← Back to the introduction

Known numbers / Literature and our study

What is known.
What we computed.

First, exact values for ordinary two-colour Ramsey numbers. Then, the circulant values reported in our paper and the computational argument establishing them.

01 / Classical two-colour numbers

Known exact values.

The nine nontrivial exact values listed in Table Ia of Radziszowski’s survey, up to exchanging the two parameters. The elementary family R(2,n) = n is omitted; R(m,n) = R(n,m).

Source: Stanisław P. Radziszowski, Small Ramsey Numbers, The Electronic Journal of Combinatorics, Dynamic Survey DS1, revision 18, 24 April 2026, §2.1, Table Ia. Read the survey table ↗

Known nontrivial exact two-colour Ramsey numbers in survey revision 18
Ramsey numberExact value
R(3,3)6
R(3,4)9
R(3,5)14
R(3,6)18
R(3,7)23
R(3,8)28
R(3,9)36
R(4,5)25

Some small cases are still open.

Selected intervals from the same survey revision; these are bounds, not exact values.

40 ≤ R(3,10) ≤ 41

47 ≤ R(3,11) ≤ 50

36 ≤ R(4,6) ≤ 40

49 ≤ R(4,7) ≤ 58

59 ≤ R(4,8) ≤ 79

43 ≤ R(5,5) ≤ 46

59 ≤ R(5,6) ≤ 85

102 ≤ R(6,6) ≤ 160

205 ≤ R(7,7) ≤ 492

See Tables Ia and Ib. This is a dated reference snapshot, checked on 16 October 2026.

02 / The circulant family

Values reported in our study.

We write RC(m,n) for one more than the largest order admitting a circulant colouring with no blue Km and no red Kn. Thus RC(m,n) ≤ R(m,n): an exact circulant value can still be only a lower bound for the ordinary Ramsey number.

Same question, a restricted family.

R(3,7) = 23, whereas RC(3,7) = 22. Our 21-vertex circulant witness proves the lower bound 22. It does not imply that the ordinary Ramsey number is 22.

The tables contain the 26 values in the paper’s circulant study. Six are classical baseline values; our computational procedure establishes the other 20. Eight values, RC(3,n) for n = 13,…,20, are new in this study. The other twelve computational values reproduce earlier results with an explicit reproducible procedure.

Study source: An Integer Programming Approach to Compute Lower Bounds for Ramsey Numbers Using Circulant Graphs, “Computational Study on Circulant Ramsey Numbers”; certificate collection and interpretation ↗. Click a witness to open its graph on a web page. The source certificate is also available to download. The graph certifies the constructive lower-bound side; the matching exclusions are computational results of the study.

The ordinary R(m,n) column gives both lower and upper bounds, or an exact value when known, from Radziszowski’s revision 18, Tables Ia, Ib and IIa. For example, RC(3,13) = 58, while 61 ≤ R(3,13) ≤ 68. See the higher-parameter survey table ↗

RC(3,n)

Circulant Ramsey numbers with first parameter 3 reported in the paper
Circulant numberExact RCOrdinary R(m,n)
Lower–upper, or exact
Witness graph
RC(3,3)Classical66Exact value
RC(3,4)Classical99Exact value
RC(3,5)Classical1414Exact value
RC(3,6)Classical1718Exact value
RC(3,7)Reproduced in this study2223Exact value
RC(3,8)Reproduced in this study2728Exact value
RC(3,9)Reproduced in this study3636Exact value
RC(3,10)Reproduced in this study3940–41Lower–upper bounds
RC(3,11)Reproduced in this study4647–50Lower–upper bounds
RC(3,12)Reproduced in this study4953–59Lower–upper bounds
RC(3,13)New in this study5861–68Lower–upper bounds
RC(3,14)New in this study6467–77Lower–upper bounds
RC(3,15)New in this study7374–87Lower–upper bounds
RC(3,16)New in this study7982–97Lower–upper bounds
RC(3,17)New in this study9292–109Lower–upper bounds
RC(3,18)New in this study98100–120Lower–upper bounds
RC(3,19)New in this study106106–132Lower–upper bounds
RC(3,20)New in this study111111–145Lower–upper bounds

RC(4,n)

Circulant Ramsey numbers with first parameter 4 reported in the paper
Circulant numberExact RCOrdinary R(m,n)
Lower–upper, or exact
Witness graph
RC(4,4)Classical1818Exact value
RC(4,5)Classical2525Exact value
RC(4,6)Reproduced in this study3436–40Lower–upper bounds
RC(4,7)Reproduced in this study4749–58Lower–upper bounds
RC(4,8)Reproduced in this study5259–79Lower–upper bounds
RC(4,9)Reproduced in this study6973–105Lower–upper bounds

RC(5,n)

Circulant Ramsey numbers with first parameter 5 reported in the paper
Circulant numberExact RCOrdinary R(m,n)
Lower–upper, or exact
Witness graph
RC(5,5)Reproduced in this study4243–46Lower–upper bounds
RC(5,6)Reproduced in this study5759–85Lower–upper bounds

03 / How the values were obtained

A witness and the exclusions above it.

For each pair (m,n), a known ordinary Ramsey upper bound u excludes every order at least u. A known circulant witness on ℓ−1 vertices gives a starting lower bound ℓ. We then investigate the remaining orders from the top down.

  1. Encode an order using distance variables.At order t, the distance-space integer program chooses the colours of ⌊t/2⌋ circular-distance classes. Branch-and-cut searches for a colouring avoiding both forbidden clique sizes.
  2. Resolve the clique questions.Target-clique heuristics search for violations; exact clique separation resolves the remaining questions and generates valid cuts. An infeasibility result must come from a completed exact solve.
  3. Inspect orders u−1, u−2, …, ℓ.Each order is either proved infeasible, produces a verified colouring, or remains unresolved at the time limit. The circulant study uses a 24-hour limit and one thread per instance.
  4. Join the two sides of the proof.If a verified colouring is first found at order t after every larger relevant order has been excluded, then t is the largest feasible circulant order and RC(m,n) = t+1. If all tested orders are excluded, the starting witness gives RC(m,n) = ℓ. A timeout leaves the exact value unresolved.

Why one infeasible order is not enough

Circulant feasibility need not be monotone in the graph order. A colouring can fail to exist at one order and exist at a larger one. To establish RC, all larger orders below u must therefore be ruled out; stopping at the first failure would not prove the claimed value.

The downloadable graph certificates independently support the lower-bound side. They do not encode the exhaustive exclusions needed for equality. Those exclusions use the paper’s formulation and completed solver runs; the repository provides the computational procedure.