Wiki
Wiki

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

Updated

Near-regular graphs at the seven-cycle threshold


Suppose

e(Gn)>⌊n2/4⌋,e(Gn)=n2/4+o(n2),δ(Gn)≥n/2−o(n).e(G_n)>\lfloor n^2/4\rfloor, \qquad e(G_n)=n^2/4+o(n^2), \qquad \delta(G_n)\ge n/2-o(n).

Then every C7C_7-rainbow coloring uses n2/8−o(n2)n^2/8-o(n^2) colors.

It suffices to argue along subsequences. Either, for every two distinct vertices and every set of at most ten other vertices, there is a three-edge path between the two vertices avoiding that set, or there is a pair x,yx,y and a forbidden set of size at most ten violating this.

Robust three-edge paths

In the first case choose a vertex pp of maximum degree, and take all edges incident to N(p)N(p), except edges incident to pp. Any two disjoint such edges, written xy,zwxy,zw with x,z∈N(p)x,z\in N(p), lie on the four-edge path

y,x,p,z,w.y,x,p,z,w.

Close it by a three-edge path from ww to yy avoiding x,p,zx,p,z. For adjacent prescribed edges, first greedily extend their two-edge path to a four-edge path, then close by a three-edge path avoiding its internal vertices. The minimum-degree assumption permits the greedy extension.

These edges therefore have distinct colors, and their number is at least

∣N(p)∣δ(G)2−O(n)≥n2/8−o(n2).\frac{|N(p)|\delta(G)}2-O(n)\ge n^2/8-o(n^2).

A pair without a robust three-edge path

In the second case let

A=N(x)∖(S∪{y}),B=N(y)∖(S∪{x}),A=N(x)\setminus(S\cup\{y\}),\qquad B=N(y)\setminus(S\cup\{x\}),

where ∣S∣≤10|S|\le10. Then ∣A∣,∣B∣≥n/2−o(n)|A|,|B|\ge n/2-o(n), and there are no edges between AA and BB, with the usual interpretation when these sets overlap. Indeed, any such edge supplies a three-edge path from xx to yy avoiding SS.

If A∩B≠∅A\cap B\ne\varnothing, a vertex z∈A∩Bz\in A\cap B has no neighbors in A∪BA\cup B. The minimum-degree condition gives ∣A∪B∣≤n/2+o(n)|A\cup B|\le n/2+o(n), and hence ∣A∣=n/2+o(n)|A|=n/2+o(n) and ∣A∖B∣=o(n)|A\setminus B|=o(n). The absence of edges from AA to BB gives e(G[A])=o(n2)e(G[A])=o(n^2). Summing the minimum-degree bound over AA shows that the cut (A,V∖A)(A,V\setminus A) has n2/4−o(n2)n^2/4-o(n^2) edges. Since the total edge count is n2/4+o(n2)n^2/4+o(n^2), only o(n2)o(n^2) edges are internal to that cut. Apply the near-bipartite theorem.

If A∩B=∅A\cap B=\varnothing, the two sets each have size n/2+o(n)n/2+o(n), and only o(n)o(n) vertices lie outside their union. A vertex in AA has no neighbors in BB, so

δ(G[A])≥n/2−o(n)=∣A∣−o(n).\delta(G[A])\ge n/2-o(n)=|A|-o(n).

Thus G[A]G[A] has n2/8−o(n2)n^2/8-o(n^2) edges, and any two are on a common C7C_7 inside AA. To verify the latter directly: any two vertices have ∣A∣−o(n)|A|-o(n) common neighbors, and any two have a three-edge path avoiding any fixed bounded set (choose a neighbor of the first, then a common neighbor of that vertex and the second). For two disjoint edges, choose the two- and three-edge connecting paths with disjoint interiors; for adjacent edges greedily extend to a four-edge path and close by a three-edge path. All these edges consequently have different colors.

Limitation

The homomorphic-cleaning reduction records a useful consequence: the same threshold conclusion holds when the normalized degree variance tends to zero, without assuming a minimum degree initially. A low-degree pruning removes only o(n)o(n) vertices, preserves strict super-Turan density, and reaches the hypotheses of this note.

Deleting low-degree vertices from a general threshold graph can increase the normalized density while decreasing the order by a positive proportion. The resulting bound in terms of the smaller order does not give the desired bound in terms of the original order. The full-density formula of Bucić, Chen and Ma, Theorem 1.2 (BCM) resolves this for longer odd cycles using a stronger all-edge induction potential; that potential is false for C7C_7, as shown by the three-branch example in the dense-curve obstruction.

A weighted minimum-degree case

There is also a direct finite-template statement. For full-support capacities, if δw≥1/2\delta_w\ge1/2 and q>1/4q>1/4, then

Φ(J23;m)≥2q2.\Phi(J_{23};m)\ge2q^2.

First suppose every vertex is triangular. If no three-walk joins x,yx,y, their neighborhoods are anticomplete. If the neighborhoods overlap, any vertex in their intersection has degree at most one minus the union mass. The minimum-degree assumption forces both neighborhoods to be the same independent set of mass 1/21/2, contradicting triangularity of xx. If they are disjoint, both have mass 1/21/2 and partition the support; minimum degree forces two isolated complete looped parts, giving q=1/4q=1/4. Thus the three-walk relation is complete. The full-triangular-vertex averaging argument in the triangle-average note gives Φ≥2q2\Phi\ge2q^2.

If a nontriangular vertex vv exists, its neighborhood AA is independent. Every neighbor has degree at most 1−d(v)1-d(v), so the minimum-degree condition forces d(v)=1/2d(v)=1/2, and every vertex of AA is completely joined to B=V∖AB=V\setminus A, also of mass 1/21/2. Thus the support consists of this complete bipartite join and an arbitrary graph inside BB. Since q>1/4q>1/4, BB has an internal edge. The rectangle anchored at any type of AA, with target AA together with the nonisolated types of BB, contains every host edge; its target is a three-walk clique. Hence in this case Φ=q≥2q2\Phi=q\ge2q^2.

This minimum-degree argument alone does not cover a positive mass of vertices with degree below 1/21/2. The general theorem uses the palette savings bound.