Wiki
Wiki

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

Updated

Uniform clique completion


Proved conditional structural theorem, not a solution of E617. The completion argument below has no dependence on the number of rows or on the total number of exceptions. Its hypotheses require a full family of as many disjoint light clique rows as the independence bound. No extraction or suppression theorem supplies that family in an arbitrary surviving core.

Use the hereditary caps from Reductions:

e(H[W])≤Dr(∣W∣),Dr(n)=(n2)−(r−1)pr(n),pr(rq+b)=r(q2)+qb.e(H[W])\le D_r(|W|),\qquad D_r(n)=\binom n2-(r-1)p_r(n),\qquad p_r(rq+b)=r\binom q2+qb.

1. An all-size row inequality and a local incidence bound

Let QQ be a clique of order r−dr-d, with d≥1d\ge1. For x∉Qx\notin Q put

mQ(x)=∣Q∖NH(x)∣,wQ(x)=dH(x,Q)+d−1r−1=1−mQ(x)r−1.m_Q(x)=|Q\setminus N_H(x)|,\qquad w_Q(x)=\frac{d_H(x,Q)+d-1}{r-1} =1-\frac{m_Q(x)}{r-1}.

These weights lie in [0,1][0,1]. For every S⊆V(H)∖QS\subseteq V(H)\setminus Q,

∑x∈SwQ(x)≤d+e‾H(S)r−1.(1)\boxed{\sum_{x\in S}w_Q(x) \le d+\frac{\overline e_H(S)}{r-1}.} \tag{1}

If s=∣S∣<ds=|S|<d, use wQ≤1w_Q\le1. Otherwise set u=s−du=s-d. The identity pr(r+u)=u+pr(u)p_r(r+u)=u+p_r(u) gives

Dr(r+u)=(r2)+(u+12)−(r−1)pr(u).D_r(r+u)=\binom r2+\binom{u+1}{2}-(r-1)p_r(u).

Subtracting the edges inside QQ and SS proves the stronger inequality

∑x∈SwQ(x)≤d−pr(s−d)+e‾H(S)r−1.(2)\sum_{x\in S}w_Q(x) \le d-p_r(s-d)+\frac{\overline e_H(S)}{r-1}. \tag{2}

Thus (1) does not require ∣Q∪S∣≤2r|Q\cup S|\le2r.

If 24d<r+524d<r+5, then

∣{x∉Q:wQ(x)≥1/3}∣≤4d−1.(3)\boxed{\bigl|\{x\notin Q:w_Q(x)\ge1/3\}\bigr|\le4d-1.} \tag{3}

Otherwise apply (1) to any 4d4d such vertices. It gives 4d/3≤d+(4d2)/(r−1)4d/3\le d+\binom{4d}{2}/(r-1), which is equivalent to r+5≤24dr+5\le24d.

The point of (3) is local: the number of substantial incidences into one light row is controlled by its own deficit, even if there are arbitrarily many exceptions elsewhere.

2. Completion from a robust transversal hypothesis

Theorem. Let a≥1a\ge1, ε∈{0,1}\varepsilon\in\{0,1\}, and

∣H∣=ar+ε,α(H)≤a.|H|=ar+\varepsilon,\qquad \alpha(H)\le a.

Assume all hereditary caps. Suppose Q1,…,QaQ_1,\ldots,Q_a are disjoint cliques, with

∣Qi∣=r−di,1≤di≤k,24k<r.|Q_i|=r-d_i,\qquad 1\le d_i\le k,\qquad24k<r.

Suppose an integer L≥1L\ge1 has the following properties:

  • L≤(r−1)/3L\le(r-1)/3 and 3(L−1)+24k−12<r−13(L-1)+24k-12<r-1;
  • a vertex in one row has at most (r−1)/3(r-1)/3 neighbors in any other single row;
  • every subfamily of the rows, with an arbitrary list of at least LL vertices in each row, has an independent transversal.

Then ε=1\varepsilon=1 is impossible. If ε=0\varepsilon=0, each row can be enlarged so that the resulting sets partition V(H)V(H) into aa cliques of order rr.

Proof. Put X=V(H)∖⋃iQiX=V(H)\setminus\bigcup_iQ_i, t=∣X∣t=|X|, and D=∑idi=t−εD=\sum_i d_i=t-\varepsilon. Every x∈Xx\in X has a row i(x)i(x) with mQi(x)(x)<Lm_{Q_{i(x)}}(x)<L: otherwise the transversal hypothesis would extend xx to an independent (a+1)(a+1)-set. Choose one such row as its owner.

Write wi=wQiw_i=w_{Q_i} and

Si={x∈X:wi(x)≥1/3}.S_i=\{x\in X:w_i(x)\ge1/3\}.

Every exception belongs to its owner's SiS_i. Call it active if it belongs to at least two of these sets, and let AA be the active set. For an inactive exception xx with owner ii,

mQj(x)>2(r−1)/3(j≠i).(4)m_{Q_j}(x)>2(r-1)/3\quad(j\ne i). \tag{4}

The row QiQ_i and all its inactive owned exceptions form a clique. Indeed, if such an xx is nonadjacent to v∈Qiv\in Q_i, their common nonneighbors in every other row number more than 2(r−1)/3−(r−1)/3≥L2(r-1)/3-(r-1)/3\ge L. If two inactive exceptions with owner ii are nonadjacent, their common nonneighbors in QjQ_j number more than

4(r−1)/3−(r−dj)=(r−4)/3+dj≥(r−1)/3≥L.4(r-1)/3-(r-d_j)=(r-4)/3+d_j\ge(r-1)/3\ge L.

In either case a transversal in the remaining rows augments the pair to size a+1a+1, a contradiction. For a=1a=1 the pair itself is already the contradiction.

Let BB be the sum of wi(x)w_i(x) over all nonowner memberships x∈Six\in S_i, and let KK be their number. Then

K≤3B,∣A∣≤K.K\le3B,\qquad |A|\le K.

Only active vertices can have owner holes, so

M:=∑x∈XmQi(x)(x)≤(L−1)∣A∣≤3(L−1)B.(5)M:=\sum_{x\in X}m_{Q_{i(x)}}(x) \le(L-1)|A|\le3(L-1)B. \tag{5}

By (3), ∣Si∣≤4di−1≤4k−1|S_i|\le4d_i-1\le4k-1. Every nonedge within SiS_i meets an active vertex, because all inactive members of SiS_i have owner ii and form a clique. Consequently

E:=∑ie‾H(Si)≤(4k−2)∑i∣A∩Si∣=(4k−2)(∣A∣+K)≤(24k−12)B.(6)\begin{aligned} E:=\sum_i\overline e_H(S_i) &\le(4k-2)\sum_i|A\cap S_i|\\ &=(4k-2)(|A|+K)\le(24k-12)B. \tag{6} \end{aligned}

Summing (1) over the SiS_i, and separating owner from nonowner memberships, gives

t−Mr−1+B≤D+Er−1.t-\frac M{r-1}+B\le D+\frac E{r-1}.

Thus

ε+B≤3(L−1)+24k−12r−1B.(7)\boxed{\varepsilon+B\le \frac{3(L-1)+24k-12}{r-1}B.} \tag{7}

Its coefficient is strictly less than one. The overloaded case is impossible. In the balanced case B=0B=0, so every exception is inactive. The owner extensions are disjoint cliques covering arar vertices. Each has order at most rr, by the cap on r+1r+1 vertices, and hence each has order exactly rr. □\square

3. Explicit sufficient transversal conditions

The robust hypothesis has several parameterized sufficient conditions. Here all degrees refer to the cross graph on ⋃iQi\bigcup_i Q_i, with edges within the rows ignored.

Self-contained local-lemma version. Let Δ\Delta be its maximum total degree. It is enough, in addition to the numerical conditions of the theorem, that

L2≥6rΔ.L^2\ge6r\Delta.

In arbitrary lists of size at least LL, a cross edge is selected with probability at most L−2L^{-2}. A bad event has at most 2rΔ−12r\Delta-1 dependent other bad events, since each row has at most rΔr\Delta incident cross edges. The symmetric local lemma applies because e(2rΔ)/L2≤e/3<1e(2r\Delta)/L^2\le e/3<1. If there are no events, the assertion is immediate. These inequalities also imply Δ<(r−1)/54\Delta<(r-1)/54, so the theorem's individual-row degree condition is automatic.

Haxell version. One may instead use

L≥2ΔL\ge2\Delta

with the other numerical conditions unchanged. Haxell's independent transversal theorem applies to every list restriction. Its precise statement is Theorem 1, and its topological proof is supplied by Theorems 7, 14 and Corollary 15, in Haxell--Wdowinski, Degree criteria and stability for independent transversals. Those passages were checked in the primary text. The local degree condition follows from Δ≤L/2≤(r−1)/6\Delta\le L/2\le(r-1)/6.

Total-edge/local-degree version. Let E×E_\times count all cross edges, and let dd bound the number of neighbors a vertex has in any one other row. For L≥8L\ge8, it is enough that

d≤(r−1)/3,E×<L3128max⁡{d,1}.d\le(r-1)/3,\qquad E_\times<\frac{L^3}{128\max\{d,1\}}.

Apply the elementary transversal theorem in Packing tools to every subfamily and list restriction. In particular the completion theorem applies asymptotically when k=o(r)k=o(r), d=o(r)d=o(r), and E×=O(r2)E_\times=O(r^2): choose, for example, L=⌊r/6⌋L=\lfloor r/6\rfloor. This statement still assumes all aa rows exist.

4. Exact unresolved step

The earlier completion argument in Hereditary span paid for exceptional nonedges using the total exception count. Equations (3) and (6) remove that loss. The theorem above therefore works with an unbounded number of rows and exceptions, once the stated row family and transversal control are available.

The existing cluster extractions can instead leave fewer than aa light rows, or rows with large deficits or concentrated cross degrees. The owner existence argument then fails: one exception and a transversal of only q<aq<a rows need not exceed the independence bound. Likewise a nonadjacent same-owner pair need not augment to size a+1a+1. No deletion of those light rows, preservation of caps under folding, or extension across a heavy residual has been proved here.

The local-lemma proof and all incidence and endpoint calculations were checked by the author and are author-recorded. No finite search was used. This is neither an unrestricted infinite family of solved parameters nor a counterexample to the problem.