Off-diagonal book Ramsey numbers

David Conlon∗ Jacob Fox† Yuval Wigderson‡

arXiv:2110.14483v2 [[Link]] 23 Nov 2022

Abstract
(k)
The book graph Bn consists of n copies of Kk+1 joined along a common Kk . In
(k) (k)
the prequel to this paper, we studied the diagonal Ramsey number r(Bn , Bn ). Here
(k) (k)
we consider the natural off-diagonal variant r(Bcn , Bn ) for fixed c ∈ (0, 1]. In this
more general setting, we show that an interesting dichotomy emerges: for very small c,
a simple k-partite construction dictates the Ramsey function and all nearly-extremal
colorings are close to being k-partite, while, for c bounded away from 0, random col-
orings of an appropriate density are asymptotically optimal and all nearly-extremal
colorings are quasirandom. Our investigations also open up a range of questions about
what happens for intermediate values of c.

1 Introduction
Given two graphs H1 and H2 , their Ramsey number r(H1 , H2 ) is the smallest positive integer
N such that every red/blue coloring of the edges of KN is guaranteed to contain a red copy
of H1 or a blue copy of H2 . One of the main open problems in Ramsey theory is to determine
the asymptotic order of r(K √ nn, Kn ). However, despite intense and longstanding interest, the
n
lower and upper bounds 2 ≤ r(Kn , Kn ) ≤ 4 for this problem have remained largely
unchanged since 1947 and 1935, respectively [11, 13].
Another major question in graph Ramsey theory, which has seen more progress, is to
determine the growth rate of the off-diagonal Ramsey number r(Ks , Kn ), where we think
of s as fixed and let n tend to infinity. The first non-trivial case is when s = 3, where it is
known that  2 
n
r(K3 , Kn ) = Θ ,
log n
∗
Department of Mathematics, California Institute of Technology, Pasadena, CA 91125, USA. Email:
dconlon@[Link]. Research supported by NSF Award DMS-2054452.
†
Department of Mathematics, Stanford University, Stanford, CA 94305, USA. Email:
jacobfox@[Link]. Research supported by a Packard Fellowship and by NSF Awards DMS-
1800053 and DMS-2154169.
‡
School of Mathematics, Tel Aviv University, Tel Aviv 69978, Israel. Email: yuvalwig@[Link].
Research supported by NSF GRFP Grant DGE-1656518, NSF-BSF Grant 20196, and by ERC Consolidator
Grants 863438 and 101044123.

1
with the upper bound due to Ajtai, Komlós, and Szemerédi [1] and the lower bound to
Kim [17]. Subsequent work of Shearer [23], Bohman–Keevash [3], and Fiz Pontiveros–
Griffiths–Morris [14] has led to a better understanding of the implicit constant, which is
now known up to a factor of 4 + o(1). However, the successes in estimating r(K3 , Kn ) have
not carried over to r(Ks , Kn ) for any other fixed s and a polynomial gap persists between the
upper and lower bounds for all s ≥ 4 (though see [18] for a promising approach to improving
the lower bound).
(k)
The book graph Bn is the graph obtained by gluing n copies of the clique Kk+1 along
(2)
a common Kk . The “book” terminology comes from the case k = 2, where Bn consists
of n triangles glued along a common edge. Continuing the analogy, each Kk+1 is called a
page of the book and the common Kk is called the spine. Ramsey numbers of books arise
naturally in the study of r(Kn , Kn ); indeed, Ramsey’s original proof [21] of the finiteness
of r(Kn , Kn ) proceeds inductively by establishing the finiteness of certain book Ramsey
numbers, while the Erdős–Szekeres bound [13] and its improvements [8, 22] are also best
interpreted through the language of books. Because of this, Ramsey numbers of books have
attracted a great deal of attention over the years, starting with papers of Erdős, Faudree,
Rousseau, and Schelp [12] and of Thomason [25]. Both of these papers prove bounds of the
(k) (k)
form 2k n − ok (n) ≤ r(Bn , Bn ) ≤ 4k n, where we think of k as fixed and n → ∞, with
Thomason conjecturing that the lower bound is closer to the truth. This was confirmed in
a recent breakthrough result of the first author [9], who proved that, for every fixed k,

r(Bn(k) , Bn(k) ) = 2k n + ok (n).

The original proof of this result relied heavily on an application of Szemerédi’s celebrated
regularity lemma, leading to rather poor control on the error term. In the prequel to this
paper [10], we gave two alternative proofs of this result, one a simplified version of the first
author’s original proof and the other a proof which avoids the use of the full regularity
lemma, allowing us to gain significantly better control over the error term (for a discussion
of how further improvements might ultimately impinge on the estimation of r(Kn , Kn ), we
refer the reader to [10]). We also proved a stability result, saying that extremal colorings for
this Ramsey problem are quasirandom.
In this paper, we study a natural off-diagonal generalization of this problem. Specifically,
we fix some k ∈ N and some c ∈ (0, 1] and we wish to understand the asymptotics of the
(k) (k)
Ramsey number r(B⌊cn⌋ , Bn ) as n → ∞. Note that for c = 1 this is precisely the question
(k) (k)
considered above. Henceforth, we omit the floor signs and write Bcn instead of B⌊cn⌋ .
(k) (k)
Our results reveal that the behavior of the function r(Bcn , Bn ) varies greatly as c moves
from 0 to 1. As we shall see, for c sufficiently small, the behavior of this Ramsey number is
determined by a simple block construction, while, for c sufficiently far from 0, its behavior is
determined by a random coloring. There is also an intermediate range of c where our results
say nothing, but where several interesting questions arise. We will say more about this in
the concluding remarks.
To describe our results in detail, we begin by observing that for any positive integers k,

2
m, and n with m ≤ n, we have
(k)
r(Bm , Bn(k) ) ≥ k(n + k − 1) + 1. (1)

Indeed, let N = k(n + k − 1). We partition the vertices of KN into k blocks, each of size
n + k − 1. We color all edges inside a block blue and all edges between blocks red. Then any
(k) (k)
blue Bn must appear inside a block, which it cannot, since Bn has n + k vertices. On the
other hand, since the red graph is k-partite, it does not contain any red Kk+1 and so cannot
(k)
contain a red Bm .
This simple inequality is a special case of a more general lower bound, usually attributed
to Chvátal and Harary [7], that r(H1 , H2 ) ≥ (χ(H1 ) − 1)(|V (H2 )| − 1) + 1 provided H2 is
connected. In general, this lower bound is far from optimal,1 but it is tight for certain sparse
graphs. The study of when it is tight goes under the name of Ramsey goodness, a term
introduced by Burr and Erdős [5] in their first systematic investigation of the concept. One
of the central results in the field of Ramsey goodness is due to Nikiforov and Rousseau [19],
who proved an extremely general theorem about when this lower bound is tight. As a very
special case of their theorem, one has the following result; see also [15] for a new proof with
better quantitative bounds.
Theorem 1.1 (Nikiforov–Rousseau [20, Theorem 2.12]). For every k ≥ 2, there exists some
c0 ∈ (0, 1) such that, for any 0 < c ≤ c0 and n sufficiently large,
(k)
r(Bcn , Bn(k) ) = k(n + k − 1) + 1.

Moreover, Nikiforov and Rousseau’s proof shows that the unique coloring on k(n + k − 1)
(k) (k)
vertices with no red Bcn and no blue Bn is the coloring we described, where the red graph
is a balanced complete k-partite graph (meaning that all the parts have orders as equal as
possible). By adapting their proof, we are able to prove a corresponding structural stability
result, which says that any coloring on N = (k + o(1))n vertices is either “close” to being
balanced complete k-partite in red or contains monochromatic books with substantially
more pages than what is guaranteed by Theorem 1.1. Note that if N is sufficiently large
and congruent to 1 modulo k, then Theorem 1.1 says that any red/blue coloring of E(KN )
contains a red Kk with at least kc (N − 1) − c(k − 1) extensions to a red Kk+1 or a blue Kk
with at least k1 (N − 1) − (k − 1) extensions to a blue Kk+1 .
Theorem 1.2. For every k ≥ 2 and every θ > 0, there exist c, γ ∈ (0, 1) such that the
following holds for any sufficiently large N and any red/blue coloring of E(KN ). Either one
can recolor at most θN 2 edges to turn the red graph into a balanced complete k-partite graph
or else the coloring contains one of the following:
• at least γN k red Kk , each with at least ( kc + γ)N extensions to a red Kk+1 , or

• at least γN k blue Kk , each with at least ( k1 + γ)N extensions to a blue Kk+1.

1
For example, for H1 = H2 = Kn , it gives a lower bound of r(Kn , Kn ) = Ω(n2 ), whereas the truth is
Θ(n)
2 .

3
 Informally, this theorem says that either the coloring is close to complete k-partite in
red or else a constant fraction of the k-tuples induce a clique that forms the spine of a
monochromatic book with at least γN more pages than what is guaranteed by the Ramsey
bound alone.
However, once c is sufficiently far from 0, the deterministic construction that yields (1)
stops being optimal. Indeed, as in the diagonal case, we can get another lower bound on
(k) (k)
r(Bcn , Bn ) by considering random colorings. More precisely, let us fix k ∈ N and c ∈ (0, 1]
and define
1
∈ 12 , 1 .


p = 1/k
c +1
−k
We set N = (p − o(1))n and independently color every edge of KN blue with probability
p and red with probability 1 − p. Given a blue Kk in this coloring, the expected number
of extensions to a blue Kk+1 is pk (N − k) = n − o(n). Similarly, the expected number of
extensions of a red Kk to a red Kk+1 is (1 − p)k (N − k) = ((1 − p)/p)k n − o(n) = cn − o(n),
by our choice of p. A standard application of the Chernoff bound and the union bound then
(k) (k)
implies that w.h.p.2 this coloring contains no blue Bn and no red Bcn , assuming the o(n)
terms are chosen appropriately. This implies that for any k ∈ N and any c ∈ (0, 1],
(k)

k
r(Bcn , Bn(k) ) ≥ c1/k + 1 n − ok (n),
(k) (k)
while the lower bound in (1) is that r(Bcn , Bn ) ≥ (k + o(1))n. If c > ((1 + o(1)) logk k )k , then
the quantity (c1/k + 1)k is larger than k + o(1), where the logarithm is to base e. Thus, once
c is sufficiently far from 0, the bound in (1) is smaller than the random bound.
Our next main result shows that the random bound actually becomes asymptotically
tight at this point.
Theorem 1.3. For every k ≥ 2, there exists some c1 = c1 (k) ∈ (0, 1] such that, for any fixed
c1 ≤ c ≤ 1,
(k)

k
r(Bcn , Bn(k) ) = c1/k + 1 n + ok (n).
Moreover, one may take c1 (k) = ((1 + o(1)) logk k )k .
Our third main result is a corresponding structural stability theorem, which says that all
near-extremal Ramsey colorings (i.e., colorings on roughly (c1/k + 1)k n vertices) must either
contain a monochromatic book substantially larger than what is guaranteed by Theorem 1.3
or be “random-like”. The latter possibility is captured by the notion of quasirandomness,
introduced by Chung, Graham, and Wilson [6]. For parameters p, θ ∈ (0, 1), a red/blue
coloring of E(KN ) is said to be (p, θ)-quasirandom if, for every pair of disjoint sets X, Y ⊆
V (KN ), we have that
|eB (X, Y ) − p|X||Y || ≤ θN 2 ,
where eB (X, Y ) denotes the number of blue edges between X and Y . Note that since
the colors are complementary, this is equivalent to the analogous condition requiring that
2
As usual, we say that an event E happens with high probability (w.h.p.) if Pr(E) → 1 as n → ∞, where
the implicit parameter n will be clear from context.

4
eR (X, Y ) is within θN 2 of (1−p)|X||Y |. In their seminal paper, Chung, Graham, and Wilson,
building on previous results of Thomason [25], showed that this condition is essentially
equivalent to a large number of other conditions, all of which encapsulate some intuitive idea
of what it means for a coloring to be similar to a random coloring with blue density p. With
this notion in hand, we can state our structural stability result.
Theorem 1.4. For every p ∈ [ 21 , 1), there exists some k0 ∈ N such that the following holds
for every k ≥ k0 . For every θ > 0, there exists some γ > 0 such that if a red/blue coloring
of E(KN ) is not (p, θ)-quasirandom, then it contains one of the following:
• at least γN k red Kk , each with at least ((1 − p)k + γ)N extensions to a red Kk+1 , or
• at least γN k blue Kk , each with at least (pk + γ)N extensions to a blue Kk+1 .
Remark. As stated, this theorem does not mention the “off-diagonalness” parameter c from
the previous theorem. But c can easily be recovered as ((1 − p)/p)k and the theorem can
then be restated to be about blue books with slightly more than n pages or red books with
slightly more than cn pages. However, since p is what matters while c plays no real role in
the argument, we instead choose to use this language and avoid c entirely.
In Theorem 5.7, we also prove a converse to Theorem 1.4, which implies that for p fixed
and k sufficiently large in terms of p, a coloring of KN (or, more accurately, a sequence of
colorings with N tending to infinity) is (p, o(1))-quasirandom if and only if all but o(N k )
red Kk have at most ((1 − p)k + o(1))N extensions to a red Kk+1 and all but o(N k ) blue
Kk have at most (pk + o(1))N extensions to a blue Kk+1 . Thus, we derive a new equivalent
formulation for (p, o(1))-quasirandomness.
The rest of the paper is organized as follows. In Section 2, we quote (mostly without
proof) a number of key results that we will use repeatedly. In Section 3, we establish
Theorem 1.2, the stability result for small c. We prove Theorem 1.3, that the random bound
is asymptotically tight once c is not too small, in Section 4 and Theorem 1.4, that extremal
colorings are quasirandom in this range, in Section 5. We end with some concluding remarks
and open problems.

1.1 Notation and Terminology

If X and Y are two vertex subsets of a graph, let e(X, Y ) denote the number of pairs in
X × Y that are edges. We will often normalize this and consider the edge density
e(X, Y )
d(X, Y ) = .
|X||Y |
If we consider a red/blue coloring of the edges of a graph, then eB (X, Y ) and eR (X, Y ) will
denote the number of pairs in X × Y that are blue and red edges, respectively. Similarly, dB
and dR will denote the blue and red edge densities, respectively. Finally, for a vertex v and
a set Y , we will sometimes abuse notation and write d(v, Y ) for d({v}, Y ) and similarly for
dB and dR .

5
 An equitable partition of a graph G is a partition of the vertex set V (G) = V1 ⊔ · · · ⊔ Vm
with ||Vi | − |Vj || ≤ 1 for all 1 ≤ i, j ≤ m. A pair of vertex subsets (X, Y ) is said to be
ε-regular if, for every X ′ ⊆ X, Y ′ ⊆ Y with |X ′ | ≥ ε|X|, |Y ′ | ≥ ε|Y |, we have

|d(X, Y ) − d(X ′ , Y ′ )| ≤ ε.

Note that we do not require X and Y to be disjoint. In particular, we say that a single
vertex subset X is ε-regular if the pair (X, X) is ε-regular. We will often need a simple
fact, known as the hereditary property of regularity, which asserts that for any 0 < α ≤ 1, if
(X, Y ) is ε-regular and X ′ ⊆ X, Y ′ ⊆ Y satisfy |X ′ | ≥ α|X|, |Y ′ | ≥ α|Y |, then (X ′ , Y ′ ) is
(max{ε/α, 2ε})-regular.
For real numbers a, b, we denote by a ± b any quantity in the interval [a − b, a + b].
All logarithms are base e unless otherwise specified. For the sake of clarity of presentation,
we systematically omit floor and ceiling signs whenever they are not crucial. In this vein,
whenever we have an equitable partition of a vertex set, we will always assume that all of
the parts have exactly the same size, rather than being off by at most one. Because the
number of vertices in our graphs will always be “sufficiently large”, this has no effect on our
final results.

2 Results from earlier work

In this section, we collect some useful tools for the study of book Ramsey numbers, all
of which have appeared in previous works. We begin with several results from the theory
around Szemerédi’s regularity lemma and then quote two simple analytic inequalities.

2.1 Tools from regularity

We begin with a strengthened form of Szemerédi’s regularity lemma taken from our first
paper [10, Lemma 2.1].

Lemma 2.1. For every ε > 0 and M0 ∈ N, there is some M = M(ε, M0 ) ≥ M0 such that, for
every graph G with at least M0 vertices, there is an equitable partition V (G) = V1 ⊔ · · · ⊔ Vm
into M0 ≤ m ≤ M parts such that the following hold:

1. Each part Vi is ε-regular and

2. For every 1 ≤ i ≤ m, there are at most εm values 1 ≤ j ≤ m such that the pair (Vi , Vj )
is not ε-regular.

To complement the regularity lemma, we will also need a standard counting lemma (see,
e.g., [26, Theorems 2.6.2 and 4.5.1]).

6
Lemma 2.2. Suppose that V1 , . . . , Vk are (not necessarily distinct) subsets of a graph G such
that all pairs (Vi , Vj ) are ε-regular. Then the number of labeled copies of Kk whose i-th vertex
is in Vi for all i is
 ! Yk
Y k
d(Vi , Vj ) ± ε |Vi |.
1≤i<j≤k
2 i=1

We will frequently use the following consequence of the counting lemma, proved in [10,
Corollary 2.6], designed to count monochromatic extensions of cliques and thus estimate the
size of monochromatic books.

Lemma 2.3. Fix k ≥ 2 and let η, α ∈ (0, 1) be parameters with η ≤ α3 /k 2 . Suppose

U1 , . . . , Uk are (not necessarily
Q distinct) vertex sets in a graph G and suppose that all pairs
(Ui , Uj ) are η-regular with 1≤i<j≤k d(Ui , Uj ) ≥ α. Let Q be a uniformly random copy of Kk
with one vertex in each Ui , for 1 ≤ i ≤ k, and say that a vertex u extends Q if u is adjacent
to every vertex of Q. Then, for any u ∈ V (G),
k
Y
Pr(u extends Q) ≥ d(u, Ui ) − 4α.
i=1

The final result in this subsection is actually a simple consequence of Markov’s inequality
and so does not require any regularity tools to prove. Nonetheless, we will always use it in
conjunction with Lemmas 2.2 and 2.3, which is why we include it here. Both the statement
and proof are very similar to [10, Lemma 5.2].

Lemma 2.4. Let κ, ξ ∈ (0, 1), let 0 < ν < ξ, and suppose that Q is a set of at least κN k
copies of Kk in an N-vertex graph. Suppose that a uniformly random Q ∈ Q has at least ξN
extensions to a Kk+1 in expectation. Then the graph contains at least (ξ − ν)κN k copies of
Kk , each with at least νN extensions.

Proof. Let X be the random variable counting the number of extensions of a random Q ∈ Q
and let Y = N − X. Then Y is a non-negative random variable with E[Y ] = N − E[X] ≤
(1 − ξ)N. By Markov’s inequality,

E[Y ] (1 − ξ)N 1−ξ

Pr(X ≤ νN) = Pr (Y ≥ (1 − ν)N) ≤ ≤ = .
(1 − ν)N (1 − ν)N 1−ν

Thus,
1−ξ ξ−ν
Pr(X ≥ νN) ≥ 1 − = ≥ ξ − ν,
1−ν 1−ν
which implies that the number of Q ∈ Q with at least νN extensions is at least (ξ − ν)|Q| ≥
(ξ − ν)κN k , as desired.

7
2.2 Analytic inequalities
The following lemma is a multiplicative form of Jensen’s inequality and is a simple conse-
quence of the standard version. For a proof, see [10, Lemma A.1].

Lemma 2.5 (Multiplicative Jensen inequality). Suppose 0 < a < b are real numbers and
x1 , . . . , xk ∈ (a, b). Let f : (a, b) → R be a function such that y 7→ f (ey ) is strictly
Qk convex on
k k
the interval (log a, log b). Then, for any z ∈ (a , b ), subject to the constraint i=1 xi = z,
k
1X
f (xi )
k i=1

is minimized when all the xi are equal (and thus equal to z 1/k ).

The following theorem is the well-known “defect” or “stability” version of Jensen’s in-
equality. For a proof, see [24, Problem 6.5].

Theorem 2.6 (Hölder’s Defect Formula). Suppose ϕ : [a, b] → R is a twice-differentiable

function with ϕ′′ (y) ≥ m > 0 for all y ∈ (a, b). For any y1 , . . . , yk ∈ [a, b], let
k k
1X 2 1X
µ= yi and σ = (yi − µ)2
k i=1 k i=1

be the empirical mean and variance of {y1 , . . . , yk }. Then

k
1X mσ 2
ϕ(yi ) − ϕ(µ) ≥ .
k i=1 2

3 The k-partite regime

In this section, we analyze what happens when c is very small. Recall, from the introduction,
(k) (k)
that a simple k-partite construction yields a lower bound for r(Bcn , Bn ) and, by a result
of Nikiforov and Rousseau [20], this construction is tight for c sufficiently small.

Theorem 1.1 (Nikiforov–Rousseau, [20, Theorem 2.12]). For every k ≥ 2, there exists some
c0 ∈ (0, 1) such that, for any 0 < c ≤ c0 and n sufficiently large,
(k)
r(Bcn , Bn(k) ) = k(n + k − 1) + 1.

Our aim here is to adapt the methods of [20] to prove a stability version of this theorem,
our Theorem 1.2. We first make the following definition.

Definition 3.1. For c, γ > 0, we say that a red/blue coloring of E(KN ) contains (c, γ)-many
books if it contains

8
 • at least γN k red Kk , each with at least ( kc + γ)N extensions to a red Kk+1 , or

• at least γN k blue Kk , each with at least ( k1 + γ)N extensions to a blue Kk+1 .

With this definition in place, we may restate Theorem 1.2 as follows.

Theorem 1.2’. For every k ≥ 2 and every θ > 0, there exist c, γ ∈ (0, 1) such that the
following holds. If a red/blue coloring of E(KN ) does not have (c, γ)-many books, then one
can recolor at most θN 2 edges to turn the red graph into a balanced complete k-partite graph.

As well as referring to Section 2, we will need the following classical result of Andrásfai,
Erdős, and Sós [2] (see also [4] for a simpler proof).
Theorem 3.2 (Andrásfai–Erdős–Sós [2]). For every k ≥ 2, there exists ρ > 0 such that if
G is a Kk+1-free graph on m vertices with minimum degree greater than (1 − k1 − ρ)m, then
G is k-partite. Moreover, one may take ρ = 1/(3k 2 − k).

This is a stability version of Turán’s theorem. Indeed, Turán’s theorem implies that if a
graph on m vertices has minimum degree at least (1 − k1 )m, then it contains a copy of Kk+1 ,
while the Andrásfai–Erdős–Sós theorem says that as long as the minimum degree is not too
far below (1 − k1 )m, every such graph must be k-partite.
Before proceeding to the technical details, let us briefly sketch the proof of Theorem 1.2.
We are given a red/blue coloring of E(KN ) and we wish to prove that either the coloring
is close to complete k-partite in red or it contains (c, γ)-many books for some c, γ > 0.
We begin by applying Lemma 2.1 to the red graph of the coloring to obtain an equitable
partition V (KN ) = V1 ⊔ · · · ⊔ Vm , where each part Vi and most pairs (Vi , Vj ) are η-regular for
some small η > 0. We now wish to improve our understanding of the coloring with respect
to this partition.
First, we show that all the parts Vi must have very low internal red density. Indeed, if
some part Vi has dR (Vi ) ≥ δ, for some fixed δ > 0, then the counting lemma, Lemma 2.2,
implies that Vi contains many red Kk+1 . By a simple averaging argument, this implies that
some k-tuple of vertices in Vi lies in many red Kk+1 , yielding a red book with kc N pages. In
fact, by using Lemma 2.4 in place of the averaging argument, we find (c, γ)-many books if
dR (Vi ) ≥ δ, so we may assume that dR (Vi ) < δ for all i.
We next build a reduced graph G with vertex set v1 , . . . , vm , where we make vi vj an edge
if and only if (Vi , Vj ) is η-regular and dR (Vi , Vj ) ≥ δ. We claim that every vertex of G has
degree at least (1 − k1 − σ)m for some small σ > 0. Indeed, if some vertex vi of G has degree
smaller than this, then we find that Vi has very high blue density to roughly ( k1 + σ)m of
the remaining parts Vj . Since Vi also has very high internal blue density, we can use this to
find many blue books with spines in Vi and ( k1 + γ)N pages for some 0 < γ < σ. This again
yields (c, γ)-many books in the coloring.
So we may assume that the graph G has high minimum degree. By applying Theorem 3.2,
we find that either G is k-partite or it contains a copy of Kk+1 . In the former case, we can
show that the coloring itself is close to k-partite in red. In the latter case, this Kk+1 yields
k + 1 parts, say V1 , . . . , Vk+1, such that all pairs are η-regular and have red density at least

9
δ. By another application of the counting lemma and an averaging argument, we can then
show that this structure again yields (c, γ)-many red books.
We now turn to the details of the proof. We will need the following fact about bipartite
graphs, which is a simple consequence of a double-counting technique first introduced by
Kővári, Sós, and Turán [16].
Lemma 3.3. Let k ≥ 2 and d ∈ (0, 1) and let ζ = (d/4)k . Let H be a bipartite graph with
parts A, B, where |B| ≥ 2k/d, and suppose that H has at least
d|A||B| edges. Let H be a
k-uniform hypergraph with vertex set B and at least (1 − ζ) |B|
k
edges. Then there are at
|B|


least ζ k edges of H such that the vertices of each such edge have at least ζ|A| common
neighbors in A.
Proof. For a k-tuple Q ∈ Bk , let ext(Q) denote the number of common neighbors of Q in

A. We double-count the number of stars K1,k in H whose central vertex is in A to find that
1 P
a∈A deg(a)
X X deg(a)  
d|B|

|A|
ext(Q) = ≥ |A| ≥ |A| ,
a∈A
k k k
Q∈( k )
B

where the first inequality follows from convexity. If we split the left-hand side into a sum
over tuples Q which are non-edges of H, a sum over tuples Q that are edges of H with fewer
than ζ|A| extensions, and the remainder, we find that
 
d|B| X X X
|A| ≤ ext(Q) + ext(Q) + ext(Q)
k
Q∈E(H)
/ Q∈E(H) Q∈E(H)
ext(Q)<ζ|A| ext(Q)≥ζ|A|
   
|B| |B|
≤ ζ|A| + ζ|A| + |A||{Q ∈ E(H) : ext(Q) ≥ ζ|A|}|.
k k
d|B|


Therefore, the number of edges of H with at least ζ|A| common neighbors is at least k
−
2ζ |B|


k
. We note that
d|B|  k


d|B| d|B| − 1 d|B| − (k − 1) d
k
|B|
= · ··· ≥ = 2k ζ,
|B| |B| − 1 |B| − (k − 1) 2


k

where we used our assumption that |B| ≥ 2k/d. Thus, the number of edges of H with at
least ζ|A| common neighbors in A is at least
       
d|B| |B| k |B| |B|
− 2ζ ≥ (2 ζ − 2ζ) ≥ζ .
k k k k
With these preliminaries in place, we are now ready to prove Theorem 1.2.
Proof of Theorem 1.2. Fix some k ≥ 2, θ ∈ (0, 1), and a red/blue coloring of E(KN ). Let
2
σ = (θ/(12k 4))k , M0 = k, δ = σ 2 , and η = δ k . Let M = M(η, M0 ) be the parameter

10
 2 2
from Lemma 2.1 and let c = δ k /M 4 and γ = δ 2k /M 2k . We apply Lemma 2.1 to the
red graph in our coloring with parameters M0 and η. This yields an equitable partition
V (KN ) = V1 ⊔ · · · ⊔ Vm with M0 ≤ m ≤ M such that each part Vi is η-regular in red and,
for each i, there are at most ηm values of j 6= i for which (Vi , Vj ) is not η-regular in red.
First suppose that some part, say V1 , has internal red density at least δ. By the counting
(δ ( 2 ) − η k+1
k+1
1
)|V1 |k+1 red Kk+1 .


lemma, Lemma 2.2, we see that V1 contains at least (k+1)! 2
Since each red Kk+1 contains exactly k + 1 red Kk , this implies that an average k-tuple of
vertices in V1 lies in at least

(δ ( ) − η k+1
)|V |k+1 
k+1
k+1  
(k+1)!
2
2 1
( k+1
) k+1
|V1 |
≥ δ 2 −η |V1 | =: ξM|V1 |
2


k

red Kk+1 . That is, if we pick a uniformly random k-tuple of vertices from V1 , then the
number of red Kk+1 containing it is at least ξM|V1 |. If we also define κ = (δ (2) −
k
expected
η k2 )/(k!M k ), then Lemma 2.2 implies that V1 contains at least κN k red Kk , with an average

one having at least ξN extensions to a red Kk+1 , where we use the fact that |V1 | ≥ N/M
since the partition is equitable and has m ≤ M parts. If we now set ν = ξ/2 and apply
Lemma 2.4, we conclude that V1 contains at least (ξκ/2)N k red Kk , each with at least
(ξ/2)N extensions to a red Kk+1 . By our choice of parameters,
2
δk
  
1 ( k+1
) k+1
ξ= δ 2 −η ≥
M 2 2M
2
and, therefore, ξ/2 ≥ c/k + γ. Similarly, κ ≥ δ k /M k and, therefore, ξκ/2 ≥ γ. Thus, we
find that in this case the coloring contains (c, γ)-many books.
Therefore, we may assume that all Vi have dR (Vi ) < δ. We build a reduced graph G with
vertex set v1 , . . . , vm and declare {vi , vj } ∈ E(G) if (Vi , Vj ) is η-regular and dR (Vi , Vj ) ≥ δ.
Suppose that some vertex of G, say v1 , has degree less than (1 − k1 − σ)m. Since at most
ηm non-neighbors of v1 can come from irregular pairs, we find that dB (V1 , Vi ) ≥ 1 − δ for at
least ( k1 + σ − η)m choices of i ∈ [m]. Let I ⊆ [m] be the set of such i. Since dB (V1 ) ≥ 1 − δ
and δ ≤ 1/k 2 , we see that, for α = kδ/4,
 
( k
) ( k
) k
dB (V1 ) 2 ≥ (1 − δ) 2 ≥ 1 − δ ≥ α.
2

Moreover, we have that η < α3 /k 2 by our choice of η. Therefore, we may apply Lemma 2.3,
which implies that if Q is a randomly chosen blue Kk in V1 and u is some vertex inSKN , then
Pr(u extends Q) ≥ d(u, V1)k − 4α. In particular, if we sum this up over all u ∈ i∈I Vi , we
find that the expected number of blue extensions of Q is at least
 
XX
k N k 1
+ σ − η (1 − δ)k − 4α N,



(d(u, V1) − 4α) ≥ |I| (1 − δ) − 4α ≥
i∈I u∈V
m k
i

11
where the first inequality follows from the convexity of the function x 7→ xk . Using η < σ/2,
we have that
   
1 k

1 σ 1 σ
+ σ − η (1 − δ) − 4α ≥ + (1 − 2kδ) ≥ + − 2kδ,
k k 2 k 2

where the last step follows from the bound 1/k + σ ≤ 1/k + 1/k ≤ 1. By our choice of
δ = σ 2 ≤ σ/(8k), we see that the expected number of blue extensions of Q is at least ( k1 + σ4 )N.
Moreover, by Lemma 2.2, the number of choices for Q is at least 1 ((1−δ)(2) −η k )(N/M)k ≥
k

k! 2
κN k . Therefore, if we apply Lemma 2.4 with parameters κ, ξ = k1 + σ4 , and ν = k1 + γ, then
we find that the coloring contains (c, γ)-many books.
Therefore, we may assume that every vertex in G has degree greater than (1 − k1 − σ)m,
so, by Theorem 3.2 and the fact that σ < 1/(3k 2 − k), we see that either G contains a Kk+1
or G is k-partite. Assume first that there is a Kk+1 in G. By relabeling the vertices, we may
assume that v1 , . . . , vk+1 form a clique. By the counting lemma, Lemma 2.2, we have that
V1 , . . . , Vk span at least (δ (2) − η k2 )(N/m)k ≥ κN k red Kk and V1 , . . . , Vk+1 span at least
k

(δ ( 2 ) − η k+1 )(N/m)k+1 red K . Every such red K

k+1

2 k+1 contains exactly one red K with
k+1 k
one vertex in each of V1 , . . . , Vk , so an average k-tuple (v1 , . . . , vk ) ∈ V1 × · · · × Vk lies in at
least
(δ ( 2 ) − η k+1
k+1
)(N/m)k+1

  
2 ( k+1
) k + 1 N
k
≥ δ 2 −η = ξN.
(N/m) 2 M
Thus, we have a set of at least κN k red Kk with at least ξN extensions on average and so,
applying Lemma 2.4 as before, our coloring has (c, γ)-many books.
Thus, we may assume that G is k-partite. Let this k-partition of V (G) be A1 ⊔ · · · ⊔ Ak .
Note that |Aℓ | ≤ ( k1 + σ)m for every ℓ, since the minimum degree of G is at least (1 − k1 −σ)m
1
and each Aℓ is an independent P set in G. 1This in turn implies that |Aℓ | ≥ ( k − kσ)m for
every ℓ, since |Aℓ | = m − ℓ′ 6=ℓ |Aℓ′ | ≥ ( k − kσ)m. We lift this S partition to a partition
of the vertices of KN into k parts X1 , . . . , Xk by letting Xℓ = vi ∈Aℓ Vi , noting that our
observations above imply that |Xℓ | = ( k1 ± kσ)N for all ℓ. We claim that each Xℓ contains
at most 3δ2 |X2ℓ | red edges. Indeed, observe that if vi , vj are two (not necessarily distinct)

vertices of G that are in the same part Aℓ , then they must be non-adjacent in G. This means
that either (Vi , Vj ) is an irregular pair or dR (Vi , Vj ) < δ. There are at most ηm2 irregular
pairs, so the irregular pairs can contribute at most ηN 2 ≤ 4k 2 η|Xℓ |2 ≤ 10k 2 η |X2ℓ | red edges

inside Xℓ , where we used that |Xℓ | ≥ ( k1 − kσ)N ≥ N/(2k). All other pairs of parts inside
each
Xℓ have red density at
most δ, so the total number of red edges inside Xℓ is at most
δ |X2ℓ | + 10k 2 η |X2ℓ | ≤ 3δ2 |X2ℓ | , since η ≤ δ/(20k 2). This implies that the number of ordered

pairs of (not necessarily distinct) vertices in Xℓ which do not form a blue edge is at most
2δ|Xℓ |2 .
This already implies that the red graph can be made k-partite by recoloring at most
2δN 2 edges, so it only remains to show that by recoloring a small number of additional
edges, we can make the red graph balanced complete k-partite. For this, suppose that
dB (X1 , X2 ) ≥ θ/k 2 . If we sample (with repetition) a random k-tuple Q of vertices from

12
X2 , then the probability that it does not form a blue clique is at most k2 · 2δ ≤ k 2 δ,

since each random pair of vertices does not span a blue edge with probability at most
2δ. Moreover, the expected number of common blue neighbors of Q inside X2 is at least
(1 − 2δ)k |X2 | − k ≥ (1 − 2kδ)|X2|, by convexity. By applying Markov’s √ inequality as in
the proof of Lemma 2.4, the probability
√ that Q has fewer than (1 − δ)|X2 | common blue
neighbors in √ X2 is at most 2k δ. Therefore, the probability that Q2 is a blue √ clique with√at
least (1
√ − δ)|X 2 | common blue neighbors in X 2 is at least 1 − k δ − 2k δ ≥ 1 − 3k δ,
since δ ≤ 1/k. Let H be the k-uniform √ hypergraph with vertex set X2 whose edges are
all blue Kk in√X2 with at least (1 − δ)|X 2 | common blue neighbors in X2 . Then H has at
|X2 |


least (1 − 3k δ) k edges.
We now apply Lemma 3.3 to the hypergraph H and to the bipartite graph of blue
edges between X1 and X2 , which has edge √ density d ≥ θ/k 2 by assumption. We have
that (d/4)k ≥ (θ/(4k 2 ))k ≥ 3kσ = 3k δ by our choice
of σ and δ, so we may indeed
apply Lemma 3.3 to conclude that at least (θ/(4k 2 ))k |Xk2 | of the edges of H have at least
(θ/(4k 2 ))k |X1 | common blue neighbors in X1 . This yields at least
 k    k  k
θ |X2 | θ|X2 | θ
≥ ≥ N k ≥ γN k
4k 2 k 4k 3 8k 4

blue Kk , each of which has at least

k k ! 
√ √
  
θ θ 1
(1 − δ)|X2 | + |X1 | ≥ 1− δ+ − kσ N
4k 2 4k 2 k
k !
√

1 θ
≥ + − δ − 2kσ N
k 4k 3
 k !
1 θ
≥ + − 3kσ N
k 4k 3
 
1
≥ +γ N
k

extensions to a blue Kk+1, where in both computations

√ we used the fact that |X1 |, |X2 | ≥
1
( k − kσ)N ≥ N/(2k), as well as our choices of δ = σ = (θ/(12k 4 ))k . Thus, in this case, we
have again found (c, γ)-many books, a contradiction.
Hence, we may assume that dB (X1 , X2 ) < θ/k 2 . By the same argument, all the blue
densities between different parts Xℓ can be assumed to be at most θ/k 2 . Since we have
already argued

that the red density inside each part is at most 2δ, we see that, by recoloring
at most ( k2 θ/k 2 + 2kδ)N 2 edges, we can make the red graph complete k-partite. Finally,
we recall that each part Xℓ has size |Xℓ | = ( k1 ± kσ)N. Therefore, by moving at most
k 2 σN arbitrary vertices into another part, we see that we can make our partition equitable.
We then recolor the edges incident with any moved vertex to obtain a balanced complete
k-partite red graph. Doing so entails recoloring at most k 2 σN 2 additional edges. Thus, in

13
total, we recolor at most
    
k θ 2 2 θ
2
+ 2kδ + k σ N ≤ + 3k σ ≤ θN 2
2
2 k 2

edges, where we used that δ ≤ σ and σ ≤ (θ/(12k 4 ))k ≤ θ/(6k 2 ).

4 An upper bound matching the random bound

In this section, we prove Theorem 1.3, which says that when c is not too small, the random
(k) (k)
lower bound for r(Bcn , Bn ) is asymptotically tight. To prove this theorem, we will mimic
our simplified proof of the diagonal result from [10, Section 3], though it needs to be adapted
to the off-diagonal setting in several ways. Before proceeding with the details, we sketch the
proof at a high level, indicating which parts require new ideas beyond those already present
in [10].
A key notion used in the proof is that of a red-blocked configuration. Informally, a red-
blocked configuration consists of k disjoint vertex sets such that each set and all pairs are
η-regular for some small η, every set has red density at least δ for some small δ, and every pair
has blue density at least δ. A blue-blocked configuration is defined similarly, except with the
roles of red and blue interchanged. Like the good and great configurations defined in [10], we
care about such configurations because their existence automatically implies the existence
of large monochromatic books. The precise statement is given in Lemma 4.3, but, roughly,
it says that if we have a red/blue coloring of the complete graph on (c1/k + 1)k n + o(n)
vertices which contains a red-blocked or a blue-blocked configuration and k is sufficiently
(k) (k)
large with respect to c, then the coloring contains a red Bcn or a blue Bn . This is the key
lemma which underlies the entire proof. Its proof is similar to that of [10, Lemma 3.3], but
requires a few modifications. First, the analytic inequality which yields the result is more
complicated in the off-diagonal setting and this is where the (necessary) assumption that k
is large with respect to c comes from. Second, the averaging arguments used in the proof
of Lemma 4.3 require a little more care than those used in the proof of [10, Lemma 3.3],
because we must take (p, 1−p)-weighted averages here. Finally, though in principle one needs
separate arguments to deal with red-blocked configurations and blue-blocked configurations,
it turns out that the same proof works for both cases, simply by interchanging the roles of
red and blue and of p and 1 − p.
The remainder of the proof now comes down to finding a red-blocked or blue-blocked
configuration or else finding a large monochromatic book directly. To do this, we begin by
applying Lemma 2.1 to the red graph of the coloring, obtaining a regular equitable partition
V (KN ) = V1 ⊔ · · · ⊔ Vm . Call a part red if it contains more red edges than blue edges and
blue otherwise. We assume for now that at least pm of the parts are blue; the case where
at least (1 − p)m of the parts are red runs similarly. We build a reduced graph G whose
vertices are in bijection with the blue parts and where edges represent pairs of parts that
are regular and have red density at least δ for some small δ > 0. By defining G in this way,

14
we see that a Kk in G corresponds to a blue-blocked configuration in the original coloring,
so it suffices to find a Kk in G.
To do this, we first show that every vertex in G must have degree at least roughly
(1 − pk−1 )|V (G)|. Indeed, if this is not the case, since |V (G)| ≥ pm, we find that there is
some blue part Vi which has very high blue density to at least roughly pk m other parts.
(k)
This can then be used to find a blue Bn , where n ≈ pk N. So we may conclude that every
vertex in G has high degree. But, by Turán’s theorem, plus the fact that pk−1 < 1/(k − 1)
for sufficiently large k, this implies that G contains a copy of Kk , as desired.
We now begin the detailed proof of Theorem 1.3. The following result generalizes a key
analytic inequality from the diagonal case [10, Lemma 3.4].
Lemma 4.1. For every p ∈ (0, 1), there exists some k1 ∈ N such that if k ≥ k1 and
x1 , . . . , xk ∈ [0, 1], then
k k
Y (1 − p)1−k X
p1−k xi + (1 − xi )k ≥ 1.
i=1
k i=1

Moreover, one may take


6 if p ≥ 1 − 5/(4e)
k1 (p) = 5−log log 1 1
+log(− log log 1−p )
1 + 1−p
1
log 1−p
otherwise.

1
Proof. First suppose that xj ≤ k
for some j ∈ [k]. Then we have that
k k
(1 − p)1−k X (1 − p)1−k (1 − p)1−k (1 − p)1−k

k k 1
(1−xi ) ≥ (1−xj ) ≥ 1− ≥ =: f (p, k),
k i=1
k k k e2 k

where we used the inequality 1−x ≥ e−2x for x ∈ [0, 12 ]. If p ≥ 1−5/(4e), then 1−p ≤ 5/(4e),
so f (p, k) ≥ (4/5)k−1ek−3 /k. Once k ≥ 6, this last expression is at least 1, so in the case
where p ≥ 1 − 5/(4e), we may take k1 (p) = 6.
1
For p < 1 − 5/(4e), let λ = λ(p) = log 1−p and
1 1
5 − log λ + log log λ1 5 − log log 1−p + log(− log log 1−p )
k1 (p) = 1 + =1+ 1 .
λ log 1−p

We now claim that

f (p, k) ≥ 1 if k ≥ k1 (p). (2)
1
By differentiating, we see that f (p, k) is monotonically increasing in k for k ≥ 1/ log 1−p =
1
1/λ. Since p < 1 − 5/(4e), we have that 5 − log λ + log log λ > 1 and so we are in the
monotonicity regime. It therefore suffices to prove the statement for k = k1 (p). Note now
that
1 e5 log λ1
(1 − p)1−k1 (p) = (1 − p)(5−log λ+log log λ )/ log(1−p) =
λ
15
and let g(p) = 1 + λ/ log λ1 + 5/ log λ1 + log log λ1 / log λ1 . Then we have

e5 log λ1 1 e3 log λ1 e3
f (p, k1(p)) = · 2 = = .
λ e k1 (p) λ + 5 + log λ1 + log log λ1 g(p)

Thus, to prove that f (p, k1 (p)) ≥ 1, it suffices to prove that g(p) ≤ e3 for all p < 1 − 5/(4e).
By differentiating, one can check that g(p) is monotonically increasing in p ∈ [0, 1 − 5/(4e)].
Thus, it suffices to check that g(1 − 5/(4e)) ≤ e3 . But g(1 − 5/(4e)) ≈ 18.4 < e3 , so
f (p, k1(p)) ≥ 1, as claimed. Hence, from now on, we may assume that all the xi are in ( k1 , 1].
For the moment, let’s assume that all the xi are in ( k1 , 1). Note that the function ϕ : y 7→
(1 − ey )k is strictly convex on the interval (log k1 , 0). By the multiplicative Jensen inequality,
Lemma 2.5, this implies that, subject to the constraint ki=1 xi = z, the term k1 ki=1 (1−xi )k
Q P
is minimized when all the xi are equal to z 1/k . Therefore,
k k
1−k
Y (1 − p)1−k X
p xi + (1 − xi )k ≥ p1−k z + (1 − p)1−k (1 − z 1/k )k .
i=1
k i=1

So it suffices to minimize this expression as a function of z. Changing variables to w = z 1/k ,

it suffices to minimize
ψ(w) = p1−k w k + (1 − p)1−k (1 − w)k
as a function of w. By differentiating, we find that ψ is minimized at w = p, where ψ(p) = 1.
This proves the desired result as long as all the xi are in [0, 1). By continuity, the result then
extends to all xi ∈ [0, 1].

Definition 4.2. Fix parameters k ∈ N and η, δ ∈ (0, 1) and suppose that we are given a
red/blue coloring of E(KN ). Then a k-tuple of pairwise disjoint vertex sets C1 , . . . , Ck ⊆
V (KN ) is called a (k, η, δ)-red-blocked configuration if the following properties are satisfied:

1. Each Ci is η-regular with itself,

2. Each Ci has internal red density at least δ, and

3. For all i 6= j, the pair (Ci , Cj ) is η-regular and has blue density at least δ.

Similarly, we say that C1 , . . . , Ck is a (k, η, δ)-blue-blocked configuration if properties (1–3)

hold, but with the roles of red and blue interchanged.

The reason we care about these configurations is that, for appropriate choices of the
parameters η and δ, their existence yields the existence of the required monochromatic
books. This idea (or, rather, the version of it when red and blue play symmetric roles)
already appears implicitly in the work of the first author [9], but was made much more
explicit in the prequel to this paper [10]. The precise statement we will need here is given
by the next lemma.

16
Lemma 4.3. For every p ∈ [ 12 , 1), there is k2 ∈ N such that the following holds. Let k ≥ k2 ,
2
c = ((1 − p)/p)k , and 0 < ε < 21 and suppose 0 < δ ≤ (1 − p)ε and 0 < η ≤ δ 4k . Suppose
that the edges of KN with N = (p−k + ε)n are red/blue colored and this coloring contains
either a (k, η, δ)-red-blocked configuration or a (k, η, δ)-blue-blocked configuration. Then, in
(k) (k)
either case, the coloring contains either a red Bcn or a blue Bn . Moreover, one may take
k2 (p) = k1 (1 − p), where k1 is the constant from Lemma 4.1.
Proof. We tackle the two cases separately: suppose first that the coloring has a (k, η, δ)-red-
blocked configuration, say C1 , . . . , Ck . By the counting lemma, Lemma 2.2, we know that
the number of blue Kk with one vertex in each Ci is at least
 ! Y k    Yk
Y k (k
) k
dB (Ci , Cj ) − η |Ci | ≥ δ 2 − η |Ci | > 0,
1≤i<j≤k
2 i=1
2 i=1

so there is at least one blue Kk with one vertex in each of C1 , . . . , Ck . By a similar compu-
tation, we see that each Ci contains at least one red Kk .
For a vertex v and i ∈ [k], let xi (v) := dB (v, Ci ) ∈ [0, 1]. We observe that from the
definition in Lemma 4.1, we have that k1 (1 − p) ≥ k1 (p) for all p ≥ 21 . Therefore, Lemma 4.1
implies that since k ≥ k2 ≥ k1 (p), we have that
k
! k
!
−k X
Y (1 − p)
p p−k xi (v) + (1 − p) (1 − xi (v))k ≥ 1
i=1
k i=1

for all v ∈ V . Summing this fact up over all v, we find that

k
! k X
!
−k X
XY (1 − p)
p p−k xi (v) + (1 − p) (1 − xi (v))k ≥ N. (3)
v∈V i=1
k i=1 v∈V

This says that a (p, 1 − p)-weighted average of two numbers is at least N, which means that
at least one of them is at least N. Suppose first that the first term is at least N, i.e., that
k
XY
xi (v) ≥ pk N.
v∈V i=1

Let Q be a uniformly random blue Q Kk spanning C1 , . . . , Ck , which must 4k

exist by our com-
k2 2
putations above. Let α = δ ≤ i<j dB (Ci , Cj ) and observe that η ≤ δ = α4 ≤ α3 /k 2 .
Thus, for any v, we can
Q apply Lemma 2.3 to conclude that the probability v extends Q to a
blue Kk+1 is at least i xi (v) − 4α. Therefore, the expected number of extensions of Q to a
blue Kk+1 is at least
k
!
X Y
xi (v) − 4α ≥ (pk − 4α)N
v∈V i=1

≥ (pk − 4α)(p−k + ε)n

≥ (1 + pk ε − 8αp−k )n
≥ n, (4)

17
 2 2 2
where (4) uses that α = δ k ≤ ((1 − p)ε)k ≤ (pε)k ≤ p2k ε/8. Therefore, Q has at least n
extensions in expectation, so there must exist some blue Kk with at least n extensions, i.e.,
(k)
a blue Bn .
Now assume that the other term in the weighted average in (3) is at least N, i.e., that
k
1 XX
(1 − xi (v))k ≥ (1 − p)k N.
k i=1 v∈V
Then there must exist some i for which
X
(1 − xi (v))k ≥ (1 − p)k N.
v∈V

Therefore, if Q is a random red Kk inside this Ci , then, by Lemma 2.3, the expected number
of extensions of Q is at least3
X
(1 − xi (v))k − 4α ≥ (1 − p)k − 4α (p−k + ε)n
  
v∈V
 k !
1−p
≥ + (1 − p)k ε − 8αp−k n
p
≥ cn, (5)
where we use the fact that c = ((1 − p)/p)k and that
2 2
α = δ k = ((1 − p)ε)k ≤ pk (1 − p)k ε/8,
since 1 − p ≤ p. Thus, the expected number of red extensions of a red Kk in Ci is at least
(k)
cn, so there must exist a red Bcn . This concludes the proof under the assumption that the
coloring contains a (k, η, δ)-red-blocked configuration.
Now, we instead assume that the coloring contains a (k, η, δ)-blue-blocked configuration
and aim to conclude the same result; the proof is more or less identical, but with the role of
p now played by q = 1 − p. As before, we find that there is at least one red Kk spanning
C1 , . . . , Ck and that each Ci contains at least one blue Kk . For a vertex v and i ∈ [k], let
yi (v) = dR (v, Ci ) ∈ [0, 1] and write q = 1 − p. Since k ≥ k2 = k1 (q), we can sum the result
of applying Lemma 4.1 over all v ∈ V to find that
k
! k X
!
−k X
XY (1 − q)
q q −k yi (v) + (1 − q) (1 − yi (v))k ≥ N.
v∈V i=1
k i=1 v∈V

As before, this is a (q, 1 − q)-weighted average of two terms, which means that one of the
terms must be at least N. Suppose first that the first term is at least N, i.e., that
k
XY
yi (v) ≥ q k N.
v∈V i=1
3
Strictly speaking, if v ∈ Ci , then dR (v, Ci ) 6= 1 − xi (v), as v has no edge to itself. However, this tiny
loss can be absorbed into the error terms and the result does not change.

18
 2
If Q is a uniformly random red Kk spanning C1 , . . . , Ck and α = δ k , then, as before, we find
that the expected number of extensions of Q to a red Kk+1 is at least
k
!
X Y
yi (v) − 4α ≥ (q k − 4α)N ≥ ((1 − p)k − 4α)(p−k + ε)n ≥ cn,
v∈V i=1

by the computation in (5). Therefore, in this case, there must exist some red Kk with at
(k)
least cn red extensions, giving the desired red Bcn . So we may assume instead that
k
1 XX
(1 − yi (v))k ≥ (1 − q)k N,
k i=1 v∈V

which implies that for some i ∈ [k],

X
(1 − yi (v))k ≥ pk N.
v∈V

Thus, if Q is a random blue Kk inside this Ci , we find that the expected number of blue
extensions of Q is at least
X
(1 − yi (v))k − 4α ≥ (pk − 4α)N ≥ (pk − 4α)(p−k + ε)n ≥ n,

v∈V

(k)
by the same computation as in (4). This gives us our blue Bn , completing the proof.
With this result in hand, we can now prove Theorem 1.3.
Proof of Theorem 1.3. Given an integer k ≥ 2, let c1 (k) be the infimum of c ∈ (0, 1] such
that k2 ((c1/k + 1)−1 ) ≤ k, where k2 is the constant from Lemma 4.3. Note that we declare
this infimum to equal 1 if no c ∈ (0, 1] satisfies this condition (as happens for k = 2). In this
case, there is nothing to prove, since Theorem 1.3 for c = 1 is already known [9]. We now
fix c ∈ [c1 , 1] and p = 1/(c1/k + 1) ∈ ( 21 , 1], noting that we have k ≥ k2 (p).
Fix 0 < ε < 21 and suppose we are given a red/blue coloring of E(KN ) where N =
−k
(p + ε)n. Our goal is to prove that if n is sufficiently large in terms of ε, then this coloring
(k) (k)
contains a red Bcn or a blue Bn . To do this, we fix parameters δ = (1 − p)2k ε/(4k) and
2
η = min{δ 4k , (1 − p)/(4k)} depending on c, k, and ε.
We apply Lemma 2.1 to the red graph from our coloring with parameters η and M0 = 1/η
to obtain an equitable partition V (KN ) = V1 ⊔ · · · ⊔ Vm , where each Vi is η-regular and, for
each i, there are at most ηm values 1 ≤ j ≤ m such that the pair (Vi , Vj ) is not η-regular.
Moreover, M0 ≤ m ≤ M = M(η, M0 ). Note that since the colors are complementary, the
same properties also hold for the blue graph. Call a part Vi blue if dB (Vi ) ≥ 12 and red
otherwise.
Suppose first that at least m′ ≥ pm of the parts are blue and rename the parts so that
V1 , . . . , Vm′ are these blue parts. We build a reduced graph G whose vertex set is v1 , . . . , vm′

19
by making {vi , vj } an edge if and only if (Vi , Vj ) is η-regular and dR (Vi , Vj ) ≥ δ. Suppose
that some vertex in G, say v1 , has degree at most (1 − pk−1 − η/p)m′ − 1. Since v1 has at
most ηm ≤ ηm′ /p non-neighbors coming from irregular pairs (V1 , Vj ), this means that there
are at least pk−1 m′ parts Vj such that S (V1 , Vj ) is η-regular and dB (V1 , Vj ) ≥ 1 − δ. Let J be
the set of all these indices j and U = j∈J Vj be the union of all of these Vj . We then have
X X
eB (V1 , U) = eB (V1 , Vj ) ≥ (1 − δ)|V1 ||Vj | = (1 − δ)|V1 ||U|. (6)
j∈J j∈J

Let V1′ ⊆ V1 denote the set of vertices v ∈ V1 with eB (v, U) ≥ (1 − 2δ)|U|. Then we may
write
X X
eB (V1 , U) = eB (v, U) + eB (v, U) ≤ |V1′ ||U| + (1 − 2δ)|V1 \ V1′ ||U|. (7)
v∈V1′ v∈V1 \V1′

Combining inequalities (6) and (7), we find that |V1′ | ≥ 12 |V1 |, where every vertex in V1′ has
blue density at least 1 − 2δ into U. Moreover, since η < 61 , we may apply the η-regularity of
V1 to conclude that the internal blue density of V1′ is at least 21 − η ≥ 31 , while the hereditary
property of regularity implies that V1′ is 2η-regular. Then the counting lemma, Lemma 2.2,
implies that V1′ contains at least
     
1 ′ (k2) k ′ k 1 −(k2) k
dB (V1 ) − 2η |V1 | ≥ 3 − 2η |V1′ |k > 0
k! 2 k! 2
blue Kk , so that V1′ contains at least one blue Kk . Every vertex of this blue Kk has at least
(1 − 2δ)|U| blue neighbors in U, so the blue Kk has at least (1 − 2kδ)|U| blue extensions into
U. Moreover, since we assumed that |J| ≥ pk−1 m′ ≥ pk m and the partition is equitable, we
find that |U| ≥ pk N. Therefore,
(1 − 2kδ)|U| ≥ (1 − 2kδ)pk (p−k + ε)n
= (1 − 2kδ)(1 + pk ε)n
≥ (1 + pk ε − 4kδ)n
≥ n,
since our choice of δ yields 4kδ = (1 − p)2k ε ≤ pk ε. Thus, we find that any blue Kk inside
(k)
V1′ must have at least n blue extensions, giving us our blue Bn .
So we may assume that every vertex in G has degree at least (1 − pk−1 − η/p)m′ . Recall
from (2) that f (1 − p, k) = p1−k /(e2 k) ≥ 1 for k ≥ k1 (1 − p). Since we assume that
k ≥ k2 (p) = k1 (1 − p), this implies that
1 1
pk−1 ≤ ≤ . (8)
e2 k 3(k − 1)
Additionally, by our choice of η ≤ (1 − p)/(4k) ≤ p/(4k), we know that
η 1
≤ .
p 3(k − 1)

20
The previous two inequalities imply that
η 1
1 − pk−1 − >1− ,
p k−1
so that G contains a Kk by Turán’s theorem. Let vi1 , . . . , vik be the vertices of this Kk
and let Cj = Vij for 1 ≤ j ≤ k. Then we claim that C1 , . . . , Ck is a (k, η, δ)-blue-blocked
configuration. The fact that each Ci is η-regular follows immediately from our application of
Lemma 2.1 and the fact that dB (Ci ) ≥ δ follows from the fact that we assumed dB (Ci ) ≥ 21 .
Finally, the definition of edges in G implies that (Ci , Cj ) is η-regular with dR (Ci , Cj ) ≥ δ for
all i 6= j. Thus, our coloring contains a (k, η, δ)-blue-blocked configuration with δ ≤ (1 − p)ε
2 (k)
and η ≤ δ 4k , so Lemma 4.3 implies that the coloring contains either a red Bcn or a blue
(k)
Bn .
We have now finished the proof if at least pm of the parts Vi are blue. Therefore, we
may assume instead that at least m′′ ≥ (1 − p)m of the parts are red and again rename the
parts so that these red parts are V1 , . . . , Vm′′ . We construct a reduced graph G on vertices
v1 , . . . , vm′′ by connecting vi to vj if (Vi , Vj ) is η-regular with dB (Vi , Vj ) ≥ δ. Suppose that
some vertex in G, say v1 , has degree at most (1 − (1 − p)k−1 − η/(1 − p))m′′ − 1. As before,
v1 has at most ηm ≤ ηm′′ /(1 − p) non-neighbors coming from irregular pairs. Thus, if we
let J denote the set of indices j for which (V1 , Vj ) is η-regular with S dR (V1 , Vj ) ≥ 1 − δ, then
we find that |J| ≥ (1 − p)k−1m′′ ≥ (1 − p)k m. Thus, if U = j∈J Vj , then we see that
|U| ≥ (1 − p)k N, since the partition is equitable. Next, as above, we let V1′ ⊆ V1 denote the
set of vertices v ∈ V1 with eR (v, U) ≥ (1 − 2δ)|U| and find that |V1′ | ≥ 21 |V1 |. Therefore, as
above, we know that V1′ contains at least one red Kk and this red Kk has at least (1−2kδ)|U|
red extensions in U. Moreover,
(1 − 2kδ)|U| ≥ (1 − 2kδ)(1 − p)k N
= (1 − 2kδ)(1 − p)k (p−k + ε)n
= (1 − 2kδ)(c + (1 − p)k ε)n (9)
≥ (c + (1 − p)k ε − 4kδ)n
≥ cn, (10)
where in (9) we used the definition of p, which implies that ((1 − p)/p)k = c, and in (10) we
used our choice of δ to see that δ ≤ (1 − p)k ε/(4k). Thus, in this case, we can find a red
(k)
Bcn .
We may therefore assume that every vertex in G has degree at least (1 − (1 − p)k−1 −
η/(1 − p))m′′ . As before, we know that, since k ≥ k2 (p),
1
(1 − p)k−1 ≤ pk−1 ≤
3(k − 1)
and our choice of η ≤ (1 − p)/(4k) implies that
η 1
≤ .
1−p 3(k − 1)

21
Thus, by Turán’s theorem, G must contain a Kk , with vertices vi1 , . . . , vik . If we let Cj = Vij ,
then C1 , . . . , Ck will be a (k, η, δ)-red-blocked configuration, by the definition of edges in G
and the assumption that N is sufficiently large in terms of ε. Thus, by Lemma 4.3, we can
(k) (k)
again conclude that the coloring contains either a red Bcn or a blue Bn .
To finish, we note that, as claimed, we may take c1 (k) ≤ ((1+o(1)) logk k )k . Indeed, for any
c and k, let p(c, k) = (c1/k + 1)−1 and y = y(c, k) = 1/ log[1/p(c, k)]. Then, from Lemmas 4.1
and 4.3, we see that k2 (p(c, k)) = 1 + y(5 + log y + log log y). Thus, if y ≤ (1 + o(1))k/ log k,
then k ≥ k2 (p(c, k)). Since y = 1/ log(1 + c1/k ), this condition is equivalent to c1/k ≥
exp((1 + o(1)) logk k ) − 1 = (1 + o(1)) logk k , which yields the desired bound.

5 Quasirandomness
In the previous section, we showed that for a certain range of c and k, the Ramsey number
(k) (k)
r(Bcn , Bn ) is, as n → ∞, asymptotically equal to the lower bound coming from a p-
random construction. In this section, we strengthen this result, showing that all colorings
whose number of vertices is close to the Ramsey number must either be quasirandom or else
contain substantially larger books than the Ramsey property implies. We make the following
definition.

Definition 5.1. For p ∈ [ 21 , 1) and γ > 0, we say that a red/blue coloring of E(KN ) contains
(p, γ)-many books if it contains

• at least γN k red Kk , each with at least ((1 − p)k + γ)N extensions to a red Kk+1 , or

• at least γN k blue Kk , each with at least (pk + γ)N extensions to a blue Kk+1 .

Here is the restatement of Theorem 1.4 in terms of (p, γ)-many books that we will prove.

Theorem 1.4’. For every p ∈ [ 21 , 1), there exists some k0 ∈ N such that the following holds
for every k ≥ k0 . For every θ > 0, there exists some γ > 0 such that if a red/blue coloring
of E(KN ) is not (p, θ)-quasirandom, then it contains (p, γ)-many books.

To prove Theorem 1.4, we will need a few technical lemmas. At a high level, the proof
closely follows the proof of the main quasirandomness theorem in [10, Section 5], as follows.
First, we prove a strengthening of Lemma 4.1, which can be thought of as a stability version of
that result; it says that if our vector (x1 , . . . , xk ) is bounded in ℓ∞ away from the minimizing
point (p, . . . , p), then the value of the function in Lemma 4.1 is bounded away from its
minimum of 1. Using this, we can strengthen Lemma 4.3 to say that not only does a blocked
configuration imply the existence of the desired monochromatic book, but in fact it implies
the existence of a larger book unless every part of the blocked configuration is ε-regular to
(k)
the entire vertex set. Therefore, assuming our coloring does not contain many blue B(pk +γ)N
(k)
or red B((1−p)k +γ)N , we will be able to repeatedly pull out vertex subsets that are ε-regular
to the entire vertex set until we have almost partitioned all the vertices into such subsets. At

22
that point, we can use the structure coming from this partition to deduce that the coloring
is (p, θ)-quasirandom, as desired.
We begin with the strengthening of Lemma 4.1.
Lemma 5.2. For p ∈ (0, 1), let k1 = k1 (p) be as in Lemma 4.1. Then, for every integer
k ≥ k1 and any ε0 > 0, there exists some δ0 > 0 such that if x1 , . . . , xk ∈ [0, 1] are numbers
with |xj − p| ≥ ε0 for some j, then
k k
1−k
Y (1 − p)1−k X
p xi + (1 − xi )k ≥ 1 + δ0 .
i=1
k i=1

Proof. Let
k k
1−k
Y (1 − p)1−k X
F (x1 , . . . , xk ) = p xi + (1 − xi )k
i=1
k i=1

and ϕ(y) = (1 − ey )k . The goal is to apply Hölder’s defect formula, Theorem 2.6, using the
strict convexity of the function ϕ. However, ϕ is only strictly convex on the interval (log k1 , 0)
and, in order to apply Theorem 2.6, we in fact need a positive lower bound on ϕ′′ , but no
such bound exists for the whole interval (log k1 , 0). Because of this, we need to separately
analyze the cases where all the variables are inside a large subinterval of ( k1 , 1) and when one
of them is outside such a subinterval.
First, suppose that one of the variables, say x1 , is in the interval [0, 1+ε
k
1
], for some small
constant ε1 > 0. Then we have that
k
(1 − p)1−k (1 − p)1−k

k 1 + ε1
F (x1 , . . . , xk ) ≥ (1 − x1 ) ≥ 1− .
k k k
From the proof of Lemma 4.1, we see that this quantity is strictly larger than 1 for all
k ≥ k1 (p), so, by choosing δ0 appropriately, we see that F (x1 , . . . , xk ) ≥ 1 + δ0 in this case.
We may therefore assume from now on that all the variables are at least 1+ε k
1
.
1+ε1
Next, suppose that there exist values x1 , . . . , xk−1 ∈ [ k , 1] such that F (x1 , . . . , xk−1 , 1) =
1. We observe that
" k−1
# k−1
∂F Y Y
= p1−k xi − (1 − p)1−k (1 − xk )k−1 = p1−k xi > 0.
∂xk xk =1 i=1 i=1
xk =1

This implies that if we move from xk = 1 to xk = 1 − ε2 for some sufficiently small ε2 ,

the value of F will decrease. Therefore, there will exist a vector (x1 , . . . , xk ) for which
F (x1 , . . . , xk ) < 1, contradicting Lemma 4.1 as long as k ≥ k1 (p). Thus, for every choice of
x1 , . . . , xk−1 ∈ [ 1+ε
k
1
, 1], we have that F (x1 , . . . , xk−1 , 1) > 1. Since the space [ 1+ε
k
1
, 1]k−1 ×{1}
is compact, we in fact find that F (x1 , . . . , xk−1 , 1) ≥ 1 + δ1′ for all x1 , . . . , xk−1 ∈ [ 1+ε k
1
, 1],
′
for some sufficiently small δ1 depending on p and k. Finally, by continuity of F , we have
that F (x1 , . . . , xk ) ≥ 1 + δ1 whenever xk ≥ 1 − ε2 for some other δ1 , ε2 > 0. Since F is a
symmetric function of its variables, the same conclusion holds if xi ≥ 1 − ε2 for any i. Thus,

23
as long as we take the δ0 in the lemma statement to be smaller than δ1 , we can assume from
now on that xi ∈ [ 1+ε k
1
, 1 − ε2 ] for all i.
By Lemma 2.5, subject to the constraint ki=1 xi = z, the term k1 ki=1 (1 − xi )k is
Q P
minimized when xi = z 1/k for all i. As in the proof of Lemma 4.1, this shows that
F (x1 , . . . , xk ) ≥ ψ(z 1/k ), where ψ(w) = p1−k w k + (1 − p)1−k (1 − w)k . The function ψ
has a global minimum at w = p, where its value is 1. This shows that F (x1 , . . . , xk ) ≥ 1 + δ0
if |z 1/k − p| ≥ ε3 , for some ε3 > 0 depending on p, k, and δ0 . Moreover, by picking δ0
sufficiently small, we can make ε3 as small as we wish. Therefore, we may now assume that
z 1/k = p ± ε3 , which implies that log(z 1/k ) = (log p) ± ε4 for some ε4 > 0, which can also be
made arbitrarily small by picking δ0 appropriately.
We are now ready to apply Hölder’s defect formula. First, we observe that for y ∈
[log 1+εk
1
, log(1 − ε2 )], we have

1 + ε1 k−2
ϕ′′ (y) = key (1 − ey )k−2 (key − 1) ≥ k · · ε2 · ε1 =: m,
k
where
Pk m is a fixed, strictly positive constant. Let yi = log xi for 1 ≤ i ≤ k, so that
1 1/k
k i=1 yi = log(z ). We assumed that |xj − p| ≥ ε0 for some j, which implies that
|yj − log p| ≥ ε0 as well, since the derivative of log x is bounded below by 1 on the interval
(0, 1). Therefore, choosing δ0 small enough that ε4 < ε0 , we see that
k
1X 1 1
(yi − log(z 1/k ))2 ≥ (yj − log(z 1/k ))2 ≥ (ε0 − ε4 )2 ,
k i=1 k k

since log(z 1/k ) = (log p) ± ε4 and |yj − log p| ≥ ε0 . Hence, by Theorem 2.6, we have that
k
1−k (1 − p)1−k X
F (x1 , . . . , xk ) = p z+ (1 − xi )k
k i=1
k
1−k 1X
1−k
= p z + (1 − p) · ϕ(yi )
k i=1
m
≥ p1−k z + ϕ(log(z 1/k )) + (ε0 − ε4 )2
2k
1/k m 2
= ψ(z ) + (ε0 − ε4 )
2k
≥ 1 + δ0 ,

where we use the fact that ψ(w) ≥ 1 for all w ∈ [0, 1] and take δ0 sufficiently small.
Using Lemma 5.2, we can now prove the following strengthening of Lemma 4.3, which
says that if we have a blocked configuration C1 , . . . , Ck and many vertices whose blue density
into Ci is far from p, then we can find a substantially larger monochromatic book than what
is guaranteed by Lemma 4.3.

24
Lemma 5.3. Fix p ∈ [ 21 , 1) and let k ≥ k2 (p), where k2 is the constant from Lemma 4.3.
Suppose 0 < ε0 < 14 and let δ0 = δ0 (ε0 ) be the parameter from Lemma 5.2. Let 0 < δ ≤
2
(1 − p)δ0 ε0 and 0 < η ≤ δ 4k and suppose that C1 , . . . , Ck is either a (k, η, δ)-red-blocked
configuration or a (k, η, δ)-blue-blocked configuration in a red/blue coloring of KN . Define
Bi = {v ∈ KN : |dB (v, Ci ) − p| ≥ ε0 }.
Then the following hold:
(k) (k)
(a) If |Bi | ≥ ε0 N for some i, then the coloring contains a blue B(pk +β)N or a red B((1−p)k +β)N ,
where β = (1 − p)k δ0 ε0 /2.
(b) If, in addition, |Ci | ≥ τ N for all i and some τ > 0, then there exists some 0 < γ < β
depending on ε0 , τ, and δ such that the coloring contains (p, γ)-many books.
Proof. We may assume without loss of generality that |B1 | ≥ ε0 N. As in the proof of
Lemma 4.3, we need to split into two cases, depending on whether C1 , . . . , Ck is blue-blocked
or red-blocked. We begin by assuming that it is (k, η, δ)-red-blocked.
First, as in the proof of Lemma 4.3, observe that each Ci contains at least one red Kk and
there is at least one blue Kk spanning C1 , . . . , Ck . Moreover, if we assume that |Ci | ≥ τ N
for all i, then Lemma 2.2 shows that the number of blue Kk spanning C1 , . . . , Ck is at least
 ! Y (k2) τ k
k    !
k k δ
|Ci | ≥ δ (2) − η
Y k
dB (Ci , Cj ) − η (τ N)k ≥ Nk
1≤i<j≤k
2 i=1
2 2

and similarly, with an additional factor of 1/k!, for the number of red Kk inside each Ci .
For a vertex v and i ∈ [k], let xi (v) = dB (v, Ci ). Lemma 4.1 implies that, for any v ∈ V ,
k
! k
!
−k X
Y (1 − p)
p p−k xi (v) + (1 − p) (1 − xi (v))k ≥ 1.
i=1
k i=1

Additionally, if v ∈ B1 , then |x1 (v) − p| ≥ ε0 , so Lemma 5.2 implies that, for v ∈ B1 ,

k k
! !
−k X
Y (1 − p)
p p−k xi (v) + (1 − p) (1 − xi (v))k ≥ 1 + δ0 .
i=1
k i=1

Adding these two equations up over all v ∈ V shows that

k
! k X
!
−k X
XY (1 − p)
p p−k xi (v) + (1 − p) (1 − xi (v))k ≥ N + δ0 |B1 | ≥ (1 + δ0 ε0 )N.
v∈V i=1
k i=1 v∈V

That is, a (p, 1 − p)-weighted average of two quantities is at least (1 + δ0ε0 )N, which implies
that one of the quantities must itself be at least (1 + δ0 ε0 )N. Suppose first that
k
XY
−k
p xi (v) ≥ (1 + δ0 ε0 )N.
v∈V i=1

25
 2
Let Q be a uniformly random blue Kk with one vertex in each of C1 , . . . , Ck . Let α = δ k ≤
4k 2
= α4 ≤ α3 /k 2 . Therefore, applying Lemma 2.3 to each v
Q
i<j dB (Ci , Cj ), so that η ≤ δ
and summing up the result, we find that the expected number of blue extensions of Q is at
least
k
!
X Y
xi (v) − 4α ≥ (pk + pk δ0 ε0 − 4α)N.
v∈V i=1

Next, observe that

2 δk ((1 − p)δ0 ε0 )k (1 − p)k δ0 ε0 pk δ0 ε0
4α = 4δ k ≤ ≤ ≤ ≤ , (11)
2 2 2 2
which implies that the expected number of blue extensions of Q is at least (pk + β)N, where
(k)
β = (1 − p)k δ0 ε0 /2. Thus, there exists a blue B(pk +β)N , proving (a) in this case. Moreover,
if we assume that |Ci | ≥ τ N for all i, then our earlier computation shows that Q is chosen
uniformly at random from a set of at least κN k monochromatic cliques, where κ = δ (2) τ k /2.
k

We may therefore apply Lemma 2.4 with ξ = pk + β and ν = pk + γ, for some appropriately
chosen 0 < γ < β, to conclude that in this case our coloring contains at least γN k blue
cliques, each with at least (pk + γ)N blue extensions, proving (b).
Therefore, we may assume that the second term in the weighted average is the large one,
i.e., that
k
(1 − p)−k X X
(1 − xi (v))k ≥ (1 + δ0 ε0 )N,
k i=1 v∈V

which implies that, for some i,

X
(1 − xi (v))k ≥ (1 − p)k (1 + δ0 ε0 )N.
v∈V

Therefore, if Q is now a random red Kk inside this Ci , Lemma 2.3 implies that the expected
number of red extensions of Q is at least
X
(1 − xi (v))k − 4α ≥ (1 − p)k + (1 − p)k δ0 ε0 − 4α N.
  
v∈V

But, by (11), 4α ≤ (1 − p)k δ0 ε0 /2, which implies that the expected number of red extensions
of Q is at least ((1 − p)k + β)N, proving (a). As before, if we also assume that |Ci | ≥ τ N for
all i, then we may apply Lemma 2.4 with κ = δ (2) τ k /2k!, ξ = (1−p)k +β, and ν = (1−p)k +γ
k

to find that our coloring contains at least γN k red Kk , each with at least ((1 − p)k + γ)N red
extensions for some appropriately chosen γ ∈ (0, β), yielding (b). This concludes the proof
of the lemma in the case where C1 , . . . , Ck is a (k, η, δ)-red-blocked configuration.
As in the proof of Lemma 4.3, the other case, where C1 , . . . , Ck is a (k, η, δ)-blue-blocked
configuration, follows in an almost identical fashion. We define yi (v) = dR (v, Ci ) for all
v ∈ V and i ∈ [k] and let q = 1 − p. We then apply Lemmas 4.1 and 5.2 with these y
variables and with q instead of p. The remaining details are exactly the same.

26
 Next, we strengthen Lemma 5.3 by showing that not only does every part of a blocked
configuration have density roughly p to most vertices, but it is in fact (p, ε)-regular to the
entire vertex set. Here, by saying that a pair of vertex subsets (X, Y ) is (p, ε)-regular, we
mean that |d(X ′, Y ′ ) − p| ≤ ε for every X ′ ⊆ X, Y ′ ⊆ Y with |X ′ | ≥ ε|X|, |Y ′ | ≥ ε|Y |. Note
that (p, ε)-regularity is equivalent, up to a linear change in the parameters, to ε-regularity
with density p ± ε.

Lemma 5.4. Fix p ∈ [ 12 , 1) and let k ≥ k2 (p). Suppose 0 < ε1 < 14 , ε0 = ε21 /2, and let δ0 =
2 2
δ0 (ε0 ) be the parameter from Lemma 5.2. Let 0 < δ ≤ (1−p)δ0 ε0 and 0 < η ≤ ε1 2−4k δ 4k and
suppose that C1 , . . . , Ck is either a (k, η, δ)-red-blocked or a (k, η, δ)-blue-blocked configuration
in a red/blue coloring of KN . Then the following hold:

(a) If, for some i, the pair (Ci , V ) is not (p, ε1)-regular in blue, then the coloring contains
(k) (k)
a blue B(pk +β)N or a red B((1−p)k +β)N , where β = (1 − p)k δ0 ε0 /2.

(b) If, in addition, |Ci | ≥ τ N for all i and some τ > 0, then the coloring contains (p, γ)-
many books for some 0 < γ < β depending on ε1 , δ, and τ .

Proof. Without loss of generality, suppose that (C1 , V ) is not (p, ε1 )-regular in blue. Then
there exist C1′ ⊆ C1 , D ⊆ V with |C1′ | ≥ ε1 |C1 |, |D| ≥ ε1 N such that |dB (C1′ , D) − p| > ε1 .
Assume first that dB (C1′ , D) ≥ p + ε1 . Let D1 ⊆ D denote the set of vertices v ∈ D with
dB (v, C1′ ) < p + ε21 and let D2 = D \ D1 . Then we have that

′
X
′
X
′
 ε1  ′
(p + ε1 ) |C1 ||D| ≤ eB (v, C1 ) + eB (v, C1 ) ≤ p + |C1||D| + |C1′ ||D2 |,
v∈D v∈D
2
1 2

ε2
which implies that |D2 | ≥ ε21 |D| ≥ 21 N = ε0 N, where each v ∈ D2 has dB (v, C1′ ) ≥ p + ε21 .
Now, consider the k-tuple of sets C1′ , C2 , . . . , Ck ; by the hereditary property of regularity,
we see that this is a (k, η ′ , δ ′ )-blocked configuration, where η ′ = η/ε1 and δ ′ = δ − η ≥ δ/2.
2
This implies that δ ′ ≤ (1 − p)δ0 ε0 and η ′ ≤ (δ ′ )4k . Therefore, we may apply Lemma 5.3(a)
to the (k, η ′ , δ ′ )-blocked configuration C1′ , C2 , . . . , Ck to conclude that the coloring contains
(k) (k)
a blue B(pk +β)N or a red B((1−p)k +β)N . Moreover, if we assume that |Ci | ≥ τ N for all i, then
|Ci′ | ≥ ε1 τ N for all i, where Ci′ = Ci if i ≥ 2. Thus, Lemma 5.3(b) implies that in this case
the coloring contains (p, γ)-many books for some 0 < γ < β depending on ε1 , δ, and τ .
To complete the proof of the lemma, we also need to check the case where dB (C1′ , D) ≤
p − ε1 . However, the proof is essentially identical: we find a subset D2 ⊆ D such that every
vertex v ∈ D2 has dB (v, C1′ ) ≤ p − ε21 and such that |D2 | ≥ ε21 |D| and then the rest of the
proof is as above.
Our next technical lemma gives the inductive step for our proof of Theorem 1.4. The
proof mimics that of Theorem 1.3, except that the vertex set is split into parts that were
already pulled out as regular and a part that has not yet been touched. Inside the untouched
part, we build a reduced graph and use it to find either many large monochromatic books or
a blocked configuration, at which point Lemma 5.4 implies that the induction can continue.

27
Lemma 5.5. Fix p ∈ [ 12 , 1) and let k ≥ k2 (p). Fix 0 < ε ≤ p/(20k) and suppose that
the edges of the complete graph KN with vertex set V have been red/blue colored. Suppose
that A1 , . . . , Aℓ are disjoint subsets of V such that (Ai , V ) is (p, ε2)-regular for all i. Let
W = V \ (A1 ∪ · · · ∪ Aℓ ) and suppose that |W | ≥ εN. Then either there is some Aℓ+1 ⊆ W
such that (Aℓ+1 , V ) is (p, ε2 )-regular or else the coloring contains (p, γ)-many books for some
γ > 0 depending on ε, p, and k.

Proof. Let ε1 = ε2 , ε0 = ε21 /2, and δ0 = δ0 (ε0 ) be the parameter from Lemma 5.2 and set
2 2
δ = (1 − p)δ0 ε0 , η = ε2 2−4k δ 4k , β = kpk−1 ε2 , and β ′ = 4ε. We apply Lemma 2.1 to the
subgraph induced on W , with parameters η and M0 = 1/η, to obtain an equitable partition
W = W1 ⊔ · · · ⊔ Wm , where M0 ≤ m ≤ M = M(η, M0 ). Call a part Wi blue if dB (Wi ) ≥ 12
and red otherwise. As in the proof of Theorem 1.3, we first assume that at least m′ ≥ pm
of the parts are blue and rename them so that W1 , . . . , Wm′ are the blue parts.
We build a reduced graph G on vertex set w1 , . . . , wm′ , connecting wi1 and wi2 by an
edge if (Wi1 , Wi2 ) is η-regular and dR (Wi1 , Wi2 ) ≥ δ. Suppose that w1 has at most (1 −
pk−1 − β ′ /p − η/p)m′ − 1 neighbors in G. Since w1 has at most ηm ≤ ηm′ /p non-neighbors
coming from irregular pairs, this means that there are at least (pk−1 + β ′/p)m′ parts Wj with
2 ≤ j ≤ m′ such that (W1 , W Sj ) is η-regular and dB (W1 , Wj ) ≥ 1 − δ. Let J be the set of
these indices j and set U = j∈J Wj . By the counting lemma, Lemma 2.2, W1 contains at
 k

least 1 2−(2) − η k |W |k blue copies of K and
k! 2 1 k

2  k  k
2−k |W |
  
1 −(k2) k k εN
2 −η |W1 | ≥ ≥ ,
k! 2 k! M k2k M

where we use that η ≤ δ 4k ≤ δ (2) / k2 and that 2−(2) − δ (2) > 2−k , along with our
2 k
k k 2

assumption that |W | ≥ εN. If we set κ = (ε/k2k M)k , then this implies that W1 contains
at least κN k blue Kk . If we pick a uniformly random such blue Kk , then Lemma 2.3 with
α = δ k ≤ 2−(2) ≤ dB (W1 )(2) implies that its expected number of blue extensions inside U
2 k k

is at least
X
dB (u, W1)k − 4α ≥ (1 − δ)k − δ k |U| ≥ (1 − 2kδ)|U|,

 
u∈U

where we first use Jensen’s inequality applied to the convex function x 7→ xk to lower bound
k k k k2
≤ δ k ≤ kδ. Since
P
u dB (u, W1 ) by (1 − δ) |U| and then use that (1 − δ) ≥ 1 − kδ and 4δ
we assumed that J was large and the partition is equitable, we find that

|U| ≥ (pk−1 + β ′ /p)m′ |Wj | ≥ (pk + β ′ )|W |.

Thus, a random blue Kk inside W1 has at least (1 − 2kδ)(pk + β ′ )|W | blue extensions in W .
Now, suppose that instead of just w1 having low degree in G, we have a set of at least
k−1 ′ ′
εm vertices wj ∈ V (G), each with
Sat most (1 − p − β /p − η/p)m − 1 neighbors in G. Let
S be the set of these j and T = j∈S Wj . By the above argument, for every j ∈ S, we have

28
that Wj contains at least κN k blue Kk such that a uniformly average one among them has
at least (1 − 2kδ)(pk + β ′ )|W | blue extensions into W . Moreover, we have that
|W |
|T | = |S||Wj | ≥ εm = ε|W | ≥ ε2 |V |.
m
We may therefore apply the (p, ε2) regularity of (Ai , V ) to conclude that dB (Ai , T ) = p ± ε2
for all i. Thus, if we pick j ∈ S randomly, then E[dB (Wj , Ai )] = p ± ε2 . Therefore, if we first
sample j ∈ S randomly and then pick a random blue Kk inside Wj , then Lemma 2.3 implies
that this random blue Kk will have in expectation at least
X 2
 h
k i
dB (a, Wj )k − 4δ k ≥ p − ε2 − δ k |Ai |
a∈Ai
kε2

  
k k
≥ p 1− − δ |Ai |
p
2kε2
 
k
≥p 1− |Ai |
p
blue extensions into Ai , again by Jensen’s inequality. This implies that this random Kk has in
expectation at least (1 − 2kε2 /p)pk |A1 ∪ · · ·∪ Aℓ | extensions into A1 ∪ · · · ∪ Aℓ . Adding up the
extensions into this set and into W , its complement, shows that this random blue Kk has in
expectation at least ξN blue extensions, where ξ is a weighted average of (1 − 2kε2 /p)pk and
(1 − 2kδ)(pk + β ′ ), and where the latter quantity receives weight at least ε, since |W | ≥ εN.
Thus,
2kε2
 
ξ ≥ (1 − ε) 1 − pk + ε(1 − 2kδ)(pk + β ′ )
p
2kε2
 
≥ 1− − ε pk + ε(1 − 2kδ)(1 + p−k β ′ )pk
p
2kε2
   
k 3kε k
≥ 1− −ε p +ε 1+ p
p p
kε2
 
k
=p 1+
p
k
= p + β,
where we used the definition of β, the fact that 2kδ < p−k β ′ /4, that (1−x/4)(1+x) ≥ 1+x/2
for all x ∈ [0, 1], and that p−k β ′ ≥ 6kε/p, which follows since β ′ = 4ε and, as in the proof of
Lemma 4.1, p1−k ≥ e2 k ≥ 32 k for k ≥ k2 (p). Therefore, by Lemma 2.4, we can find at least
γN k blue Kk , each with at least (pk + γ)N blue extensions, for some γ < β depending on ε
and β and, thus, only on ε, p, and k.
Therefore, we may assume that in G, all but εm ≤ εm′ /p of the vertices have degree at
least (1 − pk−1 − β ′ /p − η/p)m′ . Hence, the average degree in G is at least
β′ η
      
ε k−1 ′ k−1 6ε ′ k−1 1
1− 1−p − − m ≥ 1−p − m ≥ 1−p − m′ ,
p p p p 3k

29
since β ′ = 4ε, η ≤ ε, and ε ≤ p/(20k). By (8), the fact that k ≥ k2 (p) implies that
pk−1 ≤ 1/(3k). Therefore, the average degree in G is greater than (1 − 1/(k − 1))m′ , so,
by Turán’s theorem, G will contain a Kk . Let wi1 , . . . , wik be the vertices of this Kk and
let Cj = Wij for 1 ≤ j ≤ k. Then, by the definition of G, we see that C1 , . . . , Ck is a
(k, η, δ)-blue-blocked configuration with |Ci | ≥ τ N for all i, where τ = ε/M depends only
on ε, p, and k. Thus, by Lemma 5.4(b), we see that either the coloring contains (p, γ)-many
books for some γ depending on ε, p, and k or else (Cj , V ) is (p, ε2 )-regular for all j. In the
latter case, we can set Aℓ+1 = C1 (or any other Cj ) and get the desired result.
Now, we need to assume instead that at least m′′ ≥ (1 − p)m of the parts Wi are red.
However, just as in the proof of Theorem 1.3, the argument is essentially identical: we first
rule out the existence of too many low-degree vertices in the reduced graph by counting
extensions to W and to A1 ∪ · · · ∪ Aℓ and then apply Turán’s theorem to find a Kk in the
reduced graph, which completes the proof by Lemma 5.4(b).
By repeatedly applying Lemma 5.5 until W has fewer than εN vertices, we can partition
KN into a collection of subsets Ai such that (Ai , V ) is (p, ε2 )-regular, plus a small remainder
set Aℓ+1 about which we have no such information. Our final technical lemma shows that
such a structural decomposition suffices to conclude that the coloring is (p, θ)-quasirandom.

Lemma 5.6. Let ε ≤ θ/3. Suppose we have a partition

V (KN ) = A1 ⊔ · · · ⊔ Aℓ ⊔ Aℓ+1

where (Ai , V ) is (p, ε)-regular for each 1 ≤ i ≤ ℓ and |Aℓ+1 | ≤ εN. Then the coloring is
(p, θ)-quasirandom.

Proof. Fix disjoint X, Y ⊆ V (KN ). We need to check that

|eB (X, Y ) − p|X||Y || ≤ θN 2 .

First, observe that if |Y | ≤ εN, then

|eB (X, Y ) − p|X||Y || ≤ |X||Y | ≤ εN 2 ≤ θN 2 .

Therefore, from now on, we may assume that |Y | ≥ εN. For 1 ≤ i ≤ ℓ + 1, let Xi = Ai ∩ X
and define IX = {1 ≤ i ≤ ℓ : |Xi | ≥ ε|Ai |}. Then we have that

X ℓ
X
|Xi | ≤ |Aℓ+1 | + ε |Ai | ≤ 2εN.
i∈I
/ X i=1

We now write
ℓ+1
X
eB (X, Y ) − p|X||Y | = (eB (Xi , Y ) − p|Xi ||Y |) .
i=1

30
We will split this sum into two parts, depending on whether i ∈ IX or not. First, suppose
that i ∈ IX . Then |Xi | ≥ ε|Ai | and |Y | ≥ ε|V |, so we may apply the (p, ε)-regularity of
(Ai , V ) to conclude that
X X X
|eB (Xi , Y ) − p|Xi ||Y || = |dB (Xi , Y ) − p| |Xi ||Y | ≤ ε|Xi ||Y | ≤ ε|X||Y | ≤ εN 2 .
i∈IX i∈IX i∈IX
P
On the other hand, since i∈I / X |Xi | ≤ 2εN, we have that
X X
|eB (Xi , Y ) − p|Xi ||Y || ≤ |Y | |Xi | ≤ |Y |(2εN) ≤ 2εN 2 .
i∈I
/ X i∈I
/ X

Adding these together, we conclude that

|eB (X, Y ) − p|X||Y || ≤ 3εN 2 ≤ θN 2 ,
as desired.
With all these pieces in place, the proof of Theorem 1.4 becomes quite straightforward.
Proof of Theorem 1.4. Fix p ∈ [ 21 , 1) and suppose k ≥ k0 := k2 (p). Fix θ > 0 and set
ε = min{θ/3, p/(20k)}. Let γ = γ(θ, p, k) be the parameter from Lemma 5.5. Suppose we
are given a coloring of KN without (p, γ)-many books. We wish to prove that the coloring
is (p, θ)-quasirandom. We inductively apply Lemma 5.5 to find a sequence A1 , . . . , Aℓ of
vertex subsets such that (Ai , V ) is (p, ε2 )-regular for all i and, therefore, (p, ε)-regular for
all i. We continue until the remainder set W = V \ (A1 ∪ · · · ∪ Aℓ ) satisfies |W | ≤ εN,
at which point the assumptions of Lemma 5.5 are no longer met, so we set Aℓ+1 = W .
However, at this point, we can apply Lemma 5.6 to conclude that our coloring is indeed
(p, θ)-quasirandom.

5.1 The converse

In this section, we prove a converse to Theorem 1.4, which implies that not containing
(p, γ)-many books is an equivalent characterization of p-quasirandomness.
Theorem 5.7. Fix k ≥ 2 and p ∈ (0, 1). Then, for every γ > 0, there exists some θ > 0 such
that the following holds for every (p, θ)-quasirandom coloring of E(KN ) with N sufficiently
large. Apart from fewer than γN k exceptions, every red Kk has ((1 − p)k ± γ)N extensions
to a red Kk+1 and every blue Kk has (pk ± γ)N extensions to a blue Kk+1 . In particular, the
coloring does not contain (p, γ)-many books.
Remark. In this direction, there is no dependence between p and the range of k for which
the result holds. As we know from the fact that the k-partite structure is the extremal
structure for small c, such a dependence is necessary in the forward direction. However,
here, all we are saying is that almost all monochromatic books in a quasirandom coloring
are of essentially the correct size, that is, asymptotic to what they would be in a random
coloring.

31
Proof. We will use the well-known result of Chung, Graham, and Wilson [6], that a quasir-
andom coloring contains roughly the correct count of any fixed monochromatic subgraph.
Specifically, for every δ > 0, there is some θ > 0, such that, in any (p, θ)-quasirandom
coloring of E(KN ),
 
N
B(Kk ) := #(blue Kk ) = p(2)
k
± δN k ,
k
 
( k+1
) N
B(Kk+1 ) := #(blue Kk+1 ) = p 2 ± δN k+1 ,
k+1
  
( k+2
) −1 N k+2
B(Kk+2 − e) := #(blue Kk+2 − e) = p 2 ± δN k+2 ,
k+2 2

where Kk+2 − e is the graph formed
by deleting one edge from Kk+2 ; note that for this
count we have an extra factor of k+2 2
to account for the fact that this graph is not vertex-
transitive. On the other hand, we can observe that every blue copy of Kk+2 − e corresponds
to two distinct extensions of a single blue Kk to a blue Kk+1 . Therefore,
X #(blue extensions of Q)
B(Kk+2 − e) = ,
Q
2

where the sum is over all blue KPk . Let extB (Q) denote the number of blue extensions of Q.
Then we can also observe that Q extB (Q) counts the total number of ways of extending a
blue Kk into a blue Kk+1 , which is precisely (k +1)B(Kk+1), since each blue Kk+1 contributes
exactly k + 1 terms to this sum.
Now, we consider the quantity
X
E= (extB (Q) − pk N)2 .
Q a blue Kk

On the one hand, we have that if δ ≥ 1/N, then

X X X
E= extB (Q)2 − 2pk N extB (Q) + p2k N 2
Q Q Q
!
X extB (Q) X
= 2 + extB (Q) − 2pk N(k + 1)B(Kk+1) + p2k N 2 B(Kk )
Q
2 Q

= 2B(Kk+2 − e) + (1 − 2pk N)(k + 1)B(Kk+1 ) + p2k N 2 B(Kk )

      
( k+2
) −1 N k+2 k (k+1
) N 2k 2 (k2) N
≤ 2p 2 − 2p N(k + 1)p 2 +p N p + 5kδN k+2
k+2 2 k+1 k
 
2
k +3k N
=p 2 (−N + k 2 + k) + 5kδN k+2
k
< 5kδN k+2 .

32
On the other hand, suppose there were at least γN k /2 blue Kk with at least (pk + γ)N or
at most (pk − γ)N blue extensions. Then, by only keeping these cliques in the sum defining
E, we would have that
X γN k γ3
E= (extB (Q) − pk N)2 ≥ (γN)2 = N k+2 .
Q
2 2

Therefore, if we pick δ < γ 3 /10k, we get a contradiction. The same argument with p replaced
by 1 − p and blue replaced by red shows that there are also fewer than γN k /2 red Kk with
at least ((1 − p)k + γ)N or at most ((1 − p)k − γ)N red extensions. This proves the theorem,
since the total number of exceptional cliques is at most γN k .

6 Concluding remarks
Putting together the main results of this paper, we obtain the following picture. For ev-
ery k ≥ 2, there exist two numbers c0 (k), c1 (k) ∈ (0, 1] such that if 0 < c ≤ c0 , then
(k) (k) (k) (k)
r(Bcn , Bn ) = k(n + k − 1) + 1, while if 1 ≥ c ≥ c1 , then r(Bcn , Bn ) = (c1/k + 1)k n + ok (n).
Moreover, in both these regimes, there are stability results: there exist c′0 (k) ≤ c0 (k) and
c′1 (k) ≥ c1 (k) such that for 0 < c ≤ c′0 , all the near-extremal colorings are close to k-partite,4
while for all 1 ≥ c ≥ c′1 , all near-extremal colorings are quasirandom. Of course, the most
natural question remaining is to understand what happens in the interval (c′0 , c′1 ), where our
results say nothing. Note that this gap is real, since below c′0 all extremal colorings must
be k-partite, whereas above c′1 all extremal colorings must be quasirandom. On the other
hand, it is possible that there is no gap between c0 and c1 , since it is conceivable that at
the point where the random and k-partite constructions yield comparable lower bounds on
(k) (k)
r(Bcn , Bn ), both are tight.
This question about the gap really comprises at least two separate questions: what
happens for fixed k and what happens as k → ∞? To address the second question first, our
results give some indication. Indeed, we have shown that both c0 (k) and c1 (k) tend to 0 as
k → ∞ and thus the gap interval shrinks as k → ∞. More precisely, we have that
 k  k
log k log k
c0 (k) ≤ (1 + o(1)) ≤ c1 (k) ≤ (1 + o(1)) .
k k

Moreover, the results of [15] show that 1/c0 is at most single-exponential in a power of k.
On the other hand, because we used the regularity lemma, our upper bound for 1/c′0 is only
of tower-type. However, it seems likely that the methods of [15] could also be adapted to
improve this.
The other question is what happens for fixed k. Here, our understanding is much more
limited, even for the simplest case k = 2. In this case, Nikiforov and Rousseau [19] proved
(2) (2)
that c0 (2) = 1/6, in the sense that, for all c < 1/6 and all n sufficiently large, r(Bcn , Bn ) =
4
For concreteness, we can fix c′0 (k) as coming from an application of Theorem 1.2 with θ = 1/k 3 .

33
2n + 3, whereas, for any c > 1/6 and all n sufficiently large, there is a construction showing
(2) (2)
that r(Bcn , Bn ) > 2n + 3. Curiously, our results do not say anything non-trivial about
c1 (2), other than the fact that the random bound is correct for c = 1; in other words, we
cannot prove that c1 (2) < 1 and in fact believe this to not be the case.
(2) (2)
Conjecture 6.1. For every c < 1, the random bound for r(Bcn , Bn ) is not tight. In other
(2) (2) √
words, there exists some β = β(c) > 0 such that r(Bcn , Bn ) ≥ (( c + 1)2 + β)n for all n
sufficiently large.

Of course, this conjecture is really only the tip of an iceberg, with the general open
(2) (2)
question being to understand r(Bcn , Bn ) for c ∈ (1/6, 1) and n → ∞. There are many
conjectures one could make about the behavior of this quantity as a function of c; for
instance, perhaps there are a number of thresholds in the interval (1/6, 1) at which new
(2) (2)
extremal structures emerge, each dictating the value of r(Bcn , Bn ) until the next threshold.
Because we know that the random bound is correct for c = 1 and that quasirandom colorings
are the only extremal ones, such a sequence of extremal examples would need to converge,
in some appropriate sense, to the quasirandom coloring as c → 1. However, at the moment
we are not even able to conjecture a single such extremal structure or threshold.

Acknowledgments. We are grateful to the anonymous referee for helpful comments which
improved the presentation of this paper.

References
[1] M. Ajtai, J. Komlós, and E. Szemerédi, A note on Ramsey numbers, J. Combin. Theory
Ser. A 29 (1980), 354–360.

[2] B. Andrásfai, P. Erdős, and V. T. Sós, On the connection between chromatic number,
maximal clique and minimal degree of a graph, Discrete Math. 8 (1974), 205–218.

[3] T. Bohman and P. Keevash, Dynamic concentration of the triangle-free process, Random
Structures Algorithms 58 (2021), 221–293.

[4] S. Brandt, On the structure of graphs with bounded clique number, Combinatorica 23
(2003), 693–696.

[5] S. A. Burr and P. Erdős, Generalizations of a Ramsey-theoretic result of Chvátal, J.

Graph Theory 7 (1983), 39–51.

[6] F. R. K. Chung, R. L. Graham, and R. M. Wilson, Quasi-random graphs, Combinatorica

9 (1989), 345–362.

[7] V. Chvátal and F. Harary, Generalized Ramsey theory for graphs. III. Small off-diagonal
numbers, Pacific J. Math. 41 (1972), 335–345.

34
 [8] D. Conlon, A new upper bound for diagonal Ramsey numbers, Ann. of Math. 170
(2009), 941–960.

[9] D. Conlon, The Ramsey number of books, Adv. Combin. (2019), Paper No. 3, 12 pp.

[10] D. Conlon, J. Fox, and Y. Wigderson, Ramsey numbers of books and quasirandomness,
Combinatorica 42 (2022), 309–363.

[11] P. Erdős, Some remarks on the theory of graphs, Bull. Amer. Math. Soc. 53 (1947),
292–294.

[12] P. Erdős, R. J. Faudree, C. C. Rousseau, and R. H. Schelp, The size Ramsey number,
Period. Math. Hungar. 9 (1978), 145–161.

[13] P. Erdős and G. Szekeres, A combinatorial problem in geometry, Compos. Math. 2

(1935), 463–470.

[14] G. Fiz Pontiveros, S. Griffiths, and R. Morris, The triangle-free process and the Ramsey
number R(3, k), Mem. Amer. Math. Soc. 263 (2020), v+125 pp.

[15] J. Fox, X. He, and Y. Wigderson, Ramsey goodness of books revisited. Preprint available
at arXiv:2109.09205.

[16] T. Kővári, V. T. Sós, and P. Turán, On a problem of K. Zarankiewicz, Colloq. Math. 3

(1954), 50–57.

[17] J. H. Kim, The Ramsey number R(3, t) has order of magnitude t2 / log t, Random Struc-
tures Algorithms 7 (1995), 173–207.

[18] D. Mubayi and J. Verstraëte, A note on pseudorandom Ramsey graphs. J. Eur. Math.
Soc., to appear. Preprint available at arXiv:1909.01461.

[19] V. Nikiforov and C. Rousseau, Book Ramsey numbers I, Random Structures Algorithms
27 (2005), 379–400.

[20] V. Nikiforov and C. C. Rousseau, Ramsey goodness and beyond, Combinatorica 29

(2009), 227–262.

[21] F. P. Ramsey, On a problem of formal logic, Proc. London Math. Soc. 30 (1929), 264–
286.

[22] A. Sah, Diagonal Ramsey via effective quasirandomness. Duke Math. J., to appear.
Preprint available at arXiv:2005.09251.

[23] J. B. Shearer, A note on the independence number of triangle-free graphs, Discrete

Math. 46 (1983), 83–87.

35
[24] J. M. Steele, The Cauchy-Schwarz master class: An introduction to the art of mathe-
matical inequalities, MAA Problem Books Series, Mathematical Association of America,
Washington, DC, Cambridge University Press, Cambridge, 2004.

[25] A. Thomason, On finite Ramsey numbers, European J. Combin. 3 (1982), 263–273.

[26] Y. Zhao, Graph theory and additive combinatorics, 2022.

[Link]

36