QiaoShen
← BMO Preparation details

Combinatorics & Games

Start topic practice →

Topic overview

Combinatorics and games appear extremely frequently in BMO proof questions. The main weapons are: **the pigeonhole principle** (the key lies in designing the boxes), **the extremal principle** (take the maximal/minimal configuration and then tweak it), **invariants and monovariants** (find a conserved quantity to prove unreachability, find a monotone quantity to prove termination), **double counting**, **induction and construction** (divide and conquer, insertion, stripping), as well as **Ramsey / Erdős–Szekeres**-type existence results. The mnemonic for the approach: **"pick the right boxes, take extremes, find the conserved quantity"**.

Key concepts

Putting n+1n+1 objects into nn boxes, there must be one box with ≥2\ge2. The difficulty always lies in designing the boxes:

  • By remainders: among n+1n+1 integers there must be two that are congruent  mod n\bmod n (their difference is divisible by nn).
  • By odd part: taking n+1n+1 out of {1,…,2n}\{1,\dots,2n\}, one must divide another (write m=2a⋅(odd)m=2^a\cdot(\text{odd}), the odd part has only nn kinds).
  • By odd/even lattice points (Z/2)2(\mathbb Z/2)^2: among any 5 lattice points there must be two whose midpoint is a lattice point.
  • Region splitting: a unit square divided into 4 small squares ⇒ among 5 points there must be two at distance ≤22\le\tfrac{\sqrt2}{2}.

After using the pigeonhole principle, don't forget a tightness construction to show the bound cannot be improved.

Key points

  • ◆Common dimensions for designing pigeonholes: remainders, greatest odd factor, parity of coordinates, small geometric regions
  • ◆"Number of objects > number of values" type: e.g. number of subsets > number of possible sums ⇒ two subsets have equal sum
  • ◆A tightness construction (such as $\{n+1,\dots,2n\}$) proves the bound is optimal

Choose an object that maximises or minimises a quantity, then prove it has the required property: otherwise a small adjustment would improve it, giving a contradiction.

For example, a maximum cut partitions a graph's vertices into two groups with as many crossing edges as possible. If a vertex had more neighbours in its own group than in the opposite group, moving it would increase the number of crossing edges, a contradiction. Thus each person has a number of friends in the opposite group ≥\ge the number in their own group. This also proves that a maximum cut of any graph contains ≥\ge half its edges.

Key points

  • ◆'Take an extreme + tweak to a contradiction' is a universal opening move for existence/construction problems
  • ◆Often paired with "move one element / swap two elements" as fine-tuning
  • ◆Maximum cut, longest path and minimal counterexample all belong to this class

Invariant: find a quantity that is conserved under each step of the operation (often some modulus), and use it to prove that 'the target state is unreachable' (numbers changed on a blackboard, coin flipping, parity of a puzzle).

Monovariant: find a quantity that is strictly monotone and bounded at each step, and prove that the process must terminate (e.g. regard the states of a row of coins as a binary number, which each operation strictly decreases).

Key points

  • ◆To prove "unreachable", find an invariant (mod 2 / mod 3 / colouring sum)
  • ◆To prove "termination / finitely many steps", find a monotone quantity (a non-negative integer cannot decrease indefinitely)
  • ◆First ask: what is invariant under the operation? What changes monotonically?

Count the same quantity in two ways and set them equal. In graph theory, ∑vdeg⁡(v)=2∣E∣\sum_v\deg(v)=2|E| (the handshaking lemma).

Application: Mantel's theorem — the number of edges in a triangle-free graph is ≤⌊n2/4⌋\le\lfloor n^2/4\rfloor. Proof: triangle-free ⇒ the two endpoints of every edge are deg⁡(u)+deg⁡(v)≤n\deg(u)+\deg(v)\le n; summing over all edges gives ∑deg⁡2≤n∣E∣\sum\deg^2\le n|E|, then by Cauchy–Schwarz ∑deg⁡2≥(2∣E∣)2n\sum\deg^2\ge\tfrac{(2|E|)^2}{n}, and combining gives ∣E∣≤n2/4|E|\le n^2/4.

Key points

  • ◆"One quantity, two ways of counting" = double counting
  • ◆$\sum\deg=2|E|$; the number of odd-degree vertices is even
  • ◆Local constraints (inequalities on the edges) → sum → close with Cauchy–Schwarz

Divide-and-conquer induction: split the 2n+12^{n+1} problem into 2n2^n subproblems — e.g. an 2n×2n2^n\times2^n board with one square missing can be tiled by L-shaped trominoes (place one in the centre, creating a gap in each of the other three quadrants); nn lines divide the plane into 1+n+(n2)1+n+\binom n2 regions (the kkth line adds kk regions).

Insertion method: insert the new element into an already ordered sequence — every tournament has a Hamiltonian path (walk along the chain and insert at the point where the win/loss flips).

Stripping method: find a pair of adjacent targets, join them, strip them off, and induct — 2n2n points on a circle with nn red and nn blue can be joined by nn non-crossing red-blue chords.

Key points

  • ◆Divide and conquer: the difficulty lies in 'making every subproblem satisfy the induction hypothesis' (often a bridging piece is needed)
  • ◆Incremental counting: work out 'how much is added by including the $k$-th object', then sum
  • ◆Insert / strip the existence used for 'linearisable / matchable'

Ramsey R(3,3)=6R(3,3)=6: among 66 people there must be 3 who know each other or 3 who do not know each other (use pigeonhole on one person to force 3 edges of the same colour, then look inside these 3 people). 55 people is not enough (pentagon + pentagram counterexample).

Erdős–Szekeres: n2+1n^2+1 distinct real numbers must contain a monotone subsequence of length n+1n+1 (label each term with (xi,yi)=(x_i,y_i)= (the longest increasing/decreasing length ending at ii), pigeonhole + local extendability).

Happy Ending: 5 points in general position must contain a convex quadrilateral (classified by convex hull).

Key points

  • ◆Ramsey: first use the pigeonhole principle to force a monochromatic neighbourhood, then classify the interior
  • ◆"Attach coordinate labels + pigeonhole" is the Erdős–Szekeres paradigm
  • ◆Combinatorial geometry existence → first classify by the shape of the convex hull

Worked examples

Try each example yourself before checking the explanation.

1Example 1: Ramsey $R(3,3)$ — six people at a party

At a party there are 6 people, and any two of them are either "acquainted" or "strangers". Prove that there must be three people who are pairwise acquainted, or three people who are pairwise strangers.

2Example 2: odd-part pigeonhole—choosing $n+1$ out of $\{1,\dots,2n\}$ necessarily gives a pair where one divides the other

Prove: from any n+1n+1 integers chosen from {1,2,…,2n}\{1,2,\dots,2n\}, there must be one that divides another; and explain why n+1n+1 cannot be reduced to nn.

Ready to practise this topic?

Practise the available questions for this topic. Difficulty labels are platform references.

Practise Combinatorics & Games →