Wiki
Wiki

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

Updated


Statement

Let FF be a finite nonempty bipartite graph with a fixed coloring and let r≥2r\ge2. For every ε>0\varepsilon>0 and integer B0≥0B_0\ge0, there is an integer B≥B0B\ge B_0 such that every Hr−1(F)H_{r-1}(F)-free graph HH on NN vertices with maximum degree at most an integer D≥0D\ge0 satisfies

∣Bj∣≤εNDj(2≤j≤r).(1)|\mathcal B_j|\le\varepsilon ND^j\qquad(2\le j\le r). \tag{1}

Here Bj\mathcal B_j uses the admissibility thresholds Ln(B)L_n(B) from good paths. The same BB works for every N,D,HN,D,H and all these lengths simultaneously. It depends only on F,r,ε,B0F,r,\varepsilon,B_0.

Proof

Let T,K,UT,K,U be the constants for F,rF,r in heavy-path assembly. Choose an integer κ>1/ε\kappa>1/\varepsilon, so κ≥1\kappa\ge1, and take

B=B0+U+(4κT2)r−1+2K(4κ)T+1.(2)B=B_0+U+(4\kappa T^2)^{r-1}+2K(4\kappa)^{T+1}. \tag{2}

We prove the stronger integer inequality

κ∣Bj∣≤NDj(3)\kappa|\mathcal B_j|\le ND^j \tag{3}

for every 2≤j≤r2\le j\le r.

First suppose D<4κT2D<4\kappa T^2. Every length-jj endpoint fiber has at most Dj−1D^{j-1} members, by the walk bound. Since 4κT2≥14\kappa T^2\ge1 and j−1≤r−1j-1\le r-1,

Dj−1≤(4κT2)r−1≤B≤Lj(B).D^{j-1}\le(4\kappa T^2)^{r-1}\le B\le L_j(B).

The last inequality follows directly from the positive threshold recurrence for j≥1j\ge1. Thus no heavy fiber exists and (3) holds. This includes D=0D=0 and graphs with no vertices.

Otherwise put

d=⌊D2κ⌋.d=\left\lfloor\frac{D}{2\kappa}\right\rfloor.

The assumption D≥4κT2D\ge4\kappa T^2 gives d≥2T2≥1d\ge2T^2\ge1. The division inequalities give

2κd≤D<2κ(d+1)≤4κd.(4)2\kappa d\le D<2\kappa(d+1)\le4\kappa d. \tag{4}

If ∣Bj∣>2NdDj−1|\mathcal B_j|>2NdD^{j-1}, apply heavy common neighborhoods with s=Ts=T, ratio parameter 4κ4\kappa, and target constant KK. Its hypotheses hold by (2) and (4). It supplies the common-heavy configuration and a path family of size greater than KLj−1Dj−1KL_{j-1}D^{j-1}. Write r=mj+lr=mj+l by Euclidean division. Because 2≤j≤r2\le j\le r, one has m≥1m\ge1 and 0≤l<j0\le l<j. Since B≥UB\ge U, heavy-path assembly then produces a copy of Hr−1(F)H_{r-1}(F), contradicting its exclusion.

Therefore ∣Bj∣≤2NdDj−1|\mathcal B_j|\le2NdD^{j-1}. Multiplying by κ\kappa and using 2κd≤D2\kappa d\le D proves (3). Finally 1/κ<ε1/\kappa<\varepsilon implies (1). All constants were chosen before the host graph or the length jj, so the quantifiers are uniform as stated.

Source and scope

Complete reconstruction of AdmissibleFiberDegree and HubAdmissiblePruning.scaled and pruning, pinned Lean lines 9786–9926. The exposition, p. 5, describes this as discarding only a controlled fraction of long paths; the calculation above supplies a common threshold for every shorter length used there. No assumption that heavy paths themselves are mutually disjoint enters the count.

Used by. Proposition 4.1.

Bears on. #571.