sets in general random intersection graph, In:
ICALP 2004, LNCS 3142, Springer (2004) 1029-1040.
[625] S. Nikoletseas, C. Raptopoulos and P.G. Spirakis, On the independence number
and Hamiltonicity of uniform random intersection graphs, Theoret. Comput. Sci.
412 (2011) 6750-6760.
[626] I. Norros and H. Reittu, On a conditionally Poissonian graph process, Adv. in
Appl. Probab. 38 (2006) 59-75.
[627] D. Osthus, H. J. Prömel, and A. Taraz, On random planar graphs, the number
of planar graphs and their triangulations, Journal of Combinatorial Theory B 88
(2003) 119-134.
[628] Z. Palka, On the number of vertices of given degree in a random graph, J. Graph
Theory 8 (1984) 167-170.
[629] Z. Palka, Extreme degrees in random graphs, Journal of Graph Theory 11 (1987)
121-134.
[630] E. Palmer and J. Spencer, Hitting time for k edge disjoint spanning trees in a
random graph, Period Math. Hungar. 91 (1995) 151-156.
[631] A. Panholzer and H. Prodinger, The level of nodes in increasing trees revisited,
Random Structures and Algorithms 31 (2007) 203-226.
492 References

[632] A. Panholzer and G. Seitz, Ordered increasing k-trees: Introduction and analysis
of a preferential attachment network model, Discrete Mathematics and Theoret-
ical Computer Science (AofA 2010) (2010) 549-564.
[633] [PKBnV10] F. Papadopoulos, D. Krioukov, M. Boguñá and A. Vahdat, Greedy
forwarding in dynamic scale-free networks embedded in hyperbolic metric
spaces, Proceedings of the 29th Conference on Information Communications,
INFOCOM10 (2010) 2973- 2981.
[634] G. Parisi, A conjecture on Random Bipartite Matching, Pysics e-Print archive
(1998).
[635] E. Peköz, A. Röllin and N. Ross, Degree asymptotics with rates for preferential
attachment random graphs, The Annals of Applied Probability 23 (2013) 1188-
1218.
[636] M. Penrose, On k-connectivity for a geometric random graph, Random Struc-
tures and Algorithms 15 (1999) 145-164.
[637] M. Penrose, Random Geometric Graphs, Oxford University Press, 2003.
[638] N. Peterson and B. Pittel, Distance between two random k-out graphs, with and
without preferential attachment, see arxiv.org.
[639] J. Pitman, Random mappings, forest, and subsets associated with the Abel-
Cayley-Hurwitz multinomial expansions, Seminar Lotharingien de Combina-
toire 46 (2001) 46.
[640] B. Pittel, On distributions related to transitive closures of random finite map-
pings, The Annals of Probability 11 (1983) 428-441.
[641] B.Pittel, On tree census and the giant component in sparse random graphs, Ran-
dom Structures and Algorithms 1 (1990), 311-342.
[642] B. Pittel, On likely solutions of a stable marriage problem, Annals of Applied
Probability 2 (1992) 358-401.
[643] B. Pittel, Note on the heights of random recursive trees and random m-ary search
trees, Random Structures and Algorithms 5 (1994) 337-347.
[644] B. Pittel, On a random graph evolving by degrees, Advances in Mathematics 223
(2010) 619-671.
[645] B. Pittel, J. Spencer and N. Wormald, Sudden emergence of a giant k-core in a
random graph, Journal of Combinatorial Theory B 67 (1996) 111-151.
[646] B. Pittel, L. Shepp and E. Veklerov, On the number of fixed pairs in a random
instance of the stable marriage problem, SIAM Journal on Discrete Mathematics
21 (2008) 947-958.
[647] B. Pittel and R. Weishar, The random bipartite nearest neighbor graphs, Random
Structures and Algorithms 15 (1999) 279-310.
[648] D. Poole, On weak hamiltonicity of a random hypergraph, see arxiv.org.
[649] L. Pósa, Hamiltonian circuits in random graphs, Discrete Mathematics 14 (1976)
359-364.
[650] P. Pralat, N. Wormald, Meyniel’s conjecture holds for random d-regular graphs,
see arxiv.org.
[651] P. Pralat, Cleaning random graphs with brushes, Australasian Journal of Combi-
natorics 43 (2009) 237-251.
[652] P. Pralat, Cleaning random d-regular graphs with Brooms, Graphs and Combi-
natorics 27 (2011) 567-584.
 References 493

[653] V. M. Preciado and A. Jadbabaie, Spectral analysis of virus spreading in random

geometric networks, Proceedings of the 48th IEEE Conference on Decision and
Control (2009).
[654] H. Prodinger, Depth and path length of heap ordered trees, International Journal
of Foundations of Computer Science 7 (1996) 293-299.
[655] H. Prodinger and F.J. Urbanek, On monotone functions of tree structures, Dis-
crete Applied Mathematics 5 (1983) 223-239.
[656] H. Prüfer, Neuer Beweis eines Satzes über Permutationen, Archives of Mathe-
matical Physics 27 (1918) 742-744.
[657] J. Radakrishnan, Entropy and Counting, www.tcs.tifr.res.in/ jaiku-
mar/Papers/EntropyAndCounting.pdf
[658] A. Rényi and G. Szekeres, On the height of trees,Journal Australian Mathemat-
ical Society 7 (1967) 497-507.
[659] M. Radcliffe and S.J. Young, Connectivity and giant component of stochastic
Kronecker graphs, see arxiv.org.
[660] S. Rai, The spectrum of a random geometric graph is concentrated, Journal of
Theoretical Probability 20 (2007) 119-132.
[661] A. Rényi, On connected graphs I, Publ. Math. Inst. Hungar. Acad. Sci. 4 (1959)
385-387.
[662] O. Riordan, Spanning subgraphs of random graphs, Combinatorics, Probability
and Computing 9 (2000) 125-148.
[663] O. Riordan and A. Selby, The maximum degree of a random graph, Combina-
torics, Probability and Computing 9 (2000) 549-572.
[664] O. Riordan, The small giant component in scale-free random graph, Combin.
Prob. Comput. 14 (2005) 897-938.
[665] O. Riordan, Long cycles in random subgraphs of graphs with large minimum
degree, see arxiv.org.
[666] O. Riordan and L. Warnke, Achlioptas process phase transitions are continuous,
Annals of Applied Probability 22 (2012) 1450-1464.
[667] O. Riordan and N. Wormald, The diameter of sparse random graphs, Combina-
torics, Probability and Computing 19 (2010) 835-926.
[668] R.W. Robinson and N.C.Wormald, Almost all cubic graphs are Hamiltonian,
Random Structures and Algorithms 3 (1992) 117-126.
[669] V. Rödl and A. Ruciński, Lower bounds on probability thresholds for Ramsey
properties, in Combinatorics, Paul Erdős is Eighty, Vol.1, Bolyai Soc. Math.
Studies (1993), 317-346.
[670] V. Rödl, A. Ruciński and E. Szemerédi, An approximate Dirac-type theorem for
k-uniform hypergraphs, Combinatorica 28 (2008) 220-260.
[671] R.W. Robinson and N.C.Wormald, Almost all regular graphs are Hamiltonian,
Random Structures and Algorithms 5 (1994) 363-374.
[672] V. Rödl and A. Ruciński, Threshold functions for Ramsey properties Journal of
the American Mathematical Society 8 (1995) 917-942.
[673] D. Romik, The Surprising Mathematics of Longest Increasing Subsequences,
Cambridge University Press, New York (2014).
[674] S.M. Ross, A random graph, J. Applied Probability 18 (1981) 309-315.
494 References

[675] H. Rubin and R. Sitgreaves, Probability distributions related to random trans-

formations on a finite set, Tech. Rep. 19A, Applied Mathematics and Statistics
Laboratory, Stanford University, (1954).
[676] A. Ruciński, When a small subgraphs of a random graph are normally dis-
tributed, Probability Theory and Related Fields 78 (1988) 1-10.
[677] A. Ruciński and A. Vince, Strongly balanced graphs and random graphs, Journal
of Graph Theory 10 (1986) 251-264.
[678] A. Ruciński, Matching and covering the vertices of a random graph by copies of
a given graph,Discrete Mathematics 105 (1992) 185-197.
[679] A. Ruciński and N. Wormald, Random graph processes with degree restrictions,
Combinatorics, Probability and Computing 1 (1992) 169-180.
[680] A. Rudas, B,Toth and B. Való, Random trees and general branching processes,
Random Structures and Algorithms 31 (2007) 186-202.
[681] K. Rybarczyk and D. Stark, Poisson approximation of the number of cliques in
a random intersection graph Journal of Applied Probability 47 (2010) 826 - 840.
[682] K. Rybarczyk and D. Stark, Poisson approximation of counts of subgraphs in
random intersection graphs, submitted.
[683] K. Rybarczyk, Equivalence of the random intersection graph and G(n, p), Ran-
dom Structures and Algorithms 38 (2011) 205-234.
[684] K. Rybarczyk, Sharp threshold functions for random intersection graphs via a
coupling method, The Electronic Journal of Combinatorics 18(1) (2011) #P36.
[685] K. Rybarczyk, Diameter, connectivity, and phase transition of the uniform ran-
dom intersection graph, Discrete Mathematic 311 (2011) 1998-2019.
[686] K. Rybarczyk, Constructions of independent sets in random intersection graphs,
Theoretical Computer Science 524 (2014) 103-125.
[687] K. Rybarczyk, The coupling method for inhomogeneous random intersection
graphs, see arxiv.org.
[688] V.N. Sachkov, Combinatorial Methods in Discrete Mathematics, Cambridge
University Press, (1996).
[689] A.A. Saposhenko, Metric properties of almost all boolean functions, (in Rus-
sian), Diskretny Analiz 10 (1967) 91-119.
[690] A.A. Saposhenko, Geometric structure of almost all boolean functions, (in Rus-
sian), Problemy Kibernetiki 30 (1975) 227-261.
[691] D. Saxton and A. Thomason, Hypergraph containers, see arxiv.org.
[692] D. Saxton and A. Thomason, Online containers for hypergraphs with applica-
tions to linear equations, see arxiv.org.
[693] M. Schacht, Extremal results for random discrete structures, see arxiv.org.
[694] J. Schmidt and E. Shamir, A threshold for perfect matchings in random d-pure
hypergraphs, Discrete Mathematics 45 (1983) 287-295.
[695] J. Schmidt-Pruzan and E. Shamir, Component structure in the evolution of ran-
dom hypergraphs, Combinatorica 5 (1985) 81-94.
[696] R. Schneider, Convex Bodies: The Brunn-Minkowski Theory, Cambridge Uni-
versity Press, 2013.
[697] A. Scott, On the concentration of the chromatic number of random graphs (2008)
see arxiv.org.
[698] C. Seshadhri, A. Pinar, and T.G. Kolda, An in-depth analysis of stochastic Kro-
necker graphs, The Journal of the ACM, see arxiv.org.
 References 495

[699] E. Shamir and J. Spencer, Sharp concentration of the chromatic number od ran-
dom graphs Gn,p , Combinatorica 7 (1987) 121-129.
[700] C. Shannon, A Mathematical Theory of Communication, Bell System Technical
Journal 27 (1948) 379423.
[701] B. Selivanov, Pereschislenie odnorodnyckh hypergrafov c protstoǐ struktutoǐ,
Kombinatornyǐ Analiz. 2 (1972) 60-67.
[702] S. Shelah and J. Spencer, Zero-One laws for sparse random graphs, Journal of
the American Mathematical Society 1 (1988) 97-115.
[703] L.A. Shepp and S.P. Lloyd, Ordered cycle length in a random permutation, Trans.
Amer. Math. Soc. 121 (1966) 340-357.
[704] L. Shi and N. Wormald, Coloring random regular graphs, Combinatorics, Prob-
ability and Computing 16 (2007) 459-494.
[705] L. Shi and N. Wormald, Coloring random 4-regular graphs, Combinatorics,
Probability and Computing 16 (2007) 309-344.
[706] M. Šileikis and L. Warnke. A counterexample to the DeMarco-Kahn Upper Tail
Conjecture, see arxiv.org.
[707] S. Sivasubramanian, Spanning trees in complete uniform hypergraphs and a con-
nection to r-extended Shi hyperplane arrangements, see arxiv.org.
[708] B. Söderberg, General formalism for inhomogeneous random graphs, Phys. Rev.
E 66 (2002) 066121.
[709] J. Spencer, Private Communication.
[710] J. Spencer, Enumerating graphs and Brownian motion, Communications in Pure
and Applied Mathematics 50 (1997) 291-294.
[711] J. Spencer, The Strange Logic of Random Graphs, Springer, Berlin 2001.
[712] J. Spencer and N. Wormald, Birth control for giants, Combinatorica 27 (2007)
587-628.
[713] D. Stark, The vertex degree distribution of random intersection graphs, Random
Structures Algorithms 24 (2004) 249-258.
[714] J. M. Steele, On Frieze’s ζ (3) limit forlengths of minimal spanning trees, Dis-
crete Applied Mathematics 18 (1987) 99-103.
[715] C. Stein, A bound for the error in the normal approximation to the distribution of
a sum of dependent random variables, In:Proceedings of the Sixth Berkeley Sym-
posium on Mathematical Statistics and Probability , 1970, Univ. of California
Press, Berkeley, Vol. II (1972) 583-602.
[716] V.E. Stepanov, On the distribution of the number of vertices in strata of a random
tree, Theory of Probability and Applications 14 (1969) 65-78.
[717] V.E. Stepanov, Limit distributions of certain characteristics of random mappings,
Theory Prob. Appl. 14 (1969) 612-636.
[718] M. Stojaković and T. Szabó, Positional games on random graphs, Random Struc-
tures and Algorithms 26 (2005) 204-223.
[719] Ch. Su, J. Liu and Q. Feng, A note on the distance in random recursive trees,
Statistics & Probability Letters 76 (2006) 1748-1755
[720] B. Sudakov and V. Vu, Local resilience of graphs, Random Structures and Algo-
rithms 33 (2008) 409-433.
[721] J. Szymański, On the nonuniform random recursive tree, Annals of Discrete
Math. 33 (1987) 297-306.
496 References

[722] J. Szymański, On the maximum degree and the height of random recursive tree,
In:Random Graphs’87, Wiley, New York (1990) 313-324.
[723] J. Szymański, Concentration of vertex degrees in a scale free random graph pro-
cess, Random Structures and Algorithms 26 (2005) 224-236.
[724] J. Tabor, On k-factors in stochastic Kronecker graphs, (2014) unpublished
manuscript.
[725] M. Talagrand, Concentration of measures and isoperimetric inequalities in prod-
uct spaces, Publications Mathematiques de IT.H.E.S. 81 (1996) 73-205.
[726] M.A. Tapia and B.R. Myers, Generation of concave node-weighted trees, IEEE
Trans. Circuits Systems 14 (1967) 229-230.
[727] S-H. Teng and F.F. Yao, k-nearest-neighbor clustering and percolation theory,
Algorithmica 49 (2007) 192-211.
[728] A.Thomason, A simple linear expected time algorithm for Hamilton cycles, Dis-
crete Mathematics 75 (1989) 373-379.
[729] P. Turán, On an extremal problem in graph theory, Matematikai és Fizikai Lapok
(in Hungarian) 48 (1941) 436-452.
[730] T.S. Turova, Phase transition in dynamical random graphs, J. Stat. Phys. 123
(2006) 1007-1032.
[731] T.S. Turova, Diffusion approximation for the componentsin critical inhomoge-
neous random graphsof rank 1, Random Structures Algorithms 43 (2013) 486-
539.
[732] T.S. Turova, Asymptotics for the size of the largest component scaled to ”log n”
in inhomogeneous random graphs, Arkiv fur Matematik 51 (2013) 371-403.
[733] W. T. Tutte, A census of planar triangulations, Canadian Journal of Mathematics
14 (1962) 21-38.
[734] W. T. Tutte, A census of planar maps, Canadian Journal of Mathematics 15
(1963) 249-271.
[735] W.F. de la Vega, Long paths in random graphs, Studia Sci. Math. Hungar. 14
(1979) 335-340.
[736] A. M. Vershik and A. A. Shmidt, Limit measures arising in the asymptotic theory
of symmetric groups, I, Theory of Probability and its Applications 22 (1977) 70-
85.
[737] E. Vigoda, Improved bounds for sampling colorings, Journal of Mathematical
Physics 41 (2000) 1555-1569.
[738] O.V. Viskov, Some comments on branching process, Mathematical Notes 8
(1970) 701-705.
[739] V. Vu. On some degree conditions which guarantee the upper bound of chromatic
(choice) number of random graphs, Journal of Graph Theory 31 (1999) 201-226.
[740] V. Vu, Spectral norm of random matrices, Combinatorica 27 (2007) 721-736.
[741] D.W. Walkup, Matchings in random regular bipartite graphs, Discrete Mathe-
matics 31 (1980) 59-64.
[742] D.W. Walkup, On the expected value of a random asignment problem, SIAM
Journal on Computing 8 (1979) 440-442.
[743] J. Wástlund, An easy proof of the ζ (2) limit in the random assignment problem,
Electronic Communications in Probability 14 (2009) 261-269.
[744] K. Weber, Random graphs - a survey, Rostock Math. Kolloq. (1982) 83-98.
[745] D. West, Introduction to Graph Theory - Second edition, Prentice Hall 2001.
 References 497

[746] E.P. Wigner, On the distribution of the roots of certain symmetric matrices, An-
nals of Mathematics 67 (1958) 325-327.
[747] L. B. Wilson, An analysis of the stable assignment problem, BIT 12 (1972) 569-
575.
[748] E.M. Wright, The number of connected sparsely edged graphs, Journal of graph
Theory 1 (1977) 317-330.
[749] E.M. Wright, The number of connected sparsely edged graphs II, Journal of
graph Theory 2 (1978) 299-305.
[750] E.M. Wright, The number of connected sparsely edged graphs III, Journal of
graph Theory 4 (1980) 393-407.
[751] N.C. Wormald, Models of random regular graphs, In J.D. Lamb and D.A. Preece,
editors, Surveys in Combinatorics, 1999, volume 267 of London Mathematical
Society Lecture Note Series, Cambridge University Press (1999) 239-298.
[752] N. C. Wormald, The differential equation method for random graph processes
and greedy algorithms, Lectures on approximation and randomized algorithms
(M. Karoński and H.J. Prömel, eds.), Advanced Topics in Mathematics, Polish
Scientific Publishers PWN, Warsaw, 1999, pp. 73-155.
[753] O. Yagan and A. M. Makowski, Zero-one laws for connectivity in random key
graphs, IEEE Transations on Information Theory 58 (2012) 2983-2999.
[754] B. Ycart and J. Ratsaby, The VC dimension of k-uniform random hypergraphs,
Random Structures and Algorithms 30 (2007) 564-572.
[755] S.J. Young, Random Dot Product Graphs: A Flexible Model of Complex Net-
works, ProQuest (2008)
[756] S.J. Young and E.R. Scheinerman, Random dot product graph models for social
networks, In: Algorithms and Models for the Web-Graph, LNCS 4863, Springer
(2007) 138-149.
[757] Z. Zhang, L. Rong and F. Comellas, High dimensional random apollonian net-
works, Physica A: Statistical Mechanics and its Applications 364 (206) 610-618.
[758] J. Zhao, O. Yagan and V. Gligor, On k-connectivity and minimum vertex degree
in random s-intersection graphs, SIAM Meeting on Analytic Algorithmics and
Combinatorics, (2015), see arxiv.org.
[759] T. Zhou, G. Yan and B.H. Wang, Maximal planar networks with large cluster-
ing coefficient and power-law degree distribution, Physics Reviews E 71 (2005)
046141.
 26
Indices

498
 Author Index

Šilekis, M, 217 Blackwell, P., 254

Škorić, N., 154, 268 Bloznelis, M., 252, 253
Łuczak, T., vii, 8, 37, 40, 41, 47, 50, 70, 80, Bode, M., 255
110, 127, 128, 139, 141, 143, 229, 252, Boguñá, M., 255
294, 356, 364, 405, 408–410, 424, 436 Bohman, T., 47, 154, 295, 356, 405–407
Bollobás, B., vi, vii, 10, 18, 37, 40, 41, 48, 52,
Achlioptas, D., 47, 137, 229, 294
62, 74, 76, 85, 89, 106–108, 111, 123,
Aiello, W., 385
125, 128, 191–193, 195, 200, 210, 245,
Ajtai, M., 48, 89, 94, 410
364, 367, 385, 400, 406, 407, 410
Albert, R., 367
Bonato, A., 385, 405
Aldous, D., 40, 41, 394
Borgs, C., 385
Alon, N., 103, 110, 137, 176, 405, 408, 412,
Bourgain, J., 19
445
Bradonjić, M., 254
Amit, A., 409
Brautbar, M., 385
Angel, O., 400
Brightwell, G., 153
Avin, C., 411
Britton, T., 193, 253
Babai, L., 153 Broder, A., 70, 230, 409
Bal, D., 403 Bui, T., 409
Balister, P., 364 Burtin, Y., 111, 410
Balogh, J., 140, 245, 295 Callaway, D., 191
Banderier, C., 408 Candellero, E., 255
Barabási, L., 367 Canfield, R., 41, 200, 408
Barbour, A., 62, 79, 80, 253, 422–424 Capalbo, M., 412
Bartnicki, T., 405 Caro, Y., 71
Bayati, M., 121 Catlin, P., 48
Beck, J., 403 Cerny, J., 411
Behrisch, M., 251 Chakrabarti, D., 185, 194
Ben-Shimon, S., 108, 404 Chaudhuri, S., 409
Bender, E., 41, 200, 408 Chayes, J., 385
Benjamini, I., 406, 411 Chebolu, P., 109, 409
Bennett, P., 403, 407 Chen, L., 422
Berge, C., 278 Chen, W., 255
Bernstein, S., 429 Chung, F., 178, 184, 385, 457
Beveridge, A., 404 Chvátal, V., 18, 138, 403
Bhamidi, S., 193, 399 Coja–Oghlan, A., 129, 137, 409
Blackbourn, S., 252 Collevecchio, A., 386

499
500 Author Index

Conlon, D., 140, 143 Fountoulakis, N., 47, 255, 295, 351, 411
Cooley, O., 294 Fraiman, N., 192
Cooper, C., 48, 106, 107, 229, 268, 295, 296, Frankl, F., 457
365, 373, 375, 385, 386, 399, 403, 411 Friedgut, E., 19
Cover, T., 455 Friedman, J., 230
Damarackas, J., 253 Friedrich, T., 254, 255
Frieze, A., 18, 47, 70, 71, 85, 99, 106–109,
Dani, V., 119
119, 137, 138, 154, 217, 229, 230, 264,
de la Vega, W., 94
267, 268, 275, 294–296, 345, 347, 351,
Deijfen, M., 193, 253
356, 364, 365, 373, 375, 385–387,
Dellamonica, D., 412
399–401, 403–405, 407–411
DeMarco, B., 154
Frieze, A.,, 71
Devroye, L., 192
Diaz, J., 245, 254 Gale, D., 412
Diepetro, R., 252 Gamarnik, D., 47, 121, 400
Ding, J., 47, 411 Gao, P., 109, 294, 295, 386
Dudek, A., 106, 136, 217, 230, 275, 404, 406 Gao, Z., 408
Durrett, R., 191, 228, 385, 418 Gebauer, H., 404
Dyer, M., 99, 138, 294, 400, 410 Gerke, S., 110, 143, 252, 408
Gharamani, Z., 194
Ebrahimzadeh, E., 386 Gilbert, E., vi, 3
Edmonds, J., 96 Giménez, O., 408
Edmonson-Jones, M., 254 Ginibre, J., 428
Efthymiou, C., 139 Glebov, R., 107, 108, 404
Egorychev, G., 265 Glebskii, Y., 407
Elsässer, R., 254 Gligor, V., 252
England, J., 455 Godehardt, E., 252, 253
Ercal, G., 411 Goel, A., 254
Erdős, P., vi, 3, 20, 40, 48, 60, 62, 64, 68, 74, Goetze, F., 253
79, 80, 83, 85, 94, 125, 154, 295, 403, Gowers, T., 140, 143
406, 407, 410 Goyal, N., 365
Eschenauer, L., 252 Grable, D., 407
Füredi, Z., 103, 129 Graham, R., 457
Fagin, R., 407 Grandoni, F., 254
Falikman, D., 265 Gray, R., 455
Faloutsos, C., 185, 194, 367 Greenhill, C., 294, 409
Faloutsos, M., 367 Griffiths, S., 406
Faloutsos, P., 367 Grigorescu, E., 408
Fang, W., 255 Grimmett, G., 125
Farczadi, L., 386 Groër, C., 194
Feldman, V., 408 Grytczuk, J., 405
Fenner, T., 107, 108, 345, 385 Gugelmann, L., 255
Ferber, A, 156 Gupta, P., 241
Ferber, A., 265, 267, 268, 275, 293, 403, 404, Gurevich, Y., 108
412 Haber, S., 108, 230, 405
Fernholz, D., 139 Hamm, A., 295
Fill, J., 232 Harris, T., 428
Fiz Pontiveros, G., 406 Hatami, H., 137, 229
Flajolet, P., 408 He, J., 71
Flaxman, A., 47, 138, 385, 400 Heckel, A., 71
Fortuin, C., 428 Hefetz, D., 109, 268, 403, 404
Foulds, L., 410 Hoeffding, W., 431, 435
 Author Index 501

Holst, L., 423, 424 Kolda, T,, 194

Hooghiemstra, G., 193 Komlós, J., 48, 89, 94, 129, 410
Hopcroft, J., 191 Kordecki, W., 62
Horn, P., 187 Kovalenko, I., 171, 172
Hu, G., 255 Kozma, G., 385, 411
Kravitz, D., 47
Ince, N., 399
Kreuter, B., 143
Ivchenko, G., 62
Krioukov, D., 255
Jadbabaie, R., 254 Krishnamachari, B., 254
Jannsen, J., 386 Krivelevich, M., 47, 48, 94, 99, 107–109, 133,
Janson, S., vii, 40, 47, 62, 80, 191–193, 229, 137, 138, 229, 230, 245, 268, 294, 403,
387, 390, 399, 402, 403, 409, 423, 424, 404, 408
436 Krivelevich. M., 95
Janssen, J., 385 Krohmer, A., 255
Jaworski, J., 252, 253, 365 Kronenberg, G., 265, 267, 403
Jerrum, M., 138, 229 Kucera, L., 408
Jiang, J., 294 Kumar, P., 241
Johansson, A., 110, 279 Kurauskas, V., 251, 253
Johansson, T., 365
Lageras, A., 251, 253
Jonasson, J., 411
Lavrov, M., 405
Jordan, 254, 385
Lee, C., 94, 99, 108, 156
Juhász, F., 129 Leighton, T., 409
Kühn, D., 47, 107 Lelarge, M., 295
Kahle, M., 410 Lennon, C., 412
Kahn, J., 109, 110, 154, 230, 279, 295 Leskovec, J., 185, 194
Kalai, G., 19 Lev, A., 71
Kalikow, S., 191 Liagonkii, M., 407
Kamčev, N., 230 Liang, H., 71
Kang, M., 47, 193, 194, 294, 408 Lin, K., 192
Kannan, R., 408 Lindholm, M., 251, 253
Karger, D., 190 Linial, N., 409, 410
Karoński, M., 52, 62, 76, 79, 80, 193, 194, Linusson, S., 393
231, 237, 253, 294, 423, 424 Loh, P., 47, 275, 403
Karp, R., 109, 257, 349, 351, 400, 453 Long, E., 265, 267
Kasteleyn, P., 428 Lu, L., 178, 184, 385
Keevash, P., 406 Lubetzky, E., 47, 107, 109, 268, 407, 411
Kemkes, G., 230 Lucier, B., 385
Keusch, R., 405 Luczak, M., 229
Khanna, S., 385 Müller, T., 241, 245, 254, 255, 404
Khosla, M., 295 Mahdian, M., 186
Kierstead, H., 405 Mahoney, M., 255
Kim, J., 47, 217, 294, 402, 406 Makai, T., 193, 194
Kinnersley, W., 406 Makowski, A., 252
Kitsak, M., 255 Mancini, L., 252
Kleinberg, J., 185, 191, 194, 382 Marchetti-Spaccemela, A., 254
Kliem, S., 193 Marczewski, E., 231
Knox, F., 107 Martin, A., 455
Knuth, D., 40, 47, 97, 412 Martin, R., 295
Koch, C., 193, 194, 294 Martin-Löf, A., 193
Kogan, D., 407 Matoušek, J., 409
Kohayakawa, Y., 143, 278, 410, 412 Matula, D., 117, 123
502 Author Index

McColm, G., 254 Petit, J., 254

McDiarmid, C., 107, 125, 138, 246, 248, 254, Picolleli, M., 407
264, 293, 399, 400, 408 Pikhurko, O., 154
McDowell, A., 110 Pinar, A., 194
McKay, B., 41, 63, 200, 403 Pittel, B., 37, 40, 47, 97, 99, 109, 412
Mehrabian, A., 386 Poole, S., 194, 278
Mei, A., 252 Prömel, H., 143, 408
Melsted, P., 109, 351 Prabhakar, B., 393
Meshulam, R., 410 Pralat, P., 106, 136, 385, 386, 403, 405, 406
Micali, S., 96 Preciado, V, 254
Mihail, M., 385 Rödl, V., 141, 143, 154, 156, 412
Miklós, D., 385 Röllin, A., 385
Milgram, S., 381 Rényi, A., vi, 3, 20, 40, 62, 64, 68, 74, 79, 83,
Mitsche, D., 136, 245, 406 85, 94
Mitzenmacher, M., 294 Radakrishnan, J., 457
Molloy, M., 212, 229, 268, 294, 296, 402 Radcliffe, J., 187, 193
Montgomery, R., 109, 156 Rademacher, L., 365
Moore, C., 119, 137, 229 Radhakrishnan, J., 252
Morris, R., 140, 406 Rai, S., 254
Mossell, E., 139 Ramachandran, V., 139
Motwani, R., 97, 412 Raptopoulos, C., 252, 253
Mousset, F., 403 Ratsaby, J., 295
Mubayi, D., 295, 407 Reed, B., 212, 229, 248, 296, 411
Nachmias, A., 41, 411 Reinert, G., 192, 253
Nair, C., 393 Reittu, H., 193
Naor, A., 137, 404 Reyzin, L., 408
Nash-Williams, C., 109 Rinnooy-Kan, A, 349, 351
Naves, H., 108 Riordan, O., 47, 63, 71, 99, 108, 139,
Nenadov, R., 141, 154, 156, 268, 403, 412 191–193, 367, 385, 400, 406
Newman, M., 191 Robalewska, H., 402
Nicoletseas, R., 252, 253 Robinson, R., 229, 402
Noever, A., 156, 268 Roditty, Y., 71
Ross, N., 385
Norros, I., 193
Rozenman, E., 409
Noy, M., 408
Ruciński, A., vii, 50, 52, 62, 74, 76, 78–80,
Osthus, D., 47, 107, 408 110, 141, 154, 156, 217, 407, 409, 412,
Pérez-Giménez, X., 109, 230, 245 423, 424, 436, 453
Pósa, L., 89 Ruszinko, M., 295, 399
Palka, Z., 62, 63, 365 Rybarczyk, K., 232, 236, 240, 251–253
Panagiotou, K., 137, 153, 255, 295, 351 Söderberg, B., 191
Panconesi, A., 252 Sós, V., 154
Papadimitriou, C., 385 Samotij, W., 107, 140
Papadopolous, F., 255 Saposhenko, A., 410
Parisi, G., 393 Sarkar, A., 364
Pegden, W., 137, 385 Sato, C., 109, 386
Peköz, R., 385 Sauerwald, T., 254
Penrose, M., 241, 254 Saxton, D., 140, 141
Peres, Y., 41, 47, 411 Schacht, M., 140, 143
Perkins, W., 47 Schaeffer, G., 408
Person, Y., 154, 278 Scheinerman, E., 193, 231, 232, 237
Peter, U., 156, 255, 268, 403, 412 Schickinger, T., 143
 Author Index 503

Schlatter, D., 408 Thaler, J., 294

Schmidt, J., 279 Thoma, L., 399
Scott, A., 135 Thomas, J., 455
Selby, A., 63 Thomason, A., 10, 108, 140, 141
Selkow, S., 60 Trujić, M, 156
Sen, S., 193 Tsourakakis, C., 71, 230, 386
Serna, J., 254 Tsur, G., 406
Seshadhri, C., 194 Turan, P., 72
Shamir, E., 279, 409 Turova, T., 191
Shannon, C., 455 Tusnády, G., 367
Shapley, L., 412 Tutte, W., 408
Sharma, M., 393 Tuza, Z., 71
Shearer, J., 457
Ueckerdt, M., 251
Shelah, S., 108, 407
Upfal, E., 70, 230
Shepp, L., 412
Shi, L., 230 Vahdat, A., 255
Shikhelman, C., 403 van den Esker, H., 385
Shinkar, I., 406 van der Hofstad, R., 193, 399
Simonovits, M., 153, 154 van Leeuwaarden, J., 193
Singer-Cohen, K., 231, 232, 237 Vazirani, V., 96
Sipser, M., 109, 409, 453 Veklerov, E., 412
Sivasubramanian, S., 270 Vempala, S., 365, 408
Sly, A., 139, 411 Vera, J., 385
Smit, I., 365 Vigoda, E., 138
Smyth, C., 154, 230, 295 Vilenchik, D., 137
Soria, M., 408 Vince, A., 74
Sorkin, G., 47, 401 Vohra, R., 349, 351
Speiss, S., 275 Voigt, B., 141
Spencer, J., 37, 41, 47, 106, 111, 153, 367, Vu, V., 108, 110, 129, 133, 138, 217, 229, 279,
399, 407, 410, 445 385
Spirakis, P., 252, 253 Vu,V, 156
Stark, D., 240, 251 Vu,V., 155, 167
Stauffer, A., 254, 255 Wästlund, J., 393
Steger, A, 156 Wade, A., 385
Steger, A., 137, 141, 143, 153, 154, 405, 408 Walkup, D., 349, 400
Stein, C., 190, 422 Walters, M., 245, 364
Stojaković, M., 403, 404 Wang, X., 193, 386
Strogatz, S., 191 Warnke, L., 47
Sudakov, B., 47, 48, 94, 95, 99, 107, 108, 137, Weber, K., 410
138, 155, 156, 167, 229, 230, 268, 294, Weiss, B., 191
400, 405, 408 Welsh, 408
Suen, S., 70, 230, 406 West, 180
Sullivan, B., 194 Wierman, J., 40, 47
Szabó, T., 109, 403, 404 Wigner, E., 131
Szemerédi, E., 48, 89, 94, 156, 230, 410, 412 Wilson, D., 400
Talagrand, M., 445 Wilson, L., 412
Talanov, V., 407 Wilson, R., 386
Taraz, A., 251, 408 Winkler, P., 406
Teixeira, A., 411 Witkowski, M., 409
Teng, S., 364 Witkowski, Ł., 409
Tetali, P., 121, 294 Wormald, N., 38, 47, 63, 109, 138, 139, 200,
504 Author Index

210, 229, 230, 245, 295, 386, 402, 403,

405, 407, 408, 411, 453
Wright, E., 41
Xiao, Y., 408
Xu, Y., 186
Yagan, O., 252
Yan, G., 386
Yao, F., 364
Ycart, B., 295
Young, S., 193
Yuster, R., 71, 110
Zhan, Y., 194
Zhao, L., 71
Zhou, T., 386
Zhu, X., 405
Zung, J., 386
 Main Index

H-free process, 406 Convergence of Moments, 418

Gn,m , 3 Cores, 37
Gn,p , 3 Coupling Gn,m and Gn,p , 5
f -certifiable, 447 Critical Window, 39
k-Connectivity, 68 Dense Case
k-out Subgraphs of Large Graphs, 365 Greedy Algorithm, 125
Achlioptas Process, 47 Depth First Search, 94, 318
Anti-Ramsey Properties, 154 Diameter, 111
Azuma-Hoeffding Inequality, 125, 128, 371 Diameter of Sparse Random Graphs, 139
Directed k-in, `-out, 365
Balanced Graphs, 74
Disjoint Paths, 70
Balls in Boxes, 426
Dominance, 448
Bernstein’s Theorem, 434
Binomial Inequalities, 425 Edge Colored Random Graphs, 402
Binomial Random Graph, 3 Eigenvalues, 129
Bipartite Gk−out : Perfect matchings, 348 Concentration, 133
Algorithmic Proof, 351 Eigenvalues of Random Regular Graphs, 230
Existence Proof, 349 Embedding Gn,m in Gn,r , 217
Bohman-Frieze Process, 47 Entropy, 287
Erdős-Ko-Rado, 295
Canonical Labelling, 60
Evolution, 20
Chain Rule for Entropy, 455
Chebyshev’s Inequality, 12 First Cycle, 80
Chen-Stein Method, 422 First Moment Method, 12
Chernoff Inequality, 431 FKG Inequality, 428
Chromatic Index, 61 Galton-Watson Process, 258
Chromatic Number, 123 Games, 403
Dense case, 123 Geometric Preferential Attachment Graphs,
Sparse Case, 127 385
Chromatic Number of Random Regular Giant Component, 35
Graphs, 229 Graph Density, 72
Coarse Threshold, 19 Graph Homomorphism, 137
Complex Component, 46 Graph Isomorphism, 60
Conditional Entropy, 457 Graph Minors, 47
Configuration Model, 195 Graph Process, 20
Containers, 140 Graph Property, 5
Contiguity, 402 Graph Searching, 405

505
506 Main Index

Acquaintance Time, 406 Random Euclidean TSP, 399

Cops and Robbers, 405 Random Geometric Graphs, 241
Graph Cleaning, 405 Chromatic Number, 246
Graph Thickness, 48 Connectivity, 241
Graphs With Large Minimum Degree, 99 Hamiltonicity, 245
Greedy Matching Algorithm, 96 Random Knapsack Problem, 398
Hall’s Condition, 83 Random Lifts, 409
Hamilton Cycles, 89 Random Minimum Spanning Trees, 399
Hoeffding’s Theorem, 431 Random Recursive Tree, 306
Hypergeometric Distribution, 13 Random Regular Graphs, 210
Random Simplicial Complexes, 410
Janson’s Inequality, 436
Random Steiner Tree Problem, 400
Kleinberg’s Model, 382 Random Subgraphs of the n-cube, 410
Kullback-Liebler Distance, 456 Random Tournament, 267
Large Matchings, 109 Random Walks on Random Graphs, 410
Logic and Random Graphs, 230, 407 Cover Time, 411
Long Paths and Cycles, 94, 108 Mixing Time, 410
Randomly Coloring Random Graphs, 138
Markov Inequality, 12
Rank One Model, 193
Martingale Inequalities, 439
Resilience, 108, 155
Maximum Independent Set, 117
Dense Case, 117 Sampling Without Replacement, 435
Interpolation, 121 Scaling Window, 39
Sparse Case, 119 Second Moment Method, 14
Monotone Decreasing, 7 Sharp Threshold, 11
Monotone Increasing, 7 Shearer’s Lemma, 457
Small Subgraph Conditioning, 229
Nearest Neighbor Graphs, 364
SPA Model, 385
Orientability of Hypergraphs, 295 Spanning Subgraphs, 103, 108
Pósa’s Lemma, 90 Spanning Trees, 109
Parisi Conjecture, 393 Stable Matching, 411
Perfect Matchings, 82 Stirling’s Formula, 6
Bipartite Graphs, 83 Strictly Balanced, 76
Non-Bipartite Graphs, 85 Strong Components
Phase Transition, 37, 39, 47 Sub-Critical Region, 256
Planarity, 408 Super-Critical Region, 257
Plane-Oriented Recursive Trees, 316 Strong Connectivity, 263
Planted Objects, 408 Sub-Critical Phase, 20
Power Law Degree Distribution, 368 Super-Critical Phase, 33
Prüfer Code, 299 Susceptibility, 192
Preferential Attachment Graph, 367 Switching, 200
Concentration of Degree Sequence, 371 Talagrand’s Inequality, 445
Maximum Degree, 370 Threshold Function, 9
Pseudo-Graphs, 17 Triangle Free Subgraphs, 153
Rényi’s Lemma, 419 Turán Properties, 143
Rainbow Connection, 71, 230 Unicyclic Components, 28, 45
Rainbow Path, 70 Uniform Random Graph, 3
Ramsey Properties, 141 Universal graphs, 412
Random Apollonian Networks, 386
VC-Dimension, 295
Random Assignment Problem, 400
Multi-Dimensional version, 400 With High Probability (w.h.p.), 11
Random Dot Product Graph, 193 Wright Coefficients, 41