MONOCHROMATIC COMPONENTS WITH MANY EDGES

DAVID CONLON, SAMMY LUO, AND MYKHAYLO TYOMKYN

Abstract. Given an r-edge-coloring of the complete graph Kn , what is the largest number of edges
arXiv:2204.11360v2 [math.CO] 31 Aug 2022

in a monochromatic connected component? This natural question has only recently received the
attention it deserves, with work by two disjoint subsets of the authors resolving it for the first two
special cases, when r = 2 or 3. Here we introduce a general framework for studying this problem
and apply it to fully resolve the r = 4 case,
showing that any 4-edge-coloring of Kn contains a
1 n 1
monochromatic component with at least 12 2
edges, where the constant 12 is optimal only when
the coloring matches a certain construction of Gyárfás.

1. Introduction
Given an r-coloring of the edges of the complete graph Kn , how large is the largest monochromatic
connected component? A partial answer to this question was provided in 1977 by Gyárfás [6], who
showed that any such r-coloring always contains a monochromatic connected component with at
least n/(r − 1) vertices. Moreover, this estimate is best possible whenever r − 1 is a prime power
and n is a multiple of (r − 1)2 . An alternative proof of this result, as a simple corollary of his
fractional version of Ryser’s conjecture, was later found by Füredi [4, 5], who also showed that if
there is no affine plane of order r − 1, then the bound can be improved to (r − 1)n/(r2 − 2r).
There are many variants of this question. For instance, what happens when the complete graph
Kn is replaced by another graph, say a subgraph of the complete graph [8] or the complete bipartite
graph [3] of high minimum degree? Or what happens when we insist that our component has small
diameter [10]? Here we will be concerned with another variant, a rather basic one which has
received surprisingly little attention in the literature, namely, given an r-coloring of Kn , what is
the largest number of edges in a monochromatic connected component?
This question was first raised by Conlon and Tyomkyn [1] because of its close relation with
another problem, that of determining the Ramsey number for trails and circuits. However, the
components problem is arguably the more fundamental question. If we write M (n, r) for the
largest natural number such that every r-coloring of Kn contains a monochromatic connected
component with at least M (n, r) edges, then the main result of [1] may be interpreted as saying
that M (n, 2) = 29 n2 + o(n2 ). In fact, a more careful analysis of their argument implies that
M (n, 2) ≥ 91 (2n2 − n − 1), with, where divisibility allows it, the example consisting of two disjoint
red cliques of orders 2n+1
3 and n−1
3 with all blue edges between showing that this is best possible.
To say something about the general case, we first look at Gyárfás’ construction of r-colorings
where each monochromatic component has at most n/(r − 1) vertices. As noted earlier, his con-
struction, which relies on the existence of the affine plane of order r − 1, works when r − 1 is a
prime power and n is a multiple of (r − 1)2 . Concretely, the affine plane of order r − 1 corresponds
to a copy of K(r−1)2 together with r different decompositions of this graph into r − 1 vertex-disjoint
copies of Kr−1 (that is, r different Kr−1 -factors) with the property that any edge is contained in
exactly one of the r(r − 1) copies of Kr−1 . By giving the edges in the ith Kr−1 -factor color i, we

Conlon was supported by NSF Award DMS-2054452, Luo by NSF GRFP Grant DGE-1656518, and Tyomkyn by
ERC Synergy Grant DYNASNET 810115 and GAČR Grant 22-19073S.
1
obtain an r-coloring where every monochromatic component has at most r − 1 vertices. Moreover,
when n is a multiple of (r − 1)2 , we can simply blow up this coloring to obtain an r-coloring where
every monochromatic component has at most n/(r − 1) vertices.
As noted by Conlon and Tyomkyn [1], essentially the same construction works in the edge case to
show that, when r −1 is a prime
power, there are r-colorings where each monochromatic component
1 n
has at most ( r(r−1) + o(1)) 2 edges (the only caveat is that we should use each color roughly the
same number of times within each of the blown-up vertices). They also showed that this bound is
correct up to a constant and conjectured that, for r = 3, it is asymptotically tight. That this is the
case was verified by Luo [9], who proved that M (n, 3) = d 16 n2 e for n sufficiently large. Moreover,

by giving a tight lower bound for the largest number of edges

in a connected component in a graph
1 n
of given density, he was able to show that M (n, r) ≥ r2 2 in the general case, a result which was
1 n


later strengthened to M (n, r) ≥ r2 −r+ 5 2 in a revised version of the paper. Our concern here will
4
be with the following conjectural improvement to this bound.
Conjecture 1. For any natural numbers n and r with r ≥ 3,
  
1 n
M (n, r) ≥ .
r(r − 1) 2
Moreover, when there is no affine plane of order r − 1, there exists a constant εr > 0 such that
  
1 n
M (n, r) ≥ + εr .
r(r − 1) 2
The result of [9] proves this conjecture when r = 3, while the result of [1] shows that the
conjectured bound does not extend to the case r = 2. Our main result is a proof of the next open
case, when r = 4. Note that, in this case, Gyárfás’ construction corresponds to a 4-coloring of K9
where each color class is the union of three vertex-disjoint triangles. In the statement below, by
saying that a coloring matches Gyárfás’ construction, we mean that the set of components and the
intersection pattern of their vertex sets match those in this construction.
Theorem
2. In every 4-coloring of the edges
of Kn , there is a monochromatic component with at
1 n 1 n
least 12 2 edges. That is, M (n, 4) ≥ d 12 2 e. Moreover, unless the coloring

n
matches Gyárfás’s
1
construction, there is a monochromatic component with at least 12 + ε 2 edges, where ε =
2√ 1
14+ 96
− 12 > 0.0007.
Our proof of Theorem 2 consists of first showing that any 4-coloring of Kn has one of a bounded
number of component structures and then that each such component structure contains a compo-
nent with enough edges. For instance, one of the possible component structures is that each color
has precisely three components. But then one of these 12 components clearly contains at least
1/12 of the edges, as required. For the other possible component structures, our arguments are not
usually so simple, relying instead on a key observation, Proposition 3 below. This says that if a
certain union of components is large in the vertex sense, but none of these components is large in
the edge sense, then some one of the remaining components will be large in the edge sense. In fact,
even this is not quite enough and, inspired by Füredi’s approach to the vertex case, we must allow
for weighted or fractional unions of components. We will describe our general framework and how
it may be applied in more detail in the next section.

2. A general framework
Suppose r ≥ 2 and fix an r-coloring of the edges of the complete graph G = (V, E) = b Kn . For
1 ≤ i ≤ r, let Gi be the “graph of color i”, i.e., the subgraph of Kn formed by the edges with color
2
i, where we include vertices that are isolated in that color in the graph. In particular, |Gi | = n for
all i. Let Ci be the setSof connected components in Gi and let C be the set of all monochromatic
components, i.e., C = 1≤i≤r Ci . Since each vertex of G is in exactly one component of each color,
we have
X
|V (C)| = r|V (G)| = rn.
C∈C

Similarly, since each edge of G is in exactly one component of exactly one color, we have
 
X n
|E(C)| = |E(G)| = .
2
C∈C

The following observation will be central to our approach.

Proposition 3. Let X ⊆ C be a set of connected monochromatic components P in an r-coloring of

Kn and let x = |X |. Suppose ∈ ∈ + are constants such that
γ [0, r] and z R C∈X |V (C)| ≥ γn and
maxC∈C |E(C)| ≤ z n2 . Then

z(r − γ)2 ≥ max(1 − xz, 0)2 . (2.1)

P P
Proof. If γ = r, then C∈X |V (C)| ≥ rn = C∈C |V (C)|, so X = C. Then
X X |E(C)|
xz = z≥ n

= 1,
C∈C C∈C 2

so both sides of (2.1) are zero. Thus, we can assume γ < r. S

For 1 ≤ i ≤ r, let Hi be the induced subgraph of Gi on V (G) \ C∈Ci ∩X V (C). Then
r    
X X n
|E(Hi )| = |E(Kn )| − |E(C)| ≥ max (1 − xz) ,0 ,
2
i=1 C∈X

while
r
X X
|V (Hi )| = rn − |V (C)| ≤ (r − γ)n.
i=1 C∈X

Note that |V (Hi )| = 0 implies |E(Hi )| = 0. Thus, by a standard averaging argument, there is some
j with |V (Hj )| =
6 0 and
Pr  
|E(Hj )| i=1 |E(Hi )| n−1 1 − xz
≥ Pr ≥ max ,0 .
|V (Hj )| i=1 |V (Hi )| 2 r−γ

Let Cj0 be the set of components of Hj , so, by the same averaging argument, there is some C 0 ∈ Cj0
such that
P
|E(C 0 )| C∈Cj0 |E(C)|
 
|E(Hj )| n−1 1 − xz
≥P = ≥ max ,0 . (2.2)
|V (C 0 )| C∈C 0 |V (C)| |V (Hj )| 2 r−γ
j

|V (C 0 )|
Since |E(C 0 )| ≤


2 , we have

n−1 |V (C 0 )| − 1 |E(C 0 )|
≥ ≥ ,
2 2 |V (C 0 )|
3
which, combined with (2.2), implies 1−xz
r−γ ≤ 1. We therefore have

|E(C 0 )| |E(C 0 )| 2|E(C 0 )|

   
n 0 0
z ≥ |E(C )| = |V (C )| ≥ +1
2 |V (C 0 )| |V (C 0 )| |V (C 0 )|
 2|E(C 0 )|
(n − 1) max( 1−xz
  
0 +1 r−γ , 0) + 1
= |V (C )| ≥
2 2
2
 
max(1 − xz, 0) n
≥ ,
(r − γ)2 2
which rearranges to the desired bound. 
Note that taking X = ∅, we have x = γ = 0, so that z ≥ r12 , which immediately yields the simple
lower bound M (n, r) ≥ r12 n2 , as proved in Corollary 4 of [9]. For 1 − xz ≥ 0, x > 0, and 0 ≤ z ≤ 1,

(2.1) is equivalent to
p
(r − γ)2 + 2x − ((r − γ)2 + 2x)2 − 4x2 2
z≥ 2
= p . (2.3)
2x (r − γ) + 2x + ((r − γ)2 + 2x)2 − 4x2
2

1 n


In order to go beyond the bound M (n, r) ≥ r2 −r+ 5 2 from Theorem 1 of [9], we must investigate
4
the possible component structures in our r-coloring. For instance, in the case r = 3, the proof of
Theorem 2 in [9] shows that there are three possible structures for the components in our coloring.
Either:
(a) Some color has exactly one component (γ = 1, x = 1 above),
(b) Each color has exactly two components (γ = 3, x = 6), or
(c) There is a component of each color such that every vertex is covered by at least two of these
three components (γ = 2, x = 3).
Applying Proposition 3 to cases (a) and (c) yields, in each case, a lower bound on z higher than
the tight bound of 16 , while case (b), with r − γ = 0, implies 1 − xz ≤ 0, so z ≥ x1 = 16 , which is
tight.

For general r, in order to prove a lower bound of the form M (n, r) ≥ z n2 , it suffices to find, in

any given r-coloring, a set X ⊆ C yielding values of x ∈ Z+ and γ ∈ [0, r] such that x ≤ z1 and (2.1)
does not hold, that is, z(r − γ)2 ≤ (1 − xz)2 . This rearranges to
1 r−γ
x≤ − √ . (2.4)
z z
We can state our conclusions concisely as follows.
Corollary 4. Let z ∈ RP + . If there exists a set of components X ⊆ C such that, for some x ∈ Z+

and γ ∈ [0,
r], |X | = x, C∈X |V (C)| ≥ γn, and (2.4) holds, then there is a component C ∈ C with
at least z n2 edges.
In fact, a simple probabilistic argument allows us to strengthen Corollary 4 to a fractional form
allowing for non-integer values of x.
+ If there exists a function w : C → [0, 1] such that, for some x ∈ R+ and
PropositionP 5. Let z ∈ R . P
γ ∈ [0, r], C∈C w(C)
= x, C∈C w(C)|V (C)| ≥ γn, and (2.4) holds, then there is a component
C ∈ C with at least z n2 edges.
Proof. Our proof takes advantage of the fact that (2.4) is linear in x and γ. Given a function w
with the required properties, we construct a random subset X ⊆ C by taking each component C ∈ C
4
with probability w(C). Abusing notation by letting x and γ be the random variables with x = |X |
and γ = n1 C∈X |V (C)|, we have
P

r − n1 C∈C w(C)|V (C)|

  X P
r−γ
E x+ √ = w(C) + √
z z
C∈C
r−γ 1
≤x+ √ ≤ ,
z z
so there is some choice of X that satisfies the conditions of Corollary 4, yielding a component with
the desired size. 
In practice, we will apply Proposition 5 in the following form, allowing us to work with the
convenient bound (2.3).
Corollary 6. Let w : C → [0, 1], x ∈ R+ , and γ ∈ [0, r] and suppose
P
C∈C w(C) = x and
P 1
C∈C w(C)|V (C)| ≥ γn. Then (2.3) holds for z = (n) maxC∈C |E(C)|.
2
P
Note that the condition C∈C w(C)|V (C)| ≥ γn is satisfied if, for every vertex v ∈ V (G), we
have X
w(C) ≥ γ,
C3v
i.e., the function w gives a fractional cover of the vertices by components that cover each vertex
with weight at least γ. This allows us to convert our problem into aPlinear program, akin to
Füredi’s approach to the vertex case, namely, Pwe wish to minimize x = C∈C w(C) subject to the
constraints that 0 ≤ w(C) ≤ 1 for all C and C3v w(C) ≥ γ for all v ∈ V (G). In particular, to
1
p
show that z = r(r−1) works, it would suffice to show that x ≤ r(r − 1) − (r − γ) r(r − 1).

3. The case r = 4
In the case r = 4, we can investigate the set of possible component structures as in the case
r = 3, though the analysis is now considerably more intricate. Recall that our aim is to show there
1 n


exists a component with at least 12 2 edges; moreover, we would like to show that this bound
is only asymptotically tight when the components match the extremal configuration described by
Gyárfás in [7].
We will make use of the following fact shown in the course of handling the r = 3 case in [9]. An
equivalent result also appears as Lemma 4.19 in [2].
Lemma 7. In any 2-coloring of the complete bipartite graph between two vertex sets A1 and A2 ,
one of the following holds:
(a) Some color has exactly one component,
(b) Each color has exactly two components, each of which intersects both A1 and A2 , or
(c) There is one component of each color such that the intersection of their vertex sets contains
one of A1 and A2 and their union contains both.
In each of these cases, one may assign weights of 12 or 1 to the components involved to get weights
summing to at most 2 that cover every vertex involved to weight at least 1.
It will sometimes be convenient to specify further subcases within the cases of Lemma 7. We
say that a pair of vertex sets (A1 , A2 ) “satisfies case (a) for color c” if c is a color with exactly
one component covering A1 ∪ A2 . We also say that (A1 , A2 ) “satisfies case (c) directed toward
A1 ” (or, equivalently, directed away from A2 ) if A1 is contained within the intersection of the two
components.
5
 Figure 1. The cases of Lemma 7

Define the Gi and Ci as before and, for convenience, name the colors red, orange, yellow, and
blue. We begin by establishing a lower bound on maxC∈C |E(C)| in the following “degenerate” case
where some three components with distinct colors cover the entire vertex set.
Lemma 8. If, in a 4-coloring of the edges of Kn , there are three components C1 , C2 , C3 ∈ C of
distinct colors such that C1 ∪ C2 ∪ C3 = V , then
maxC∈C |E(C)| 2 1
n

≥ √ > .
2 14 + 96 12
Proof. If there is one component C1 with V (C1 ) = V , then applying Proposition 3 with X = {C1 }
and (γ, x) = (1, 1) yields, via (2.3), the bound
2 2 2
z≥ √ > > √ ,
11 + 117 22 14 + 96
as desired.
Next, suppose that there are two components C1 , C2 with C1 ∪C2 = V , but with both V1 := C1 \C2
and V2 := C2 \ C1 non-empty. Since we have not yet distinguished the colors in any way, we can
assume without loss of generality that C1 is red and C2 is orange. Then all edges between V1 and
V2 are either yellow or blue. Applying Lemma 7 to the pair of vertex sets (V1 , V2 ) yields a way to
choose weights on the yellow and blue components summing to at most 2 such that every vertex
in V1 ∪ V2 is covered by components with weights summing to at least 1. Starting from this choice
of weights and adding a weight of 1 to each of C1 and C2 then allows us to apply Corollary 6 with
(γ, x) = (2, 4), which yields the bound
2 2 2
z≥ √ > > √ ,
12 + 80 21 14 + 96
which again suffices.
The remaining case to consider is where there are three components C1 , C2 , C3 with C1 ∪C2 ∪C3 =
V , but all of V1 := C1 \ (C2 ∪ C3 ), V2 := C2 \ (C3 ∪ C1 ), and V3 := C3 \ (C1 ∪ C2 ) are non-empty.
Without loss of generality, we can assume that C1 is red, C2 is orange, and C3 is yellow. For each
pair of the Vi , only two colors are possible on the edges between them, one of which is blue. Our
aim is to apply Lemma 7 to each of these pairs and then combine the results into an appropriate
choice of weights on the components of the overall coloring of Kn .
6
 Figure 2. The vertex partition in Lemma 8

For each of the pairs (Vi , Vj ), we apply Lemma 7 to the complete bipartite graph between Vi
and Vj , yielding one of the cases (a), (b), or (c) described therein. We further split case (a) into
(a1) and (a2), depending on whether the corresponding component is blue (case (a1)) or not (some
cases are not mutually exclusive, but this will not be an issue).

Figure 3. The cases between pairs in Lemma 8

For any case constellation between the pairs (Vi , Vj ), we aim to exhibit at most two components
other than C1 , C2 , C3 covering V1 ∪ V2 ∪ V3 . Together with C1 , C2 , and C3 , this will yield a 2-cover
of the entire graph Kn by at most 5 components (since all vertices outside V1 ∪ V2 ∪ V3 are in at
least two of the Ci and so are already 2-covered). Applying Corollary 6 with (γ, x) = (2, 5) would
then give the claimed bound of
2
z≥ √ .
14 + 96
If case (a) (that is, either (a1) or (a2)) occurs for at least two pairs (Vi , Vj ), then the relevant
components readily give a cover of V1 ∪ V2 ∪ V3 . Hence, from now on we can assume that this is
not the case.
7
 Suppose case (a1) occurs on one of the pairs, say (V1 , V2 ). If (V1 , V3 ) satisfies case (b), then
V1 ∪ V2 ∪ V3 is covered by a single blue component. If (V1 , V3 ) satisfies case (c) (in either direction),
then V1 ∪ V2 is covered by a blue component B and V3 \ B is covered by a non-blue component.
Thus, we can assume from now on that case (a1) never happens.
If case (b) occurs, say on (V1 , V2 ), then one of the two remaining pairs, say (V1 , V3 ), must satisfy
case (b) or (c). If (V1 , V3 ) satisfies case (b), then V1 ∪ V2 ∪ V3 is covered by at most two blue
components. The same is true if (V1 , V3 ) satisfies case (c) directed away from V1 . Lastly, if (V1 , V3 )
satisfies (c) directed toward A1 , then V1 ∪ V2 is covered by a blue component B 0 and V3 \ B 0 is
covered by a non-blue component. Hence, we may also assume from now on that case (b) never
happens. In other words, each pair satisfies either case (a2) or (c) and the former occurs at most
once.
If all three pairs (Vi , Vj ) satisfy case (c), then the respective ‘case (c) directions’ result in an
auxiliary 3-vertex tournament, which is either cyclic or transitive. In the cyclic case we again have
that V1 ∪ V2 ∪ V3 is contained in a single blue component. In the transitive case, without loss of
generality, let each (Vi , Vj ) for i < j be directed toward Vi . Then again V1 ∪ V2 lies in a single blue
component B 00 and V3 \ B 00 can be covered by a non-blue component.
Hence, we may assume that one pair, say (V1 , V2 ), is of type (a2), with D being the non-blue
component between them, and the other two pairs are of type (c). If either of these type (c) pairs
is directed toward V3 , then V1 ∪ V2 ∪ V3 is covered by D and the corresponding blue component.
Thus, the last case to consider is when both (V1 , V3 ) and (V2 , V3 ) are directed away from V3 .
In that case, let B1 and D1 be the blue and non-blue components between (V1 , V3 ), respectively,
and define B2 and D2 similarly with respect to (V2 , V3 ). If B1 ∩ B2 ∩ V3 6= ∅, then B1 and B2
coalesce into a single component B in Kn and V3 \ B can be covered by a non-blue component. On
the other hand, if B1 ∩ B2 ∩ V3 = ∅, then

D1 ∪ D2 ⊇ (V3 \ B1 ) ∪ (V3 \ B2 ) = V3

and, since D1 and D2 cover V1 and V2 , respectively, we again obtain that two components, namely

D1 and D2 , cover V1 ∪ V2 ∪ V3 . Thus, in every case, we have a component with at least 14+2√96 n2
edges, as desired. 

The lower bound of 14+2√96 in the conclusion of Lemma 8 can be improved slightly through a
more careful analysis. However, we chose to omit this more involved proof, since the improved
estimate is not needed for the tight case in Theorem 2.
For any vertex v and any component C with v ∈ C, we can assume there is a vertex w ∈ C not
in any other component containing v; otherwise, we are done by Lemma 8. In particular, we can
assume that no component has its vertex set contained entirely within the vertex set of another
component.
For the sake of clarity, we say that two vertices v1 and v2 are joined in a given color if the edge
between them is of that color, while a set of vertices is connected in a given color if they share a
component of that color. Call two vertices v1 and v2 equivalent if they are contained in a common
component, that is, they are connected, in each color. Clearly, this is an equivalence relation, so
we may speak of equivalence classes with respect to it. Note that if v1 and v2 share exactly three
components, then, considering a vertex w outside of these three components (if there is no such
w, we are again done by Lemma 8), both v1 and v2 must be joined to w via the fourth color,
meaning that they share all four components, a contradiction. Hence, we can assume that any two
vertices are either equivalent or share at most two components. If v1 and v2 share precisely two
components, let us call them biconnected.
8
 The next

two lemmas cover a pair of general cases where we again get a lower bound of at least
2√ n
14+ 96 2
edges.

Lemma 9. If there is a pair of components of different colors that do not intersect, then there is
2√ n


a monochromatic component with at least 14+ 96 2 edges.

Proof. By symmetry, we can assume that the two disjoint components R and B are red and blue,
respectively. All edges between R and B are orange or yellow, so we can apply Lemma 7 to the
complete bipartite graph GR,B between them. If either R or B is contained entirely in a single
orange or yellow component in GR,B and thus in G, we are done by Lemma 8, so assume otherwise.
Then only case (b) of Lemma 7 can apply, where there are exactly two orange components and
two yellow components in GR,B (none of which can coalesce together in G). It is easy to see that
in this case any two vertices in R that share an orange component also share a yellow component
and are therefore equivalent. Hence, there are exactly two equivalence classes of vertices in R and,
likewise, exactly two equivalence classes in B. Taking representatives v1 , v2 ∈ R and w1 , w2 ∈ B of
these equivalence classes, let

X = {C ∈ C : C ∩ {v1 , v2 , w1 , w2 } =
6 ∅} \ {R, B}.

Define w(C) = 1 when C ∈ X and w(C) = 0 otherwise. We claim that this choice of w satisfies
the conditions of Corollary 6 with (γ, x) = (3, 8). If {v1 , v2 } ⊆ C for any component C 6= R, then
R ⊆ C, a contradiction, so v1 and v2 do not share any non-red components; likewise, w1 and w2
do Pnot share any non-blue components. Thus, X contains exactly two components P of each color,
so C∈C w(C) = |X| = 8. Moreover, by construction, for u ∈ R ∪ B, we have C3u w(C) = 3.
Now consider a fixed u ∈/ R ∪ B. Since the two equivalence classes in R do not share any non-red
components, there are at least two colors among the edges between u and R and, similarly, at
least two colors among the edges between u and B. If there are exactly two colors among the
edges between u and R ∪ B, then these colors must be orange and yellow, a contradiction because
R ∪ B cannot be covered by the union of an orange component and a yellow P one. So there are at
least three colors among the edges between u and R ∪ B, which means C3u w(C) ≥ 3. Thus,
we can apply Corollary

6 with (γ, x) = (3, 8), yielding a monochromatic component with at least
2√ n 2√ n


17+ 33 2
> 14+ 96 2 edges, as desired. 

Lemma 10. If there
is a pair of biconnected vertices, then there is a monochromatic component
with at least 14+2√96 n2 edges.

Proof. Suppose that v1 and v2 are biconnected. By symmetry, we can assume that the two com-
ponents they share are red and orange and that there are at least as many orange components as
red components in our coloring. We can also assume that there are at least two red components;
otherwise, we are done by Lemma 8. If there are exactly two orange components, then applying
Proposition 3 to the set of red and orange components with (γ, x) = (2, 4) yields z ≥ 12+2√80 and
we are again done, so we can assume that there are at least three orange components. Let R and
O be the joint red and orange components, respectively, of v1 and v2 and, for i ∈ {1, 2}, let the
yellow and blue components containing vi be Yi and Bi , respectively. Let T = V \ R.
Since {R, O, Y1 , B1 } are the four components of v1 , their union is the whole of V . Hence,
D := B1 \ (R ∪ O) 6= ∅, as otherwise {R, O, Y1 } would satisfy the assumptions of Lemma 8 and,
analogously, E := Y1 \ (R ∪ O) 6= ∅. Observe that v2 must be joined in yellow to all vertices in D
and in blue to all vertices in E. This means that D ∩ E = ∅ and, consequently, v1 is joined to all
of D in blue and to all of E in yellow. Furthermore, E = B2 \ (R ∪ O) and D = Y2 \ (R ∪ O).
9
 Figure 4. The situation in Lemma 10

If there is a vertex w ∈ Y1 ∩ B1 ∩ R \ O, then w shares exactly three components with v1 , a

contradiction. Hence, we may conclude that
Y1 ∩ B1 \ O = Y1 ∩ B1 \ (R ∪ O) = D ∩ E = ∅ (3.1)
and, similarly, Y2 ∩ B2 \ O = ∅.
Now consider a vertex v3 ∈ R \ O. If v3 is joined in orange to all of T \ O = D ∪ E, then
either there are only two orange components (O and the component of v3 ) or there is an orange
component whose vertex set is contained within R; either case is a contradiction. Therefore, v3 is
joined in blue or yellow to some vertex in T \ O, which means it is in one of B1 , B2 , Y1 , Y2 . By
symmetry, we can assume it is in B1 , so by (3.1) we have v3 ∈ / Y1 . We conclude that v3 must be
joined to all of E in orange (see Figure 4). Since v3 was chosen from R \ O arbitrarily, every vertex
in R \ O is joined either to all of D or to all of E in orange. If any orange component does not
intersect R, we are done by Lemma 9. Otherwise, we have at most three orange components in
total, so by our earlier assumption it must be exactly three. In particular, there will be a vertex
in R \ O connected in orange to all of D and none of E. It follows that every edge between D and
E must be red, so all of D ∪ E = V \ (R ∪ O) is contained in a single red component. If there is a
third red component, it must lie entirely inside O, so we are done by Lemma 8. Therefore, we can
assume there are exactly two red components.
Next, note that v1 and v3 share components in red and blue but not in orange, which means v1
and v3 are biconnected. Therefore, by the same argument as above (with orange and blue swapped),
the total number of blue components is also three. Let B3 be the third blue component and note
that B3 ⊆ R ∪ O. Consequently, we must have B3 ∩ R \ O 6= ∅, as otherwise B3 would be contained
in O. Let v4 ∈ B3 ∩ R \ O be an arbitrary vertex. By using similar reasoning as for v3 , we have
that v4 is in one of B1 , B2 , Y1 , Y2 ; since v4 ∈ B3 , it must be in one of Y1 and Y2 . We conclude, as
before for v3 and blue, that there are exactly three yellow components.
Thus, we have exactly two red components and exactly three components of every other color,
for a total of exactly 11 monochromatic components. By the pigeonhole principle, one of these
1 n 2√ n



components must have at least 11 2 > 14+ 96 2
edges, as needed. 

We are now ready to complete the proof of our main theorem.

Proof of Theorem
2. If the hypotheses of any of Lemmas 8, 9, or 10 hold, we have a component with
at least 14+2√96 n2 edges, as needed, so we can assume otherwise. Thus, every pair of components
of different colors intersects and any pair of vertices sharing at least two components are equivalent.
10
 We claim that in this case there are at most three components of each color. Without loss of
generality, let red be a color with the fewest components. Let R be a red component and fix a
vertex v ∈/ R. Let O, Y , and B be the orange, yellow, and blue components of v, respectively. By
Lemma 10, we can assume that all vertices in R ∩ O are equivalent and similarly for R ∩ Y and
R ∩ B. However, O ∪ B ∪ Y ⊇ R, as every vertex in R is adjacent to v via a non-red edge. So the
vertices of R can be partitioned into at most three equivalence classes. Thus, there are at most
three components of each color that intersect V (R) and since, by Lemma 9, we can assume that R
intersects every non-red component, there are at most three components of each color in total, as
claimed.

fewer than three components, then |C| ≤ 11 and there is some compo-
If there is a color with
1 n
nent with at least 11 2 edges. Otherwise, we will show that our coloring matches the extremal
construction claimed. Indeed, by assumption, the equivalence class of a vertex is determined by its
red and orange components. Let R = R1 , R2 , R3 be the red components and O1 , O2 , O3 the orange
components and let Vij = Ri ∩ Oj for 1 ≤ i, j ≤ 3. The Vij form a partition of V (G) and each is an
equivalence class of vertices. Since we can assume that every pair of components of different colors
intersects, the Vij are all non-empty.
Every component contains exactly three equivalence classes: one for each orange component if it
is red and one for each red component if it is not red. Each pair of the three vertex sets V11 , V22 , V33
must share at least one component, each of which must be yellow or blue; one of the two colors
occurs at least twice, say yellow. Then V11 ∪ V22 ∪ V33 is connected in yellow and in fact must form
a yellow component, since each yellow component contains exactly three equivalence classes. Then
neither V13 nor V31 shares a red, orange, or yellow component with V22 , so V13 ∪ V22 ∪ V31 forms a
blue component and, similarly, so do V12 ∪ V21 ∪ V33 and V11 ∪ V23 ∪ V32 . Repeating this argument
shows that V12 ∪ V23 ∪ V31 forms another yellow component, as does V13 ∪ V21 ∪ V32 . Hence, we are
in exactly the extremal configuration claimed.
Thus, only the claimed extremal configuration can have fewer than 14+2√96 n2 edges in every

component and in this
extremal configuration, since there are 12 components, we instead get a
1 n
lower bound of 12 2 edges, as desired. 

As in [1] and [9], it is possible to amend our argument to show that every 4-coloring of Kn
1
contains a monochromatic trail or circuit of length at least ( 24 + o(1))n2 . We omit the details, but
briefly note the main idea, which is to delete a sparse subgraph in order to guarantee that each
component is Eulerian and then work around these omitted edges.

References
[1] Conlon, D., and Tyomkyn, M. Ramsey numbers of trails and circuits. Preprint available at arXiv:2109.02633
[math.CO] .
[2] DeBiasio, L., Kamel, Y., McCourt, G., and Sheats, H. Generalizations and strengthenings of Ryser’s
conjecture. Electron. J. Combin. 28, 4 (2021), Paper No. 4.37.
[3] DeBiasio, L., Krueger, R. A., and Sárközy, G. N. Large monochromatic components in multicolored
bipartite graphs. J. Graph Theory 94, 1 (2020), 117–130.
[4] Füredi, Z. Maximum degree and fractional matchings in uniform hypergraphs. Combinatorica 1, 2 (1981),
155–162.
[5] Füredi, Z. Covering the complete graph by partitions. Discrete Math. 75, 1-3 (1989), 217–226.
[6] Gyárfás, A. Partition coverings and blocking sets in hypergraphs. Communications of the Computer and
Automation Institute of the Hungarian Academy of Sciences 71 (1977), 62 pp.
[7] Gyárfás, A. Large monochromatic components in edge colorings of graphs: a survey. In Ramsey theory, vol. 285
of Progr. Math. Birkhäuser/Springer, New York, 2011, pp. 77–96.
11
 [8] Gyárfás, A., and Sárközy, G. N. Large monochromatic components in edge colored graphs with a minimum
degree condition. Electron. J. Combin. 24, 3 (2017), Paper No. 3.54.
[9] Luo, S. On connected components with many edges. Preprint available at arXiv:2111.13342 [math.CO] .
[10] Ruszinkó, M. Large components in r-edge-colorings of Kn have diameter at most five. J. Graph Theory 69, 3
(2012), 337–340.

Department of Mathematics, California Institute of Technology, Pasadena, CA 91125, USA

Email address: [email protected]

Department of Mathematics, Stanford University, Stanford, CA 94305, USA

Email address: [email protected]

Department of Applied Mathematics, Faculty of Mathematics and Physics, Charles University,

11800 Prague, Czech Republic
Email address: [email protected]

12