ANT ´
ONIO GIR ˜
AO, GAL KRONENBERG,AND ALEX SCOTT
e−clog2k2k
k
the lower bound due to Spencer [
20
] and upper bound by Sah [
17
]. For larger
r
, the gap
between the upper and lower bound is even larger, and even for
k=3
, we do not understand the behaviour
of
Rr(3)
as
r→∞
. Very recently, Conlon and Ferber [
6
], Wigderson [
23
], and Sawin [
18
] found nice
constructions which give the best known lower bounds for
Rr(k)
when
r≥3
. In a recent breakthrough,
Campos, Griffiths, Morris and Sahasrabudhe [
4
], reduced the the long standing upper bound of
4k
to
(4−ε)k
. This is the first exponential improvement over the upper bound of Erd˝
os and Szekeres, proved
in 1935.
In this paper, we prove a “multidimensional" generalisation of Ramsey’s Theorem for cartesian
products of graphs. Given two graphs
H
and
G
, we write
G□H
for the Cartesian product of
H
and
G
,
namely the graph with vertex set
V(G)×V(H)
in which
(x,y)
is joined to
(x′,y′)
if and only if
x=x′
and
yy′∈E(H)
or
xx′∈E(G)
and
y=y′
. The Cartesian product is associative, so it makes sense to
write
G1□G2□···□Gd
(without brackets) for the Cartesian product of
d
graphs; we write
□dG
for the
product
G□···□G
of
d
copies of
G
. Note that in a Cartesian product of
d
graphs
G1,...,Gd
, there is an
edge between
v= (v1,...,vd)
and
w= (w1,...,wd)
if and only if there is some
i
such that
viwi
is an edge
of Giand vj=wjfor j̸=i, and in this case we will say that the edge vw is in direction i.
Given a colouring
c
of the edges of
□dKn
, we say that
c
is monochromatic in every direction if for
each
i∈ {1,...,d}
there is some
ci
such that all edges in direction
i
have colour
ci
. For positive integers
r,d
, we define
Rr(d,n)
to be the smallest
N
such that every
r
-colouring of the edges of
□dKN
contains a
copy of
□dKn
that is monochromatic in every direction. Note that we cannot demand a copy of
□dKn
that has the same colour in every direction, as we are asking for a full-dimensional subgraph: for example,
□dKN
could be coloured with colour
1
for all edges in direction
1
and coloured with 2 for the edges in
all other directions. It is easy to see that if we demand a monochromatic copy of
□ℓKn
then
ℓ
must be at
most ⌈d/r⌉. It will follow from Theorem 1.1 that this is also tight.
It is not too hard to prove that
Rr(d,n)
exists by an iterated application of Ramsey’s Theorem.
However, this would only give an upper bound of tower-type as a function of
d
. Our main goal is to show
that a doubly exponential bound on ndsuffices.
Theorem 1.1. Let
d
be a positive integer. There exists
Cd>0
such that for every
n,r
the following holds.
For
N≥rrCdrnd
, every
r
-colouring of
□dKN
contains a copy of
□dKn
which is monochromatic in every
direction. That is, Rr(d,n)≤rrCdrnd
.
As an immediate corollary, we see that any
r
-edge colouring of
□dKN
contains a monochromatic
copy of □ℓKnfor ℓ=⌈d/r⌉.
Corollary 1.2. Let
n,d,r
be positive integers and
ℓ=⌈d/r⌉
. For
N≥rrCdrnd
, every
r
-edge-colouring of
□dKNcontains a monochromatic copy of □ℓKn. The value of ℓis tight.
Another foundational result in Ramsey theory appears in a paper Erd˝
os and Szekeres [
8
] from 1935:
any sequence of
n2+1
distinct real numbers contains either an increasing or decreasing subsequence of
length
n+1
. There are a number of different ways to generalise the Erd˝
os-Szekeres Theorem to higher
dimensions (see, for example, [
2
,
3
,
11
,
12
,
14
,
15
,
19
,
21
]). Perhaps the most natural approach was
developed thirty years ago by Fishburn and Graham [10].
A
d
-dimensional array is an injective function
f
from
A1×... ×Ad
to
R
where
A1,...Ad
are non-
empty subsets of
Z
; we say
f
has size
|A1|×···×|Ad|
; if
|Ai|=n
for each
i
, it will be convenient to
DISCRETE ANALYSIS, 2024:21, 10pp. 2