Wiki
Wiki

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

Updated

Global clique packing


Proved necessary conditions, not a resolution. Let pp be the maximum number of vertex-disjoint monochromatic KrK_r's in one color of a hypothetical counterexample. Section 5 of the third independent input correctly proves p≤r−2p\le r-2 for r≥3r\ge3 and, in its extension, p≤r−3p\le r-3 for r≥8r\ge8. The missing boundary arguments and the triangle-support lemma are supplied below.

Combining the argument with the established private-row excess bound gives the stronger conclusion

p≤r−max⁡{3,⌊r/8⌋}−1(r≥8).(1)\boxed{p\le r-\max\{3,\lfloor\sqrt r/8\rfloor\}-1 \qquad(r\ge8).} \tag{1}

Unlike the older minority-color packing bound, these restrictions apply to every color. They do not force a packing large enough to contradict them, and they are not bounds on the total number of full cliques.

Fibers and independent-set attachments

An outsider has at most one own-color neighbor in a full clique. Fix a full color-ii clique AA. For j≠ij\ne i, let bj(A)b_j(A) count outsiders whose star into AA repeats color jj. Every such star contains each foreign color; it is either rainbow or omits ii and repeats exactly one foreign color.

For each u∈Au\in A, its private color-jj fiber consists of uu and the outsiders with unique color-jj neighbor uu in AA. It is a clique: otherwise two private nonneighbors, together with A∖{u}A\setminus\{u\}, would be a color-jj independent (r+1)(r+1)-set. The rr fibers cover r2+1−bj(A)r^2+1-b_j(A) vertices and their deficits from order rr sum to bj(A)−1b_j(A)-1. At least r−bj(A)+1r-b_j(A)+1 are full, giving

bj(A)≥r−p+1.b_j(A)\ge r-p+1.

If ti(A)t_i(A) counts color-ii edges leaving AA, star counting yields

ti(A)=r2−r+1−∑j≠ibj(A)≤(r−1)(p−1)+1.(2)t_i(A)=r^2-r+1-\sum_{j\ne i}b_j(A) \le(r-1)(p-1)+1. \tag{2}

Suppose a color has k≥2k\ge2 disjoint full rows and its residue contains an independent set II of order r+1−kr+1-k. Avoid the neighbors of II in every row. The resulting lists have order at least k−1k-1 and matching cross graphs. An independent transversal would augment II to order r+1r+1, so matching-transversal rigidity makes all lists have order exactly k−1k-1, with a perfect matching between each pair. In particular each member of II has a neighbor in every row, and these neighbors are distinct within each row.

Let SS be the residue vertices belonging to at least one such II, and write s=∣S∣s=|S|. Every pair of rows has at least k−1k-1 cross edges, and each vertex of SS attaches to every row. Thus

ti(A)≥(k−1)2+s.(3)\boxed{t_i(A)\ge(k-1)^2+s.} \tag{3}

This does not assume simultaneous choices of the different independent sets: the forced attachments are individual edges.

Excluding a packing of r minus one full cliques

The full-resolution obstruction already gives p≤r−1p\le r-1. Suppose equality holds, and take r−1r-1 rows in color ii. The residue RR has order r+1r+1. Set F=Gi[R]‾F=\overline{G_i[R]} and let ss count its nonisolated vertices. We have e(F)≥r−1e(F)\ge r-1. Applying (3) to independent pairs and comparing with (2) gives s≤r−1s\le r-1. Also

∑j≠ibj(A)≤3r−3−s.\sum_{j\ne i}b_j(A)\le3r-3-s.

Since every bj(A)≥2b_j(A)\ge2, at least ss foreign colors have bj(A)=2b_j(A)=2. Each therefore has r−1r-1 full private rows. This is the propagation step missing from a one-color edge argument.

Put T=r(r2)+rT=r\binom r2+r. Every color has at least T+r−1T+r-1 edges, by the nonpartite gap, or the coupled block count in Palette and cut bounds. The total surplus available above rTrT is r2(r−1)/2r^2(r-1)/2.

For r≥5r\ge5, the inequality e(F)≥r−1≥4e(F)\ge r-1\ge4 forces s≥4s\ge4. Every packing color therefore has at least

(r−1)(r2)+(r−2)(r−12)+4(r−1)(r-1)\binom r2+(r-2)\binom{r-1}{2}+4(r-1)

edges, with surplus A=(r3−6r2+15r−12)/2A=(r^3-6r^2+15r-12)/2 above TT. There are at least three such colors, and

3A+(r−3)(r−1)−r2(r−1)2=(r−3)(r−2)(2r−5)2>0,3A+(r-3)(r-1)-\frac{r^2(r-1)}2 =\frac{(r-3)(r-2)(2r-5)}2>0,

a contradiction.

The input's separate proof for the smaller parameters is not retained, but these cases are direct. For r=3r=3, s≤2s\le2 contradicts e(F)≥2e(F)\ge2. For r=4r=4, necessarily s=3s=3 and F=K3⊔2K1F=K_3\sqcup2K_1, so the target residue has seven edges. Propagation gives all four colors three-row packings. Each has at least

3(42)+2(32)+3⋅3+7=403\binom42+2\binom32+3\cdot3+7=40

edges, exceeding the total (172)=136\binom{17}{2}=136 when summed. Hence p≤r−2p\le r-2 for all r≥3r\ge3.

A triangle-support lemma

Lemma. Suppose a graph FF on 2r+12r+1 vertices contains a triangle, and every vertex belonging to a triangle lies in a set SS with s=∣S∣≤rs=|S|\le r. Then

e(F)≤r2+s−1.(4)e(F)\le r^2+s-1. \tag{4}

Put Y=V(F)∖SY=V(F)\setminus S. Each vertex of YY has an independent neighborhood. Choose y∈Yy\in Y of maximum total degree among vertices of YY, and put A=N(y)∩YA=N(y)\cap Y, B=Y∖AB=Y\setminus A, C=N(y)∩SC=N(y)\cap S. Replace all vertices of BB by independent twins of yy. The edge gain is

∑b∈B(d(y)−d(b))+e(F[B])≥0.\sum_{b\in B}(d(y)-d(b))+e(F[B])\ge0.

No triangle meeting YY is created: the new neighborhood of every vertex in BB is the independent set A∪CA\cup C. If AA is nonempty, make all its vertices twins of a member with the most neighbors in SS. This also does not decrease the edge count. Their common neighborhood D⊆SD\subseteq S is independent and disjoint from CC. If AA is empty, take D=∅D=\varnothing and omit this operation.

Now YY is complete bipartite on A,BA,B, its edges to SS are exactly those from AA to DD and BB to CC, and F[S]F[S] is unchanged. Because F[S]F[S] still contains a triangle, t=∣S∖(C∪D)∣≥1t=|S\setminus(C\cup D)|\ge1. Writing a,b,c,da,b,c,d for the respective sizes gives

e(F)≤ab+ad+bc+cd+t(s−t)+(t2)≤⌊(2r+1−t)24⌋+ts−t(t+1)2.\begin{aligned} e(F)&\le ab+ad+bc+cd+t(s-t)+\binom t2\\ &\le\left\lfloor\frac{(2r+1-t)^2}{4}\right\rfloor +ts-\frac{t(t+1)}2. \end{aligned}

After dropping the floor, the difference of the final expression from r2+s−1r^2+s-1 is

(t−1)(s−r−1−t+14)≤0.(t-1)\left(s-r-1-\frac{t+1}{4}\right)\le0.

This proves (4), including the empty-AA case. It is sharp: add to Kr,rK_{r,r} an apex adjacent to nonempty subsets of opposite sides whose total order is s−1s-1. Exactly ss vertices lie in triangles, and the edge count is r2+s−1r^2+s-1.

Checking the independent extension

Let r≥8r\ge8 and suppose p=r−2p=r-2. The residue has order 2r+12r+1, so two-row capacity forces an independent triple. Let ss count the vertices participating in such triples. The complement FF of the residue has

e(F)≥(r−1)(r+2)=r2+r−2.e(F)\ge(r-1)(r+2)=r^2+r-2.

If s≤rs\le r, apply (4); otherwise no lemma is needed. In both cases s≥r−1s\ge r-1. Equations (2)--(3) give

ti(A)≥(r−3)2+s,∑j≠ibj(A)≤5r−8−s.t_i(A)\ge(r-3)^2+s,\qquad \sum_{j\ne i}b_j(A)\le5r-8-s.

Every bj(A)≥3b_j(A)\ge3, so at least s−r+4≥3s-r+4\ge3 foreign colors have bj(A)=3b_j(A)=3, and therefore have r−2r-2 full rows. Each of these at least four packing colors has surplus above TT at least

A2=r3−8r2+15r−142,A_2=\frac{r^3-8r^2+15r-14}{2},

by counting row edges, (r−3)(r-3) cross edges per row pair, and r−1r-1 supported residue attachments per row. Residual edges are omitted. The budget excess is

4A2+(r−4)(r−1)−r2(r−1)2=(r−8)(3r2−5r+10)2+16>04A_2+(r-4)(r-1)-\frac{r^2(r-1)}2 =\frac{(r-8)(3r^2-5r+10)}2+16>0

for r≥8r\ge8. Thus the archived extension is valid without assuming any of its unretained proofs.

Stronger comparison using established private-row excess

For r≥7r\ge7, Cross-color cores proves that every foreign private system has excess deficit

Ej≥e(r):=⌈8r−7−52⌉.E_j\ge e(r):=\left\lceil\frac{\sqrt{8r-7}-5}{2}\right\rceil.

Its exact number of full rows is fj=r−bj+1+Ejf_j=r-b_j+1+E_j. Since fj≤pf_j\le p, the fiber bound strengthens to

bj≥r−p+1+e(r),ti(A)≤(r−1)(p−1−e(r))+1.(5)b_j\ge r-p+1+e(r),\qquad t_i(A)\le(r-1)(p-1-e(r))+1. \tag{5}

This established excess bound applies without bj≤rb_j\le r; its owner proof and the required supporting-pair size conditions are recorded in the linked note.

Suppose p=r−a≥2p=r-a\ge2 and the residue has an independent (a+1)(a+1)-set. Equation (3), even using only these a+1a+1 supported vertices, gives ti(A)≥(p−1)2+a+1t_i(A)\ge(p-1)^2+a+1. The upper bound (5) minus this lower bound is exactly

a(p−2)−(r−1)e(r).(6)a(p-2)-(r-1)e(r). \tag{6}

It is negative whenever a≤e(r)a\le e(r).

For a=1a=1, the clique cap supplies the required independent pair. For a=2,3a=2,3, the small-independence span bounds supply it when r≥8r\ge8. We have e(r)≥2e(r)\ge2 for r≥8r\ge8 and e(r)≥3e(r)\ge3 for r≥12r\ge12. When 8≤r≤118\le r\le11, e(r)=2e(r)=2 and (6) for a=3a=3 is r−13<0r-13<0. Thus p=r−1,r−2,r−3p=r-1,r-2,r-3 are all excluded for r≥8r\ge8.

Put m=max⁡{3,⌊r/8⌋}m=\max\{3,\lfloor\sqrt r/8\rfloor\}. For 4≤a≤m4\le a\le m, r≥64a2r\ge64a^2 and e(r)≥ae(r)\ge a. The hereditary span theorem forces an independent (a+1)(a+1)-set in the ar+1ar+1 residue, so (6) again contradicts (5). Together with p≠rp\ne r, this proves (1). All these cases have p≥2p\ge2; if p≤1p\le1, (1) is immediate.

At r=7r=7, assuming p=r−2p=r-2, (3) and (5) instead give s≤3s\le3, while (4) gives s≥6s\ge6. Thus p≤r−3p\le r-3 also holds at that parameter. This is not a new fixed case of the coloring problem, whose status is unchanged. The essential unresolved issue is a lower bound forcing a near-complete packing, not the validity of these upper bounds.