RAMSEY NUMBERS OF TRAILS AND CIRCUITS

DAVID CONLON AND MYKHAYLO TYOMKYN

Abstract. We show that every two-colouring of the edges of the complete graph Kn contains a
monochromatic trail or circuit of length at least 2n2 /9+o(n2 ), which is asymptotically best possible.
A trail in a graph is a walk without repeated edges and a circuit is a closed trail, with the same
first and last vertex. The length of a trail or circuit is its number of edges. Recently, Osumi [2]
investigated Ramsey numbers for trails, proving that every two-colouring of the edges of Kn contains
a monochromatic trail of length at least n − 1, while there are two-colourings where the longest
monochromatic trail has length at most n2 /4 + o(n2 ). In this note, we improve these results.
Theorem 1. Every two-colouring of the edges of Kn contains a monochromatic circuit (and so a
trail) of length at least 2n2 /9 + O(n3/2 ) and this is asymptotically tight.
Proof. For the upper bound, consider the red/blue colouring of Kn where the red edges form a com-
plete bipartite graph between two blue cliques of orders n/3 and 2n/3, each rounded appropriately.
It is easily checked that the largest monochromatic component has size 2n2 /9 + O(n) and so the
longest monochromatic trail or circuit has length at most 2n2 /9 + O(n).
For the lower bound, suppose that we are given a red/blue colouring of the complete graph Kn .
After removing, for each colour class, a suitable forest that meets all odd degree vertices (see, for
instance, [3, Proposition 2.1]), we may assume that each colour class is Eulerian, in the sense that
every vertex has even degree in both red and blue. However, this is not immediately helpful, since
the colour classes may be disconnected.
Suppose that the largest blue component U1 has order n1 , noting that the bipartite graph between
U1 and its complement
√ U1c is, apart from the at most 2n missing edges, complete in red. We claim
that if n1 ≤ n − 2 n, then c
√ the red bipartite graph between U1 and U1 has a connected component
which includes all but n vertices of U1 . To see this, note that there are at least n1 (n − n1 ) −√ 2n
2n
red edges between U1 and U1c , so there is a vertex in U1c with degree at least n1 − n−n ≥ n 1 − n
√ 1
in U1 . Therefore, all of these at least n1 − n vertices, which we label V√1 , lie in a common red
component, as required. Moreover, this component contains at least (n1 − n)(n − n1 ) − 2n edges.
If now n1 < n/3, then all blue components have order less than n/3, so the number of red edges
is at least (1/2) · n · 2n/3 − 2n = n2 /3 + O(n) and the average red degree is at least 2n/3 + O(1).
After deleting a bounded number of vertices, we may also assume that every vertex has red degree
at least n/2, which implies that the remaining graph is connected. By [3, Theorem 1.3], which
says that any connected graph with average degree t contains a trail of length 2t + O(t), this then

implies that there is a red trail of length at least 2n2 /9 + O(n), as required. Since the component
containing this trail is Eulerian, we also have a red circuit of at least the same length.
If n/3 ≤ n1 ≤ 2n/3, then the red component containing V1 has size at least
√ √
(n1 − n)(n − n1 ) − 2n ≥ (n/3 − n)2n/3 − 2n = 2n2 /9 + O(n3/2 ).
But this component is Eulerian, so we have a circuit through all of the edges of the component,
giving the required circuit (and trail) of length at least 2n2 /9 + O(n3/2 ).

Conlon was supported by NSF Award DMS-2054452 and Tyomkyn by ERC Synergy Grant DYNASNET 810115,
the H2020-MSCA-RISE Project CoSP- GA No. 823748 and GAČR Grant 19-04113Y.
1
2 DAVID CONLON AND MYKHAYLO TYOMKYN

√
If, instead, 2n/3 < n1 ≤ n − 2 n, consider the induced graph on U1 . Ignoring √colours for now,
√
this graph has at least n21 − 2n edges. Moreover, since |V1 | ≥ n1 − n, all but 2n ≤ n of these

edges are incident
with a vertex in V1 , so that the number of edges in U1 incident with a vertex in
V1 is at least n21 + O(n). Together with the edges in the bipartite graph between V1 and U1c , we
have at least
 
n1
+ n1 (n − n1 ) + O(n3/2 ) ≥ 4n2 /9 + O(n3/2 )
2
edges. Therefore, either there are at least 2n2 /9+O(n3/2 ) edges in the red component containing V1 ,
which again completes the proof, or there are at least 2n2 /9 + O(n3/2 ) edges in the blue component
in U1 . Since this component is also Eulerian, this again
√ completes the proof.
It remains to deal with the case where n1 > n√− 2 n. By symmetry, we may also assume that
the largest red component has more than n − 2 n vertices. The number of edges which are not
√
contained within the intersection of the vertex sets of these components is at most 4 n · n = 4n3/2 ,
so the total number of edges in the intersection is at least

n2
 
n
− 4n3/2 − 2n = + O(n3/2 ).
2 2
Therefore, either the largest red or the largest blue component, both of which are again Eulerian,
contains at least n2 /4 + O(n3/2 ) edges, more than required. 

Inverting the statement of Theorem 1, we see that the Ramsey number of a trail with ℓ edges,
that is, the smallest n psuch that every
√ two-colouring of the edges of Kn contains a monochromatic
trail with ℓ edges, is 3 p ℓ/2 + o( √ℓ). Similarly, the Ramsey number for the family of all circuits
with at least ℓ edges is 3 ℓ/2 + o( ℓ). It remains an interesting question to determine the Ramsey
number for circuits of a given fixed length.
It would also be interesting to investigate the analogue of Theorem 1 for more than two colours.
It is reasonably easy to see that for every natural number k, there exists ck such that every k-
colouring of the edges of Kn contains a monochromatic circuit of length at least ck n2 + o(n2 ).
n


Indeed, if we again delete a forest for each colour class, we have a graph with 2 − kn edges where
every coloured component is Eulerian. For n sufficiently large, one of the colours in this graph has
average degree at least n/k + o(n), which, by a standard folklore result, implies that this colour has
a subgraph of minimum degree n/2k + o(n). But then there is a component in this colour with at
least n2 /8k2 + o(n2 ) edges. We note that this is also close to sharp. To see this, we note that when
k − 1 is a prime power and n is a multiple of (k − 1)2 , there is a construction of Gyárfás (see, for
example, [1]) using affine planes where every monochromatic component has order at most n/(k −1)
and, hence, at most n/(k−1) edges (or, with a more careful analysis, n2 /2k(k − 1) + O(n) edges).


2
The natural next step would be to determine the best possible constant ck . Clearly, this problem
is closely related to the question of determining the largest number of edges in a monochromatic
component in any k-colouring of the edges of Kn . In fact, the answer should be asymptotically the
same in both cases. Our arguments verify this for k = 2. For the next case, k = 3, we suspect that
the bound coming from Gyárfás’ construction, n2 /12 + O(n) edges, is correct.

References
[1] A. Gyárfás, Large monochromatic components in edge colorings of graphs: a survey, in Ramsey theory, 77–96,
Progr. Math., 285, Birkhäuser/Springer, New York, 2011.
[2] M. Osumi, Ramsey numbers of trails, preprint available at arXiv:2109.00734 [cs.DM].
[3] V. Szécsi, On the minimal length of the longest trail in a fixed edge-density graph, Cent. Eur. J. Math. 11 (2013),
1831–1837.
 RAMSEY NUMBERS OF TRAILS AND CIRCUITS 3

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

Department of Applied Mathematics, Charles University, 11800 Prague, Czech Republic