The bound rounds up
20 objects into 6 boxes guarantees a box with 4, not 3. Rounding down gives a claim that is true but weaker than what holds.
With n objects in k boxes, some box holds at least ⌈n/k⌉. Sets, Relations, and Functions stated it; here it is applied to monotone subsequences and to the Ramsey number R(3,3) = 6.
Core ideas
The principle is elementary. The theorems it proves are not, and the gap between the two is the point of this lesson.
20 objects into 6 boxes guarantees a box with 4, not 3. Rounding down gives a claim that is true but weaker than what holds.
In Erdős–Szekeres the boxes are pairs of subsequence lengths, which is not a classification anyone would guess.
Six people always contain three mutual acquaintances or three mutual strangers. No probability is involved.
The material
Statement, then two theorems whose proofs are pigeonhole with a clever classification.
If n objects are placed into k boxes, some box contains at least ⌈n/k⌉ objects.
If every box held at most ⌈n/k⌉ − 1, the total would be at most k(⌈n/k⌉ − 1) < k(n/k + 1 − 1) = n, using ⌈n/k⌉ < n/k + 1. That contradicts having placed all n.
□
Among 100 people, at least ⌈100/12⌉ = 9 share a birth month. In any set of 10 integers, at least ⌈10/2⌉ = 5 share a parity. Each is one division.
Every sequence of n² + 1 distinct real numbers contains a monotone subsequence of length n + 1 — increasing or decreasing.
For each position i, let aᵢ be the length of the longest increasing subsequence ending at i, and bᵢ the length of the longest decreasing one ending at i.
Suppose no monotone subsequence of length n+1 exists. Then every aᵢ and every bᵢ lies in {1, …, n}, so the pair (aᵢ, bᵢ) takes one of n² values. With n² + 1 positions, two positions i < j share a pair.
But that is impossible. If xᵢ < xⱼ then the increasing subsequence ending at i extends to j, so aⱼ > aᵢ. If xᵢ > xⱼ then bⱼ > bᵢ. Either way the pairs differ.
□
The boxes are the n² possible pairs (aᵢ, bᵢ), and inventing them is the entire proof. Once the classification is chosen, the pigeonhole principle does the rest in one line. This is the lesson's real content: the principle is trivial and the classification is not.
Among any six people, three are mutual acquaintances or three are mutual strangers. Five people do not suffice.
Take a person P. Each of the other five is an acquaintance or a stranger, so by pigeonhole at least ⌈5/2⌉ = 3 fall in one class; say three acquaintances A, B, C.
If any two of A, B, C know each other, they and P are three mutual acquaintances. If no two of them do, A, B, C are three mutual strangers.
For five people it can fail: arrange them in a circle where each knows exactly the two neighbours. That has no triangle of either kind.
□
Both halves are needed. The construction on five people is what makes the theorem say R(3,3) = 6 rather than R(3,3) ≤ 6, and producing such constructions is generally much harder than the pigeonhole bound.
Ramsey numbers grow ferociously and are almost entirely unknown. R(4,4) = 18; R(5,5) is only known to lie between 43 and 46, despite decades of effort. Erdős's remark is the standard one: if aliens demanded R(5,5) we should marshal all our computers, and if they demanded R(6,6) we should attack them instead.
Interactive
Enumerated while enumeration is possible
Method
Objects are usually given; boxes are invented.
The people, the numbers, the positions in the sequence. Usually explicit in the problem.
The hard step. Remainders, parities, degree counts, pairs of subsequence lengths — whatever gives fewer boxes than objects.
Round up. This is the guaranteed occupancy, and rounding down weakens the conclusion for no reason.
Two objects in one box must yield the conclusion. If it does not, the classification was the wrong one.
Given any n integers, some consecutive block sums to a multiple of n.
The classification is "remainder mod n", and case 1 is what reduces the box count from n to n − 1 — the same move as the acquaintance-count argument in Sets, Relations, and Functions. Creating a surplus by ruling out one box is the recurring trick.
Show that among any 5 points in a unit square, two are within √2 / 2 of each other — then improve the bound by choosing a finer partition and see what it costs you in the object count.
Something to change about this lesson? Write it down, tick it off when it is handled, and export the list when you want to act on it. Kept in this browser only: nothing is sent anywhere, and nobody is notified.
No recommendations yet for this lesson.