ProofComplete

Proof of the Erdos-Szekeres Bound

We prove the classical Erdos-Szekeres upper bound R(s,t)≀(s+tβˆ’2sβˆ’1)R(s, t) \leq \binom{s + t - 2}{s - 1} using an elegant combinatorial argument that combines induction with Pascal's identity.


Statement

Theorem5.5Erdos-Szekeres Bound

For all integers s,tβ‰₯1s, t \geq 1, R(s,t)≀(s+tβˆ’2sβˆ’1).R(s, t) \leq \binom{s + t - 2}{s - 1}. In particular, R(n,n)≀(2nβˆ’2nβˆ’1)<4n/Ο€nR(n, n) \leq \binom{2n - 2}{n - 1} < 4^n / \sqrt{\pi n}.


Proof

Proof

We proceed by induction on s+ts + t. The base cases are R(s,1)=R(1,t)=1R(s, 1) = R(1, t) = 1 for all s,tβ‰₯1s, t \geq 1, and (sβˆ’1sβˆ’1)=(tβˆ’10)=1\binom{s - 1}{s - 1} = \binom{t - 1}{0} = 1.

For the inductive step, assume s,tβ‰₯2s, t \geq 2 and that R(sβˆ’1,t)≀(s+tβˆ’3sβˆ’2)R(s - 1, t) \leq \binom{s + t - 3}{s - 2} and R(s,tβˆ’1)≀(s+tβˆ’3sβˆ’1)R(s, t - 1) \leq \binom{s + t - 3}{s - 1}.

We first establish the recursive inequality: R(s,t)≀R(sβˆ’1,t)+R(s,tβˆ’1).R(s, t) \leq R(s - 1, t) + R(s, t - 1).

Let N=R(sβˆ’1,t)+R(s,tβˆ’1)N = R(s - 1, t) + R(s, t - 1) and consider any 2-coloring of the edges of KNK_N with colors red and blue. Fix a vertex vv and partition the remaining Nβˆ’1N - 1 vertices into:

  • R={u:vuΒ isΒ red}R = \{u : vu \text{ is red}\}
  • B={u:vuΒ isΒ blue}B = \{u : vu \text{ is blue}\}

Since ∣R∣+∣B∣=Nβˆ’1=R(sβˆ’1,t)+R(s,tβˆ’1)βˆ’1|R| + |B| = N - 1 = R(s-1, t) + R(s, t-1) - 1, either ∣R∣β‰₯R(sβˆ’1,t)|R| \geq R(s-1, t) or ∣B∣β‰₯R(s,tβˆ’1)|B| \geq R(s, t-1).

If ∣R∣β‰₯R(sβˆ’1,t)|R| \geq R(s-1, t), then the coloring restricted to RR contains either a red Ksβˆ’1K_{s-1} (which with vv gives a red KsK_s) or a blue KtK_t. If ∣B∣β‰₯R(s,tβˆ’1)|B| \geq R(s, t-1), then the coloring restricted to BB contains either a red KsK_s or a blue Ktβˆ’1K_{t-1} (which with vv gives a blue KtK_t).

Now apply the inductive hypothesis: R(s,t)≀R(sβˆ’1,t)+R(s,tβˆ’1)≀(s+tβˆ’3sβˆ’2)+(s+tβˆ’3sβˆ’1)=(s+tβˆ’2sβˆ’1),R(s, t) \leq R(s-1, t) + R(s, t-1) \leq \binom{s+t-3}{s-2} + \binom{s+t-3}{s-1} = \binom{s+t-2}{s-1}, where the last step is Pascal's identity. β–‘\square

β– 

Asymptotic Analysis

ExampleDiagonal Case Asymptotics

Setting s=t=ns = t = n gives R(n,n)≀(2nβˆ’2nβˆ’1)=(2nβˆ’2)!((nβˆ’1)!)2.R(n, n) \leq \binom{2n - 2}{n - 1} = \frac{(2n-2)!}{((n-1)!)^2}. By Stirling's approximation, (2nβˆ’2nβˆ’1)∼4nβˆ’1Ο€(nβˆ’1)\binom{2n-2}{n-1} \sim \frac{4^{n-1}}{\sqrt{\pi(n-1)}}, so R(n,n)=O(4n/n)R(n, n) = O(4^n / \sqrt{n}).

RemarkLower Bounds and the Gap

Erdos proved R(n,n)β‰₯(1+o(1))ne22n/2R(n,n) \geq (1 + o(1)) \frac{n}{e\sqrt{2}} 2^{n/2} using the probabilistic method (1947). The gap between cβ‹…2n/2c \cdot 2^{n/2} and Cβ‹…4n/nC \cdot 4^n/\sqrt{n} has been one of the most famous open problems in combinatorics. In 2023, Campos, Griffiths, Morris, and Sahasrabudhe achieved the first exponential improvement to the upper bound, showing R(n,n)≀(4βˆ’Ξ΅)nR(n,n) \leq (4 - \varepsilon)^n for some explicit Ξ΅>0\varepsilon > 0.

RemarkAsymmetric Improvements

For R(3,t)R(3, t), much tighter bounds are known: R(3,t)=Θ(t2/log⁑t)R(3, t) = \Theta(t^2 / \log t). The upper bound follows from Shearer's entropy method and the lower bound from the triangle-free process (Bohman-Keevash, Fiz Pontiveros-Griffiths-Morris).