24.5.

Notes 519

log(r−1)
t ∗ = r(r−1)
(r−2)2
n. They showed that w.h.p. if t ≥ (1 + ε)t ∗ then the maximum
component size in Gt is O(log n) and conjectured that there is a phase transition
at t ∗ . This was confirmed by Cooper and Frieze [262] who gave an asymptotic
formula for the size of the largest component. Černy and Teixeira [210] show that
in a critical window around t ∗ the size of the largest component is of order n2/3 .
Cooper and Frieze also considered the vacant set for random walks on Gn,p , see
Exercise 24.4.1.
520 Chapter 24. Random Walk on Random Graphs
Chapter 25

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.

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 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 [407] showed that w.h.p. Gτ ∗ contains
a rainbow spanning tree. This is clearly best possible. Bal, Bennett, Frieze and
Pralat [61] 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 cy-
cles. Cooper and Frieze [253] 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 [402]. Bal and Frieze [63] show that if
m ≥ Kn log n and q = n and n is even there is a rainbow Hamilton cycle w.h.p.
Ferber [352] removed the requirement that n be even. Bal and Frieze also consid-
ered rainbow perfect matchings in k-uniform hypergraphs. Janson and Wormald
[512] 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 colored so that each color is
used r times, then w.h.p. there is a rainbow Hamilton cycle. Ferber, Kronenberg,
522 Chapter 25. Brief notes on uncovered topics

Mousset and Shikhelman [355] give results on packing rainbow structures such
as Hamilton cycles. Ferber, Nenadov and Peter 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 [250] 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 con-
nectivity 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
[230] for one of the earliest papers on the subject and books by Beck [83] and
Hefetz, Krivelevich, Stojaković and Szabó [476]. Most of the analyses have con-
sidered G = Kn and to make the problem interesting [230] 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 striking 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 [427]. 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
1/m (H)


optimal threshold bias b for this game is of order Θ n 2 , Bednarska and
Łuczak [85]. For sufficiently small constant c > 0, if b ≤ cn 1/m 2 (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 [819] began research on random boards i.e.
where G is a random graph. Ben-Shimon, Ferber, Hefetz and Krivelevich [99]
prove a hitting time result for the b = 1 Hamilton cycle game on the graph pro-
cess. 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 pos-
sible. Biased Hamiltonicity games on Gn,p were considered in Ferber, Glebov,
Krivelevich and Naor [353] where it was shown that for p  logn n , the threshold
np
bias bHAM satisfies bHAM ≈ log n w.h.p. The H-game where A wins if she can
 523

create a copy of some fixed graph H was first studied by Stojakovic and Szabo
[819] in the case of H is a clique on k vertices. This was strengthened by Müller
and Stojaković [700]. 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 [771] 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 [712] 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 [106] 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 neighbor-
hood of some uncolored vertex. The game chromatic 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 [81]. Bohman, Frieze and Sudakov [142] stud-
ied χg for dense random graphs and proved that for such graphs, χg is within a
constant factor of the chromatic number. Keusch and Steger [564] proved that this
factor is asymptotically equal to two. Frieze, Haber and Lavrov [395] extended
the results of [142] 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 [643] 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 [747] 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 [179].