the ground / stratum · Pattern
Take any six people. Join every pair with a line: red if they know each other, blue if they're strangers. No matter how the world arranged those acquaintances, you are now guaranteed three people who all know each other, or three who are all strangers. You cannot draw it otherwise. That is not a fact about people — it is a fact about six, and it is the smallest visible edge of a law: in any large enough structure, order you did not put there appears anyway.
In 1930 a Cambridge philosopher named Frank Ramsey — who would die that same year at twenty-six — proved a theorem he needed only as a stepping-stone in mathematical logic. It turned out to describe something far stranger: that complete disorder is impossible. Make any structure big enough and you cannot keep it patternless; some monochromatic island always forms. The party problem is its friendliest face. Colour the friendships below and try, if you can, to avoid the trio.
Each edge is a pair of people: red = acquainted, blue = strangers. Click any edge to flip its colour. A monochromatic triangle — three people all red or all blue with each other — lights up the instant it appears. With 6 people, try to reach zero. You can't. Drop to 5 and there is exactly one way out.
Goodman proved the best you can ever do on six people is two such triangles — never fewer, never zero. The order is structural, not bad luck.
Why must six be enough? Here is the whole proof, and it fits in a breath. Pick any one person. They have five relationships, each red or blue; by the pigeonhole principle at least three are the same colour — say three red acquaintances, a, b, c. Now look at the three edges among a, b, c. If even one is red, it closes a red triangle with our person. If none is red, then a, b, c are mutual strangers — a blue triangle. Either way, the trio exists. (Press “Pigeonhole proof” above to watch it run on the current colouring.)
That argument is airtight, but a machine can also simply check every possibility. Six people have 15 pairs, each red or blue: 2¹⁵ = 32,768 possible worlds in all. Run the search — every one of them contains a trio.
Enumerate every two-colouring of the fifteen edges and count, for each, how many monochromatic triangles it holds. If even one colouring reached zero, R(3,3) would be greater than 6. None does.
x-axis: number of monochromatic triangles in a colouring · y-axis: how many of the 32,768 colourings have that many. The bar at the far left is the minimum — and it sits at 2, not 0.
Six is the answer to the simplest question: how many people force a monochromatic triangle (a group of three)? The number is written R(3,3) = 6 — Ramsey's number for “3 and 3.” But you can ask it for any sizes. R(s,t) is the fewest people that force either s mutual friends or t mutual strangers. Ramsey's theorem says every one of these numbers is finite — order is always eventually forced. Knowing the actual value is another matter entirely.
Step up by one. How many people force a monochromatic group of four? The answer is R(4,4) = 18, proved by Greenwood and Gleason in 1955. The lower half — that 17 people are not enough — has a jewel of a witness: the Paley graph on 17 points. Place 17 people in a circle; make two of them acquainted exactly when the gap between them (mod 17) is a perfect square — 1, 2, 4, 8, 9, 13, 15, or 16. That single rule produces a colouring with no monochromatic group of four at all. The instrument below builds it and checks it live.
Now take one more step, to a group of five — and walk straight off the edge of human knowledge.
The diagonal Ramsey numbers R(k,k), on a logarithmic scale. Two are known exactly; from five up, no one on Earth knows the value — only that it lies somewhere in the hatched band.
Left: the 17-vertex graph is built from the “square gaps” rule and searched for any group of four that's all-connected or all-disconnected — none exists. Right: this is one of Geoffrey Exoo's actual 1989 graphs on 42 people (from Brendan McKay's collection), searched for any monochromatic group of five — none exists, so 42 people are not enough and R(5,5) ≥ 43.
That last gap is the famous one. We know R(5,5) is at least 43 (Exoo's graph proves it) and at most 46 (a 2024 computer-assisted proof by Angeltveit and McKay, whittled down from 49). Between those, nothing. Most experts believe the true value is exactly 43 — in 1997 McKay, Radziszowski and Exoo found 656 distinct 42-person graphs and could extend none of them — but believing is not proving, and the search space is past astronomical. The next number, R(6,6), is known only to lie between 102 and 160; its lower bound hasn't moved since 1965.
Suppose aliens invade the earth and threaten to obliterate it in a year's time unless human beings can find the Ramsey number for red five and blue five. We could marshal the world's best minds and fastest computers, and within a year we could probably calculate the value. If the aliens demanded the Ramsey number for red six and blue six, however, we would have no choice but to launch a preemptive attack.Paul Erdős, as recounted by Graham & Spencer, Scientific American, 1990
The joke is exact arithmetic. Each step of R multiplies the colourings to sift by an unspeakable factor; the proofs that pinned R(4,5)=25 and bounded R(5,5) already lean on the world's graph-generation software. And yet Ramsey's theorem promises every one of these numbers is a definite, finite integer, sitting there, waiting. We just can't reach most of them. The asymptotics are barely better: for seventy-five years the best bounds on R(k,k) were trapped between roughly √2ᵏ and 4ᵏ — until 2023, when Campos, Griffiths, Morris and Sahasrabudhe finally pushed the upper bound below 4ᵏ for the first time, to about 3.993ᵏ. A breakthrough — and still exponentially far from the truth.
Ramsey's theorem guarantees the answer exists. It says nothing about whether you'll ever know it.
Recomputed live, and checked offline first. The trio-detection in Instrument I, the full enumeration of all 2¹⁵ colourings in Instrument II, the Paley-graph construction, and the search for monochromatic cliques in both witness graphs (Instrument III) all run in your browser. They are independently verified in /research/ramsey-numbers/verify.mjs (11/11): the C₅ escape; the exhaustive proof that all 32,768 colourings of K₆ contain a trio (none escape; minimum 2); the Paley-17 graph shown to have clique and independence number 3; and all 328 of Exoo's published 42-vertex graphs confirmed to contain no monochromatic K₅ (over every one of the 850,668 five-subsets), proving R(5,5) ≥ 43.
Cited, not re-proved. The upper bounds — that 6, 18, 25 people are enough (R(3,3)≤6 we do exhaust; R(4,4)=18, R(4,5)=25 are the classical proofs) and that R(5,5) ≤ 46 — are the published theorems (Greenwood–Gleason 1955; McKay–Radziszowski 1995; Angeltveit–McKay 2017/2024). The 656-graph conjecture R(5,5)=43 and the R(6,6) range are cited from the literature, with sources below.
The lower-bound graphs prove a strict inequality — 42 people are not enough, so R(5,5) ≥ 43. They do not prove the value is 43; that 43 actually suffices is conjecture, not theorem. “Complete disorder is impossible” means a monochromatic clique of any fixed size eventually becomes unavoidable — it does not say structure appears everywhere, or that we can compute when. Ramsey theory is a guarantee of existence, frequently with no way to find the witness. That gap between exists and known is the whole drama here, and we keep it loud.
Sources. F. P. Ramsey, “On a problem of formal logic,” Proc. London Math. Soc. s2-30:264–286, 1930. R. E. Greenwood & A. M. Gleason, “Combinatorial relations and chromatic graphs,” Canad. J. Math. 7:1–7, 1955 (R(3,3)=6, R(3,4)=9, R(3,5)=14, R(4,4)=18; the Paley/quadratic-residue colourings). G. Exoo, “A lower bound for R(5,5),” J. Graph Theory 13:97–98, 1989. B. D. McKay & S. P. Radziszowski, “R(4,5)=25,” J. Graph Theory 19:309–322, 1995, and their work conjecturing R(5,5)=43 (with Exoo, 1997). V. Angeltveit & B. D. McKay, “R(5,5)≤48,” arXiv 2017, and the improvement to 46, arXiv:2409.15709, 2024. The 42-vertex graphs: B. D. McKay, “Ramsey Graphs” data collection (r55_42some.g6). S. Radziszowski, “Small Ramsey Numbers,” Electronic J. Combinatorics dynamic survey DS1 (the authoritative table). The Erdős aliens quotation: R. L. Graham & J. H. Spencer, “Ramsey Theory,” Scientific American 263(1):112–117, July 1990. A. W. Goodman, “On sets of acquaintances and strangers at any party,” Amer. Math. Monthly 66:778–783, 1959 (the minimum-2 count). M. Campos, S. Griffiths, R. Morris & J. Sahasrabudhe, “An exponential improvement for diagonal Ramsey,” arXiv 2023 / Annals of Mathematics 2026 (R(k,k) ≤ (4−ε)ᵏ). The “complete disorder is impossible” epigram is attributed to T. S. Motzkin.