Wiki
Wiki

Research notes on every problem, and a library of the papers behind them. Built from the open erdos repository.

Updated


Source. Published p. 231, Fact 3.10 and its proof; the bounded-gap version also completes the dimension step in Lemma 3.4. (canonical PDF).

As printed (p. 231): let cc, α\alpha and ϵ\epsilon be fixed and let {iD}i=1∞\{iD\}_{i=1}^\infty be an infinite arithmetic progression. If VV is a finite set such that for every i≥1i\ge1 there is a set H(iD)⊆RiD\mathcal H(iD)\subseteq\mathbb R^{iD} satisfying (i)–(iii) of Definition 3.1, then VV is α\alpha-hyper Ramsey.

Bounded-gap form proved here. Fix a finite nonempty configuration XX and constants R>0R>0, c>1c>1 and 0<ϵ<10<\epsilon<1. Suppose an unbounded increasing sequence of positive integer dimensions nin_i has bounded gaps and, for every sufficiently large ii, has a nonempty witness Hni⊆S(R,ni)H_{n_i}\subseteq S(R,n_i) with ∣Hni∣<cni|H_{n_i}|<c^{n_i} such that every XX-free subset has density less than (1−ϵ)ni(1-\epsilon)^{n_i}.

Then the same fixed target and radius admit such witnesses in every sufficiently large dimension, with ϵ\epsilon replaced by a smaller positive constant. In particular, if R2=ρ(X)2+αR^2=\rho(X)^2+\alpha and ni=iDn_i=iD for a fixed positive integer DD, then XX is α\alpha-hyper-Ramsey.

Proof.

Let DD bound the gaps, and for each large integer mm choose the largest available ni≤mn_i\le m. Then ni>m−Dn_i>m-D, so ni≥m/2n_i\ge m/2 for all sufficiently large mm. Pad every vector of HniH_{n_i} with zero coordinates. This is an isometry into S(R,m)S(R,m), and ∣Hni∣<cni≤cm|H_{n_i}|<c^{n_i}\le c^m.

Set 1−ϵ′=1−ϵ1-\epsilon'=\sqrt{1-\epsilon}, so 0<ϵ′<ϵ0<\epsilon'<\epsilon. Every XX-free subset has density strictly less than

(1−ϵ)ni≤(1−ϵ)m/2=(1−ϵ′)m.(1-\epsilon)^{n_i}\le(1-\epsilon)^{m/2}=(1-\epsilon')^m.

Thus a subset whose density is at least the last quantity must contain XX, with the weak forcing endpoint preserved. The arithmetic progression in Fact 3.10 is the case ni=iDn_i=iD. The proof also permits a finite initial segment of dimensions to be absent.

Source precision.

The target and radius are fixed before the dimension varies. This argument cannot transfer witnesses for a varying sequence of noncongruent targets.

Bears on. #174.