Factoring complete graphs and hypergraphs into factors with few maximal cliques

Paul ErdΕ‘s Hungarian Academy of Sciences, Budapest, Hungary David P. Galvin david.galvin@gmail.com Somerville, MA 02143 Fred Galvin bof@sunflower.com Dept. of Mathematics, University of Kansas, Lawrence, KS 66045 Michael M. Krieger mkrieger@alumni.caltech.edu Krieger Law Office, 11209 National Blvd., #417, Los Angeles, CA 90064
Abstract

For integers r,tβ‰₯2π‘Ÿπ‘‘2r,t\geq 2 and nβ‰₯1𝑛1n\geq 1 let fr​(t,n)subscriptπ‘“π‘Ÿπ‘‘π‘›f_{r}(t,n) be the minimum, over all factorizations of the complete rπ‘Ÿr-uniform hypergraph of order n𝑛n into t𝑑t factors H1,…,Htsubscript𝐻1…subscript𝐻𝑑H_{1},\dots,H_{t}, of βˆ‘i=1tc​(Hi)superscriptsubscript𝑖1𝑑𝑐subscript𝐻𝑖\sum_{i=1}^{t}c(H_{i}) where c​(Hi)𝑐subscript𝐻𝑖c(H_{i}) is the number of maximal cliques in Hisubscript𝐻𝑖H_{i}. It is known that f2​(2,n)=n+1subscript𝑓22𝑛𝑛1f_{2}(2,n)=n+1; in fact, if G𝐺G is a graph of order n𝑛n, then c​(G)+c​(GΒ―)β‰₯n+1𝑐𝐺𝑐¯𝐺𝑛1c(G)+c(\overline{G})\geq n+1 with equality iff ω​(G)+α​(G)=n+1πœ”πΊπ›ΌπΊπ‘›1\omega(G)+\alpha(G)=n+1 where Ο‰πœ”\omega is the clique number and α𝛼\alpha the independence number. In this paper we investigate fr​(t,n)subscriptπ‘“π‘Ÿπ‘‘π‘›f_{r}(t,n) when r>2π‘Ÿ2r>2 or t>2𝑑2t>2. We also characterize graphs G𝐺G of order n𝑛n with c​(G)+c​(GΒ―)=n+2𝑐𝐺𝑐¯𝐺𝑛2c(G)+c(\overline{G})=n+2.

00footnotetext: Some of this research was conducted at the University of California, Los Angeles, and the University of Kansas, and was supported in part by grants from the National Science Foundation.00footnotetext: 2020 Mathematics Subject Classification: Primary: 05C69; Secondary: 05C6500footnotetext: Keywords: graphs, hypergraphs, maximal cliques, maximal anticliques

1 Introduction

Let c​(G)𝑐𝐺c(G) denote the number of maximal cliques and c¯​(G)=c​(GΒ―)¯𝑐𝐺𝑐¯𝐺\overline{c}(G)=c(\overline{G}) the number of maximal anticliques in a graph G𝐺G. In an earlier paper [2] two of the present authors showed that, if G𝐺G is a graph of order n𝑛n, then c​(G)+c¯​(G)β‰₯n+1𝑐𝐺¯𝑐𝐺𝑛1c(G)+\overline{c}(G)\geq n+1 with equality just in case ω​(G)+α​(G)=n+1πœ”πΊπ›ΌπΊπ‘›1\omega(G)+\alpha(G)=n+1 where Ο‰πœ”\omega is the clique number and α𝛼\alpha the independence number. (Proofs of these facts will be given here; see Theorem 28 and Theorem 34(a).) In this paper we generalize the problem in two ways: by considering factorizations of the complete graph Knsubscript𝐾𝑛K_{n} into t𝑑t factors instead of just the two factors G𝐺G and G¯¯𝐺\overline{G}, and by considering uniform hypergraphs instead of graphs. We also characterize the graphs G𝐺G of order n𝑛n with c​(G)+c¯​(G)=n+2𝑐𝐺¯𝑐𝐺𝑛2c(G)+\overline{c}(G)=n+2.

An rπ‘Ÿr-uniform hypergraph (in this paper rβ‰₯2π‘Ÿ2r\geq 2) is a structure H=(V,E)𝐻𝑉𝐸H=(V,E) comprising a nonempty finite set V=V​(H)𝑉𝑉𝐻V=V(H) of vertices and a set E=E​(H)βŠ†(Vr)𝐸𝐸𝐻binomialπ‘‰π‘ŸE=E(H)\subseteq\binom{V}{r} of edges; n​(H)=|V|𝑛𝐻𝑉n(H)=|V| is the order and e​(H)=|E|𝑒𝐻𝐸e(H)=|E| is the size of H𝐻H. (A 222-uniform hypergraph is a graph; a 333-uniform hypergraph is a triple system.) A set XβŠ†V𝑋𝑉X\subseteq V is a clique if (Xr)βŠ†Ebinomialπ‘‹π‘ŸπΈ\binom{X}{r}\subseteq E, an anticlique if (Xr)∩E=βˆ…binomialπ‘‹π‘ŸπΈ\binom{X}{r}\cap E=\varnothing. The complement of H𝐻H is the hypergraph HΒ―=(V,(Vr)βˆ–E)¯𝐻𝑉binomialπ‘‰π‘ŸπΈ\overline{H}=(V,\binom{V}{r}\setminus E). We write c​(H)𝑐𝐻c(H) for the number of maximal cliques and c¯​(H)=c​(HΒ―)¯𝑐𝐻𝑐¯𝐻\overline{c}(H)=c(\overline{H}) for the number of maximal anticliques in H𝐻H. We write D​(v)𝐷𝑣D(v) for the number of maximal cliques and D¯​(v)¯𝐷𝑣\overline{D}(v) for the number of maximal anticliques containing a vertex v𝑣v. We define d​(H)=min⁑{D​(v):v∈V}𝑑𝐻:𝐷𝑣𝑣𝑉d(H)=\min\{D(v):v\in V\} and d¯​(H)=d​(HΒ―)=min⁑{D¯​(v):v∈V}¯𝑑𝐻𝑑¯𝐻:¯𝐷𝑣𝑣𝑉\overline{d}(H)=d(\overline{H})=\min\{\overline{D}(v):v\in V\}. The complete rπ‘Ÿr-uniform hypergraph of order n𝑛n is Knr=(V,(Vr))subscriptsuperscriptπΎπ‘Ÿπ‘›π‘‰binomialπ‘‰π‘ŸK^{r}_{n}=(V,\binom{V}{r}) where |V|=n𝑉𝑛|V|=n. Note that c​(Knr)=1𝑐subscriptsuperscriptπΎπ‘Ÿπ‘›1c(K^{r}_{n})=1 and c¯​(Knr)=max⁑{(nrβˆ’1),1}¯𝑐subscriptsuperscriptπΎπ‘Ÿπ‘›binomialπ‘›π‘Ÿ11\overline{c}(K^{r}_{n})=\max\{\binom{n}{r-1},1\}. An edge {u,v}𝑒𝑣\{u,v\} of a graph may be written as u​v𝑒𝑣uv if no confusion will result. The neighborhood of a vertex v𝑣v in a graph G=(V,E)𝐺𝑉𝐸G=(V,E) is the set N​(v)=NG​(v)={u∈V:u​v∈E}𝑁𝑣subscript𝑁𝐺𝑣conditional-set𝑒𝑉𝑒𝑣𝐸N(v)=N_{G}(v)=\{u\in V:uv\in E\}, and we write N¯​(v)¯𝑁𝑣\overline{N}(v) for NG¯​(v)={u∈V:u​vβˆ‰E,uβ‰ v}subscript𝑁¯𝐺𝑣conditional-set𝑒𝑉formulae-sequence𝑒𝑣𝐸𝑒𝑣N_{\overline{G}}(v)=\{u\in V:uv\notin E,u\neq v\}. The clique number ω​(G)πœ”πΊ\omega(G) of a graph G𝐺G is the maximum number of vertices in a clique of G𝐺G; the independence number α​(G)=ω​(GΒ―)π›ΌπΊπœ”Β―πΊ\alpha(G)=\omega(\overline{G}) is the maximum number of vertices in an anticlique of G𝐺G. See West [4] for graph-theoretic notation and terminology not defined here.

For integers r,tβ‰₯2π‘Ÿπ‘‘2r,t\geq 2 and nβ‰₯1𝑛1n\geq 1 we define fr​(t,n)subscriptπ‘“π‘Ÿπ‘‘π‘›f_{r}(t,n) as the minimum of βˆ‘i=1tc​(Hi)superscriptsubscript𝑖1𝑑𝑐subscript𝐻𝑖\sum_{i=1}^{t}c(H_{i}) over all factorizations of KnrsubscriptsuperscriptπΎπ‘Ÿπ‘›K^{r}_{n} into t𝑑t edge-disjoint spanning subgraphs H1,…,Htsubscript𝐻1…subscript𝐻𝑑H_{1},\dots,H_{t}. In this notation, part of the aforementioned result of [2] may be expressed as follows. (A slightly stronger result will be proved here as Theorem 28.)

Theorem 1.

f2​(2,n)=n+1subscript𝑓22𝑛𝑛1f_{2}(2,n)=n+1.

Sometimes we find it convenient to use the language of colorings: fr​(t,n)subscriptπ‘“π‘Ÿπ‘‘π‘›f_{r}(t,n) is the least possible number of maximal monochromatic cliques in an edge coloring of KnrsubscriptsuperscriptπΎπ‘Ÿπ‘›K^{r}_{n} with t𝑑t colors, it being understood that a set which is a maximal monochromatic clique for more than one color is counted once for each color; e.g., f2​(3,2)=1+2+2=5subscript𝑓2321225f_{2}(3,2)=1+2+2=5. Most of this paper is a study of the function fr​(t,n)subscriptπ‘“π‘Ÿπ‘‘π‘›f_{r}(t,n). We have determined exact values only in some very special cases; mostly we have upper and lower bounds which are far apart.

In Β§2 we establish some general facts about the function fr​(t,n)subscriptπ‘“π‘Ÿπ‘‘π‘›f_{r}(t,n), most notably an asymptotic lower bound: for fixed integers r,tβ‰₯2π‘Ÿπ‘‘2r,t\geq 2 and any Ξ΅>0πœ€0\varepsilon>0 we show that fr​(t,n)>nrβˆ’1βˆ’Ξ΅subscriptπ‘“π‘Ÿπ‘‘π‘›superscriptπ‘›π‘Ÿ1πœ€f_{r}(t,n)>n^{r-1-\varepsilon} for all sufficiently large n𝑛n (Theorem 7).

In Β§3 we find upper bounds for f3​(2,n)subscript𝑓32𝑛f_{3}(2,n), the minimum number of maximal cliques in a triple system of order n𝑛n and its complement. The main result here is that f3​(2,n)β‰€βŒŠn24βŒ‹+5subscript𝑓32𝑛superscript𝑛245f_{3}(2,n)\leq\left\lfloor\frac{n^{2}}{4}\right\rfloor+5. For n≀12𝑛12n\leq 12 this is definitely not the best possible, e.g., f3​(2,7)=14subscript𝑓32714f_{3}(2,7)=14 is attained by the Fano plane (Theorem 11).

In Β§4 we find upper bounds and a few exact values for f2​(t,n)subscript𝑓2𝑑𝑛f_{2}(t,n) when t>2𝑑2t>2. We show that, if there is a projective plane of order qπ‘žq, then f2​(q+1,n)=q2+qβˆ’1subscript𝑓2π‘ž1𝑛superscriptπ‘ž2π‘ž1f_{2}(q+1,n)=q^{2}+q-1 for (qβˆ’1)2<n≀(qβˆ’1)​qsuperscriptπ‘ž12π‘›π‘ž1π‘ž(q-1)^{2}<n\leq(q-1)q, and f2​(q+1,n)=q2+qsubscript𝑓2π‘ž1𝑛superscriptπ‘ž2π‘žf_{2}(q+1,n)=q^{2}+q for (qβˆ’1)​q<n≀q2π‘ž1π‘žπ‘›superscriptπ‘ž2(q-1)q<n\leq q^{2} (Theorem 19). For qβ‰₯2π‘ž2q\geq 2 we observe that f2​(q+1,q2)=q2+qsubscript𝑓2π‘ž1superscriptπ‘ž2superscriptπ‘ž2π‘žf_{2}(q+1,q^{2})=q^{2}+q if and only if a projective plane of order qπ‘žq exists (Theorem 20). We obtain an upper bound of the form f2​(t,n)≀n+Ctsubscript𝑓2𝑑𝑛𝑛subscript𝐢𝑑f_{2}(t,n)\leq n+C_{t} whenever a projective plane of order t𝑑t or tβˆ’1𝑑1t-1 exists (Theorem 21); in particular see Theorem 23 for the case t=3𝑑3t=3, Theorem 25 for t=4𝑑4t=4, Theorems 26 and 27 for t=5𝑑5t=5. We have no counterexample to the conjecture that f2​(3,n)=n+2subscript𝑓23𝑛𝑛2f_{2}(3,n)=n+2 for n≑1(mod3)𝑛annotated1pmod3n\equiv 1\pmod{3} and f2​(3,n)=n+3subscript𝑓23𝑛𝑛3f_{2}(3,n)=n+3 otherwise (Question 24).

In Β§5 we characterize the graphs G𝐺G with c​(G)+c¯​(G)=n​(G)+2𝑐𝐺¯𝑐𝐺𝑛𝐺2c(G)+\overline{c}(G)=n(G)+2 (Theorem 34(b)), thereby showing that c​(G)+c¯​(G)β‰₯n​(G)+3𝑐𝐺¯𝑐𝐺𝑛𝐺3c(G)+\overline{c}(G)\geq n(G)+3 for all graphs outside of a well-defined class of exceptions. In particular, we show that c​(G)+c¯​(G)β‰₯n​(G)+3𝑐𝐺¯𝑐𝐺𝑛𝐺3c(G)+\overline{c}(G)\geq n(G)+3 if ω​(G)+α​(G)<n​(G)πœ”πΊπ›ΌπΊπ‘›πΊ\omega(G)+\alpha(G)<n(G) (Lemma 33, Corollary 35(c)).

Some of our results (Theorems 6, 7, 8, 17, 20, and 23, perhaps in a less general formulation) were stated without proof in the earlier paper [2] or in the abstract [1]. Some were presented by the third author in invited addresses to the Mid-Atlantic Mathematical Logic Seminar, Dartmouth College, Hanover, New Hampshire, October 2003, and to the CombinaTexas Conference, Houston, Texas, April 2009.

2 Generalities

In this section we establish some basic properties of the function fr​(t,n)subscriptπ‘“π‘Ÿπ‘‘π‘›f_{r}(t,n) such as monotonicity (Theorem 3), the trivial bounds (Theorem 6), and an asymptotic lower bound (Theorem 7). We begin by disposing of the trivial cases where n≀r+1π‘›π‘Ÿ1n\leq r+1. Recall that the maximum size of a t𝑑t-partite graph of order n𝑛n is attained by the TurΓ‘n graph Tn,tsubscript𝑇𝑛𝑑T_{n,t}.

Theorem 2.

For any integers r,tβ‰₯2π‘Ÿπ‘‘2r,t\geq 2 we have

fr​(t,n)={tΒ if ​1≀n<r,(tβˆ’1)​n+1Β if ​n=r,(n+12)+(tβˆ’2)​(n2)βˆ’e​(Tn,t)Β if ​n=r+1,⌊(n+1)24βŒ‹Β if ​n=r+1​ and ​t=2.subscriptπ‘“π‘Ÿπ‘‘π‘›cases𝑑 ifΒ 1π‘›π‘Ÿπ‘‘1𝑛1Β ifΒ π‘›π‘Ÿbinomial𝑛12𝑑2binomial𝑛2𝑒subscript𝑇𝑛𝑑 ifΒ π‘›π‘Ÿ1superscript𝑛124Β ifΒ π‘›π‘Ÿ1Β and 𝑑2f_{r}(t,n)=\begin{cases}t&\text{ if }1\leq n<r,\\ (t-1)n+1&\text{ if }n=r,\\ \binom{n+1}{2}+(t-2)\binom{n}{2}-e(T_{n,t})&\text{ if }n=r+1,\\ \left\lfloor\frac{(n+1)^{2}}{4}\right\rfloor&\text{ if }n=r+1\text{ and }t=2.\\ \end{cases}
Proof.

Only the case n=r+1π‘›π‘Ÿ1n=r+1 is in need of a proof. Consider a factorization H1,…,Htsubscript𝐻1…subscript𝐻𝑑H_{1},\dots,H_{t} of Knnβˆ’1=(V,(Vnβˆ’1))superscriptsubscript𝐾𝑛𝑛1𝑉binomial𝑉𝑛1K_{n}^{n-1}=\left(V,\binom{V}{n-1}\right). Let Vi={v∈V:Vβˆ–{v}∈E​(Hi)}subscript𝑉𝑖conditional-set𝑣𝑉𝑉𝑣𝐸subscript𝐻𝑖V_{i}=\{v\in V:V\setminus\{v\}\in E(H_{i})\} and let ni=|Vi|=e​(Hi)subscript𝑛𝑖subscript𝑉𝑖𝑒subscript𝐻𝑖n_{i}=|V_{i}|=e(H_{i}), so that βˆ‘i=1tni=|V|=nsuperscriptsubscript𝑖1𝑑subscript𝑛𝑖𝑉𝑛\sum_{i=1}^{t}n_{i}=|V|=n.

First suppose ni<nsubscript𝑛𝑖𝑛n_{i}<n for all i𝑖i. Then for x∈Vπ‘₯𝑉x\in V the set Vβˆ–{x}𝑉π‘₯V\setminus\{x\} is a maximal clique only for the unique factor Hisubscript𝐻𝑖H_{i} having Vβˆ–{x}𝑉π‘₯V\setminus\{x\} as an edge, while for {x,y}∈(V2)π‘₯𝑦binomial𝑉2\{x,y\}\in\binom{V}{2} the set Vβˆ–{x,y}𝑉π‘₯𝑦V\setminus\{x,y\} is a maximal clique of Hisubscript𝐻𝑖H_{i} just in case xβˆ‰Viπ‘₯subscript𝑉𝑖x\notin V_{i} and yβˆ‰Vi𝑦subscript𝑉𝑖y\notin V_{i}. Hence the factors Hisubscript𝐻𝑖H_{i} have a total of n𝑛n maximal cliques of size nβˆ’1𝑛1n-1 and (tβˆ’2)​(n2)+βˆ‘i=1t(ni2)𝑑2binomial𝑛2superscriptsubscript𝑖1𝑑binomialsubscript𝑛𝑖2(t-2)\binom{n}{2}+\sum_{i=1}^{t}\binom{n_{i}}{2} maximal cliques of size nβˆ’2𝑛2n-2, so that in this case

βˆ‘i=1tc​(Hi)=n+(tβˆ’2)​(n2)+βˆ‘i=1t(ni2)=(n+12)+(tβˆ’2)​(n2)βˆ’[(n2)βˆ’βˆ‘i=1t(ni2)].superscriptsubscript𝑖1𝑑𝑐subscript𝐻𝑖𝑛𝑑2binomial𝑛2superscriptsubscript𝑖1𝑑binomialsubscript𝑛𝑖2binomial𝑛12𝑑2binomial𝑛2delimited-[]binomial𝑛2superscriptsubscript𝑖1𝑑binomialsubscript𝑛𝑖2\begin{split}\sum_{i=1}^{t}c(H_{i})&=n+(t-2)\binom{n}{2}+\sum_{i=1}^{t}\binom{n_{i}}{2}\\ &=\binom{n+1}{2}+(t-2)\binom{n}{2}-\left[\binom{n}{2}-\sum_{i=1}^{t}\binom{n_{i}}{2}\right].\end{split} (1)

On the other hand, if n1=nsubscript𝑛1𝑛n_{1}=n, then

βˆ‘i=1tc​(Hi)=1+(tβˆ’1)​(n2)=n+(tβˆ’2)​(n2)+(nβˆ’12),superscriptsubscript𝑖1𝑑𝑐subscript𝐻𝑖1𝑑1binomial𝑛2𝑛𝑑2binomial𝑛2binomial𝑛12\sum_{i=1}^{t}c(H_{i})=1+(t-1)\binom{n}{2}=n+(t-2)\binom{n}{2}+\binom{n-1}{2},

the same number of maximal cliques as when n1=nβˆ’1subscript𝑛1𝑛1n_{1}=n-1. Hence we may assume that ni<nsubscript𝑛𝑖𝑛n_{i}<n for all i𝑖i. The quantity (1) is minimized when (n2)βˆ’βˆ‘i=1t(ni2)binomial𝑛2superscriptsubscript𝑖1𝑑binomialsubscript𝑛𝑖2\binom{n}{2}-\sum_{i=1}^{t}\binom{n_{i}}{2} is maximized, i.e., when (n2)βˆ’βˆ‘i=1t(ni2)=e​(Tn,t)binomial𝑛2superscriptsubscript𝑖1𝑑binomialsubscript𝑛𝑖2𝑒subscript𝑇𝑛𝑑\binom{n}{2}-\sum_{i=1}^{t}\binom{n_{i}}{2}=e(T_{n,t}). This shows that fnβˆ’1​(t,n)=(n+12)+(tβˆ’2)​(n2)βˆ’e​(Tn,t)subscript𝑓𝑛1𝑑𝑛binomial𝑛12𝑑2binomial𝑛2𝑒subscript𝑇𝑛𝑑f_{n-1}(t,n)=\binom{n+1}{2}+(t-2)\binom{n}{2}-e(T_{n,t}). Finally, setting t=2𝑑2t=2, we have

fnβˆ’1​(2,n)=(n+12)βˆ’e​(Tn,2)=(n+12)βˆ’βŒŠn24βŒ‹=⌊(n+1)24βŒ‹.subscript𝑓𝑛12𝑛binomial𝑛12𝑒subscript𝑇𝑛2binomial𝑛12superscript𝑛24superscript𝑛124f_{n-1}(2,n)=\binom{n+1}{2}-e(T_{n,2})=\binom{n+1}{2}-\left\lfloor\frac{n^{2}}{4}\right\rfloor=\left\lfloor\frac{(n+1)^{2}}{4}\right\rfloor.

∎

Theorem 3.

fr​(t,n)≀fr​(t,n+1)subscriptπ‘“π‘Ÿπ‘‘π‘›subscriptπ‘“π‘Ÿπ‘‘π‘›1f_{r}(t,n)\leq f_{r}(t,n+1).

Proof.

If H𝐻H is an induced subhypergraph of G𝐺G then c​(H)≀c​(G)𝑐𝐻𝑐𝐺c(H)\leq c(G). ∎

The inequality is not necessarily strict, e.g., if n𝑛n is odd then f2​(n,n)=f2​(n,n+1)=(n+12)subscript𝑓2𝑛𝑛subscript𝑓2𝑛𝑛1binomial𝑛12f_{2}(n,n)=f_{2}(n,n+1)=\binom{n+1}{2} by Corollary 15(b). See Theorem 19 for more examples.

Question 4.

Does the inequality fr​(t,n)≀fr​(t+1,n)subscriptπ‘“π‘Ÿπ‘‘π‘›subscriptπ‘“π‘Ÿπ‘‘1𝑛f_{r}(t,n)\leq f_{r}(t+1,n) hold for all integers r,tβ‰₯2π‘Ÿπ‘‘2r,t\geq 2 and nβ‰₯1𝑛1n\geq 1?

At any rate we don’t have strict inequality, since f2​(3,9)=f2​(4,9)=12subscript𝑓239subscript𝑓24912f_{2}(3,9)=f_{2}(4,9)=12 by Theorems 23 and 25.

An obvious way to get a t𝑑t-coloring from a (t+1)𝑑1(t+1)-coloring is by merging two colors. This can increase the number of maximal monochromatic cliques, i.e., if G1subscript𝐺1G_{1} and G2subscript𝐺2G_{2} are edge-disjoint graphs on the same vertex set V𝑉V, we can have c​(G1βˆͺG2)>c​(G1)+c​(G2)𝑐subscript𝐺1subscript𝐺2𝑐subscript𝐺1𝑐subscript𝐺2c(G_{1}\cup G_{2})>c(G_{1})+c(G_{2}). For example, let V𝑉V be the set of vertices of a unit cube, and let

E​(G1)𝐸subscript𝐺1\displaystyle E(G_{1}) ={{x,y}∈(V2):d​(x,y)=1},absentconditional-setπ‘₯𝑦binomial𝑉2𝑑π‘₯𝑦1\displaystyle=\{\{x,y\}\in\binom{V}{2}:d(x,y)=1\},
E​(G2)𝐸subscript𝐺2\displaystyle E(G_{2}) ={{x,y}∈(V2):d​(x,y)=2}.absentconditional-setπ‘₯𝑦binomial𝑉2𝑑π‘₯𝑦2\displaystyle=\{\{x,y\}\in\binom{V}{2}:d(x,y)=\sqrt{2}\}.

Then G1=Q3subscript𝐺1subscript𝑄3G_{1}=Q_{3}, G2=2​K4subscript𝐺22subscript𝐾4G_{2}=2K_{4}, and G1βˆͺG2=K2,2,2,2subscript𝐺1subscript𝐺2subscript𝐾2222G_{1}\cup G_{2}=K_{2,2,2,2}, so that c​(G1)=12𝑐subscript𝐺112c(G_{1})=12, c​(G2)=2𝑐subscript𝐺22c(G_{2})=2, and c​(G1βˆͺG2)=16𝑐subscript𝐺1subscript𝐺216c(G_{1}\cup G_{2})=16.

Lemma 5.

Consider a coloring of KnrsuperscriptsubscriptπΎπ‘›π‘ŸK_{n}^{r}. For each color i𝑖i, let cisubscript𝑐𝑖c_{i} denote the number of maximal i𝑖i-cliques. Then, for any two colors iβ‰ j𝑖𝑗i\neq j, we have ci​cjβ‰₯(nrβˆ’1)subscript𝑐𝑖subscript𝑐𝑗binomialπ‘›π‘Ÿ1c_{i}c_{j}\geq\binom{n}{r-1}.

Proof.

Let V𝑉V be the vertex set. For each color i𝑖i let Cisubscript𝐢𝑖C_{i} be the set of all maximal i𝑖i-cliques. Each set X∈(Vrβˆ’1)𝑋binomialπ‘‰π‘Ÿ1X\in\binom{V}{r-1} is an i𝑖i-clique; let fi​(X)subscript𝑓𝑖𝑋f_{i}(X) be some maximal i𝑖i-clique containing X𝑋X. If iβ‰ j𝑖𝑗i\neq j, then the map X↦(fi​(X),fj​(X))maps-to𝑋subscript𝑓𝑖𝑋subscript𝑓𝑗𝑋X\mapsto(f_{i}(X),f_{j}(X)) is an injection from (Vrβˆ’1)binomialπ‘‰π‘Ÿ1\binom{V}{r-1} to CiΓ—Cjsubscript𝐢𝑖subscript𝐢𝑗C_{i}\times C_{j}. ∎

Theorem 6.

For all integers r,tβ‰₯2π‘Ÿπ‘‘2r,t\geq 2 and nβ‰₯rβˆ’1π‘›π‘Ÿ1n\geq r-1,

t​(nrβˆ’1)12≀fr​(t,n)≀(tβˆ’1)​(nrβˆ’1)+1𝑑superscriptbinomialπ‘›π‘Ÿ112subscriptπ‘“π‘Ÿπ‘‘π‘›π‘‘1binomialπ‘›π‘Ÿ11t\binom{n}{r-1}^{\frac{1}{2}}\leq f_{r}(t,n)\leq(t-1)\binom{n}{r-1}+1.

Proof.

For the upper bound, give all edges the same color.

For the lower bound, let KnrsuperscriptsubscriptπΎπ‘›π‘ŸK_{n}^{r} be t𝑑t-colored so that fr​(t,n)=c1+β‹―+ctsubscriptπ‘“π‘Ÿπ‘‘π‘›subscript𝑐1β‹―subscript𝑐𝑑f_{r}(t,n)=c_{1}+\dots+c_{t} where cisubscript𝑐𝑖c_{i} is the number of maximal i𝑖i-cliques. By Lemma 5 we have ci​cjβ‰₯(nrβˆ’1)subscript𝑐𝑖subscript𝑐𝑗binomialπ‘›π‘Ÿ1c_{i}c_{j}\geq\binom{n}{r-1} for iβ‰ j𝑖𝑗i\neq j. Hence:

(c1​c2)​(c2​c3)​…​(ct​c1)β‰₯(nrβˆ’1)t;subscript𝑐1subscript𝑐2subscript𝑐2subscript𝑐3…subscript𝑐𝑑subscript𝑐1superscriptbinomialπ‘›π‘Ÿ1𝑑(c_{1}c_{2})(c_{2}c_{3})\dots(c_{t}c_{1})\geq\binom{n}{r-1}^{t};
c1​c2​…​ctβ‰₯(nrβˆ’1)t/2;subscript𝑐1subscript𝑐2…subscript𝑐𝑑superscriptbinomialπ‘›π‘Ÿ1𝑑2c_{1}c_{2}\dots c_{t}\geq\binom{n}{r-1}^{t/2};
fr​(t,n)=c1+c2+β‹―+ctβ‰₯t​(c1​c2​…​ct)1/tβ‰₯t​(nrβˆ’1)1/2.subscriptπ‘“π‘Ÿπ‘‘π‘›subscript𝑐1subscript𝑐2β‹―subscript𝑐𝑑𝑑superscriptsubscript𝑐1subscript𝑐2…subscript𝑐𝑑1𝑑𝑑superscriptbinomialπ‘›π‘Ÿ112f_{r}(t,n)=c_{1}+c_{2}+\dots+c_{t}\geq t(c_{1}c_{2}\dots c_{t})^{1/t}\geq t\binom{n}{r-1}^{1/2}.

∎

Theorem 7.

For any integers sβ‰₯1𝑠1s\geq 1 and tβ‰₯2𝑑2t\geq 2, and for any Ξ΅>0πœ€0\varepsilon>0, we have fs+1​(t,n)>nsβˆ’Ξ΅subscript𝑓𝑠1𝑑𝑛superscriptπ‘›π‘ πœ€f_{s+1}(t,n)>n^{s-\varepsilon} for all sufficiently large n𝑛n.

Proof.

Choose δ𝛿\delta so that 0<Ξ΄<min⁑{Ξ΅,s}0π›Ώπœ€π‘ 0<\delta<\min\{\varepsilon,s\}. Let Ξ±=sβˆ’Ξ΄s𝛼𝑠𝛿𝑠\alpha=\frac{s-\delta}{s} and k=⌈sβˆ’Ξ΄1βˆ’Ξ±βŒ‰π‘˜π‘ π›Ώ1𝛼k=\left\lceil\frac{s-\delta}{1-\alpha}\right\rceil. Define M𝑀M so that

mβ‰₯M⟹m​(mβˆ’1)​(mβˆ’2)​⋯​(mβˆ’s+1)msβ‰₯tβˆ’32tβˆ’1.π‘šπ‘€π‘šπ‘š1π‘š2β‹―π‘šπ‘ 1superscriptπ‘šπ‘ π‘‘32𝑑1m\geq M\implies\frac{m(m-1)(m-2)\cdots(m-s+1)}{m^{s}}\geq\frac{t-\frac{3}{2}}{t-1}.

By Ramsey’s theorem there is an integer Rβ‰₯kπ‘…π‘˜R\geq k such that any t𝑑t-coloring of the edges of the complete hypergraph KRs+1superscriptsubscript𝐾𝑅𝑠1K_{R}^{s+1} contains a monochromatic clique of size kπ‘˜k. Let Ξ²=min⁑{tβˆ’32s!,1(Rk)}𝛽𝑑32𝑠1binomialπ‘…π‘˜\beta=\min\left\{\frac{t-\frac{3}{2}}{s!},\frac{1}{\binom{R}{k}}\right\}.

Let nβˆˆβ„•π‘›β„•n\in\mathbb{N} and consider a t𝑑t-coloring of the edges of Kns+1superscriptsubscript𝐾𝑛𝑠1K_{n}^{s+1}. Let cisubscript𝑐𝑖c_{i} be the number of maximal i𝑖i-cliques. Let mπ‘šm be the size of the largest monochromatic clique.

If there is a j𝑗j-clique of size mπ‘šm, then for each color iβ‰ j𝑖𝑗i\neq j there are at least (ms)binomialπ‘šπ‘ \binom{m}{s} maximal i𝑖i-cliques. If nΞ±β‰₯Msuperscript𝑛𝛼𝑀n^{\alpha}\geq M then

βˆ‘i=1tci>(tβˆ’1)​(ms)β‰₯tβˆ’32s!​msβ‰₯β​msβ‰₯β​nα​s=β​nsβˆ’Ξ΄.superscriptsubscript𝑖1𝑑subscript𝑐𝑖𝑑1binomialπ‘šπ‘ π‘‘32𝑠superscriptπ‘šπ‘ π›½superscriptπ‘šπ‘ π›½superscript𝑛𝛼𝑠𝛽superscript𝑛𝑠𝛿\sum_{i=1}^{t}c_{i}>(t-1)\binom{m}{s}\geq\frac{t-\frac{3}{2}}{s!}m^{s}\geq\beta m^{s}\geq\beta n^{\alpha s}=\beta n^{s-\delta}.

Let p𝑝p be the number of monochromatic cliques of size kπ‘˜k. Since every monochromatic clique is contained in a maximal monochromatic clique, p≀(mk)β€‹βˆ‘i=1tci𝑝binomialπ‘šπ‘˜superscriptsubscript𝑖1𝑑subscript𝑐𝑖p\leq\binom{m}{k}\sum_{i=1}^{t}c_{i}. On the other hand, if nβ‰₯R𝑛𝑅n\geq R, then pβ‰₯(nk)/(Rk)β‰₯β​(nk)𝑝binomialπ‘›π‘˜binomialπ‘…π‘˜π›½binomialπ‘›π‘˜p\geq\binom{n}{k}/\binom{R}{k}\geq\beta\binom{n}{k} and we have

βˆ‘i=1tciβ‰₯β​(nk)(mk)β‰₯β​nkmkβ‰₯β​nknα​kβ‰₯β​nsβˆ’Ξ΄.superscriptsubscript𝑖1𝑑subscript𝑐𝑖𝛽binomialπ‘›π‘˜binomialπ‘šπ‘˜π›½superscriptπ‘›π‘˜superscriptπ‘šπ‘˜π›½superscriptπ‘›π‘˜superscriptπ‘›π›Όπ‘˜π›½superscript𝑛𝑠𝛿\sum_{i=1}^{t}c_{i}\geq\frac{\beta\binom{n}{k}}{\binom{m}{k}}\geq\frac{\beta n^{k}}{m^{k}}\geq\frac{\beta n^{k}}{n^{\alpha k}}\geq\beta n^{s-\delta}.

Thus fs+1​(t,n)β‰₯β​nsβˆ’Ξ΄>nsβˆ’Ξ΅subscript𝑓𝑠1𝑑𝑛𝛽superscript𝑛𝑠𝛿superscriptπ‘›π‘ πœ€f_{s+1}(t,n)\geq\beta n^{s-\delta}>n^{s-\varepsilon} for sufficiently large n𝑛n. ∎

3 A triple system and its complement

In this section we find upper bounds for f3​(2,n)subscript𝑓32𝑛f_{3}(2,n). For n>12𝑛12n>12 our best result is f3​(2,n)β‰€βŒŠn24βŒ‹+5subscript𝑓32𝑛superscript𝑛245f_{3}(2,n)\leq\left\lfloor\frac{n^{2}}{4}\right\rfloor+5 (Theorem 12). We know the exact value of f3​(2,n)subscript𝑓32𝑛f_{3}(2,n) only for n≀7𝑛7n\leq 7. By tedious case analysis (which we omit) we have verified that Theorem 8 is optimal for 2≀n≀62𝑛62\leq n\leq 6, so that f3​(2,n)=⌊(n+1)24βŒ‹subscript𝑓32𝑛superscript𝑛124f_{3}(2,n)=\left\lfloor\frac{(n+1)^{2}}{4}\right\rfloor in those cases, and that Theorem 11 is optimal for n=7𝑛7n=7, so that f3​(2,7)=14subscript𝑓32714f_{3}(2,7)=14. See Theorem 2 for the trivial cases n≀4𝑛4n\leq 4.

Theorem 8.

f3​(2,n)β‰€βŒŠ(n+1)24βŒ‹subscript𝑓32𝑛superscript𝑛124f_{3}(2,n)\leq\left\lfloor\frac{(n+1)^{2}}{4}\right\rfloor for nβ‰₯2𝑛2n\geq 2.

Proof.

Let n=n1+n2𝑛subscript𝑛1subscript𝑛2n=n_{1}+n_{2} where n1,n2>0subscript𝑛1subscript𝑛20n_{1},n_{2}>0. Let V=V1βˆͺV2,V1∩V2=βˆ…,|Vi|=niformulae-sequence𝑉subscript𝑉1subscript𝑉2formulae-sequencesubscript𝑉1subscript𝑉2subscript𝑉𝑖subscript𝑛𝑖V=V_{1}\cup V_{2},V_{1}\cap V_{2}=\varnothing,|V_{i}|=n_{i}, and consider the triple system H=(V,E)𝐻𝑉𝐸H=(V,E) where E={e∈(V3):|e∩V1|β‰₯2}𝐸conditional-set𝑒binomial𝑉3𝑒subscript𝑉12E=\{e\in\binom{V}{3}:|e\cap V_{1}|\geq 2\}. Then c​(H)=n2+(n22)𝑐𝐻subscript𝑛2binomialsubscript𝑛22c(H)=n_{2}+\binom{n_{2}}{2}; the maximal cliques are the sets V1βˆͺ{v}subscript𝑉1𝑣V_{1}\cup\{v\}, v∈V2𝑣subscript𝑉2v\in V_{2}, and the 222-element subsets of V2subscript𝑉2V_{2}. Taking n1=⌊n2βŒ‹subscript𝑛1𝑛2n_{1}=\left\lfloor\frac{n}{2}\right\rfloor and n2=⌈n2βŒ‰subscript𝑛2𝑛2n_{2}=\left\lceil\frac{n}{2}\right\rceil, we see that

f3​(2,n)≀c​(H)+c¯​(H)=⌈n2βŒ‰+(⌈n2βŒ‰2)+⌊n2βŒ‹+(⌊n2βŒ‹2)=⌊(n+1)24βŒ‹.subscript𝑓32𝑛𝑐𝐻¯𝑐𝐻𝑛2binomial𝑛22𝑛2binomial𝑛22superscript𝑛124f_{3}(2,n)\leq c(H)+\overline{c}(H)=\left\lceil\frac{n}{2}\right\rceil+\binom{\left\lceil\frac{n}{2}\right\rceil}{2}+\left\lfloor\frac{n}{2}\right\rfloor+\binom{\left\lfloor\frac{n}{2}\right\rfloor}{2}=\left\lfloor\frac{(n+1)^{2}}{4}\right\rfloor.

Alternatively, let H=(V,E)𝐻𝑉𝐸H=(V,E) where V=[n]𝑉delimited-[]𝑛V=[n] and E={{x,y,z}∈(V3):x<y<z,y​ odd}𝐸conditional-setπ‘₯𝑦𝑧binomial𝑉3formulae-sequenceπ‘₯𝑦𝑧𝑦 oddE=\{\{x,y,z\}\in\binom{V}{3}:x<y<z,\ y\text{ odd}\}. For integers a<bπ‘Žπ‘a<b of the same parity, let Xa,bsubscriptπ‘‹π‘Žπ‘X_{a,b} be the set consisting of aπ‘Ža and b𝑏b and all integers of the opposite parity between aπ‘Ža and b𝑏b; e.g., X3,9={3,4,6,8,9}subscript𝑋3934689X_{3,9}=\{3,4,6,8,9\}. Then the maximal cliques (anticliques) of H𝐻H are just the sets Xa,b∩[n]subscriptπ‘‹π‘Žπ‘delimited-[]𝑛X_{a,b}\cap[n] where aπ‘Ža and b𝑏b are even (odd) integers, 0≀a<b≀n+10π‘Žπ‘π‘›10\leq a<b\leq n+1. It follows that

c​(H)+c¯​(H)=(⌈n2βŒ‰+12)+(⌊n2βŒ‹+12)=⌊(n+1)24βŒ‹.𝑐𝐻¯𝑐𝐻binomial𝑛212binomial𝑛212superscript𝑛124c(H)+\overline{c}(H)=\binom{\left\lceil\frac{n}{2}\right\rceil+1}{2}+\binom{\left\lfloor\frac{n}{2}\right\rfloor+1}{2}=\left\lfloor\frac{(n+1)^{2}}{4}\right\rfloor.

∎

In fact f3​(2,n)=⌊(n+1)24βŒ‹subscript𝑓32𝑛superscript𝑛124f_{3}(2,n)=\left\lfloor\frac{(n+1)^{2}}{4}\right\rfloor holds for 2≀n≀62𝑛62\leq n\leq 6 but not for nβ‰₯7𝑛7n\geq 7; see Theorems 11 and 12.

Lemma 9.

A triple system H𝐻H of order nβ‰₯2𝑛2n\geq 2 can be extended to a triple system Hβˆ—superscript𝐻H^{*} of order n+1𝑛1n+1 such that c​(Hβˆ—)=c​(H)+d​(H)+1𝑐superscript𝐻𝑐𝐻𝑑𝐻1c(H^{*})=c(H)+d(H)+1, d​(Hβˆ—)=d​(H)+1𝑑superscript𝐻𝑑𝐻1d(H^{*})=d(H)+1, c¯​(Hβˆ—)=c¯​(H)¯𝑐superscript𝐻¯𝑐𝐻\overline{c}(H^{*})=\overline{c}(H), and d¯​(Hβˆ—)=d¯​(H)¯𝑑superscript𝐻¯𝑑𝐻\overline{d}(H^{*})=\overline{d}(H).

Proof.

Let H=(V,E)𝐻𝑉𝐸H=(V,E) be a triple system of order nβ‰₯2𝑛2n\geq 2. Choose a vertex v∈V𝑣𝑉v\in V with D​(v)=d​(H)𝐷𝑣𝑑𝐻D(v)=d(H). Choose a new vertex vβˆ—βˆ‰Vsuperscript𝑣𝑉v^{*}\notin V and let Hβˆ—=(Vβˆ—,Eβˆ—)superscript𝐻superscript𝑉superscript𝐸H^{*}=(V^{*},E^{*}) where Vβˆ—=Vβˆͺ{vβˆ—}superscript𝑉𝑉superscript𝑣V^{*}=V\cup\{v^{*}\} and Eβˆ—=Eβˆͺ{{x,y,vβˆ—}:{x,y,v}∈E}superscript𝐸𝐸conditional-setπ‘₯𝑦superscript𝑣π‘₯𝑦𝑣𝐸E^{*}=E\cup\{\{x,y,v^{*}\}:\{x,y,v\}\in E\}.

The maximal anticliques of Hβˆ—superscript𝐻H^{*} are (1)1(1) the maximal anticliques of H𝐻H not containing v𝑣v and (2)2(2) the sets Xβˆͺ{vβˆ—}𝑋superscript𝑣X\cup\{v^{*}\} where X𝑋X is a maximal anticlique of H containing v𝑣v. Plainly c¯​(Hβˆ—)=c¯​(H)¯𝑐superscript𝐻¯𝑐𝐻\overline{c}(H^{*})=\overline{c}(H) and d¯​(Hβˆ—)=d¯​(H)¯𝑑superscript𝐻¯𝑑𝐻\overline{d}(H^{*})=\overline{d}(H).

The maximal cliques of Hβˆ—superscript𝐻H^{*} are (1)1(1) the maximal cliques of H𝐻H, and (2)2(2) the sets (Xβˆ–{v})βˆͺ{vβˆ—}𝑋𝑣superscript𝑣(X\setminus\{v\})\cup\{v^{*}\} where X𝑋X is a maximal clique of H𝐻H containing v𝑣v, and (3)3(3) the set {v,vβˆ—}𝑣superscript𝑣\{v,v^{*}\}. Now it can be seen that c​(Hβˆ—)=c​(H)+DH​(v)+1=c​(H)+d​(H)+1𝑐superscript𝐻𝑐𝐻subscript𝐷𝐻𝑣1𝑐𝐻𝑑𝐻1c(H^{*})=c(H)+D_{H}(v)+1=c(H)+d(H)+1, and that DHβˆ—β€‹(v)=DHβˆ—β€‹(vβˆ—)=DH​(v)+1=d​(H)+1subscript𝐷superscript𝐻𝑣subscript𝐷superscript𝐻superscript𝑣subscript𝐷𝐻𝑣1𝑑𝐻1D_{H^{*}}(v)=D_{H^{*}}(v^{*})=D_{H}(v)+1=d(H)+1, while DHβˆ—β€‹(u)β‰₯DH​(u)+1β‰₯d​(H)+1subscript𝐷superscript𝐻𝑒subscript𝐷𝐻𝑒1𝑑𝐻1D_{H^{*}}(u)\geq D_{H}(u)+1\geq d(H)+1 for u∈Vβˆ–{v}𝑒𝑉𝑣u\in V\setminus\{v\}, whence d​(Hβˆ—)=d​(H)+1𝑑superscript𝐻𝑑𝐻1d(H^{*})=d(H)+1. ∎

Lemma 10.

If there is a triple system H𝐻H of order nβ‰₯2𝑛2n\geq 2 with c​(H)+c¯​(H)=kπ‘π»Β―π‘π»π‘˜c(H)+\overline{c}(H)=k and d​(H)=⌊m2βŒ‹π‘‘π»π‘š2d(H)=\left\lfloor\frac{m}{2}\right\rfloor and d¯​(H)=⌈m2βŒ‰Β―π‘‘π»π‘š2\overline{d}(H)=\left\lceil\frac{m}{2}\right\rceil, then there is a triple system Hβ€²superscript𝐻′H^{\prime} of order n+1𝑛1n+1 with c​(Hβ€²)+c¯​(Hβ€²)=k+⌊m+22βŒ‹π‘superscript𝐻′¯𝑐superscriptπ»β€²π‘˜π‘š22c(H^{\prime})+\overline{c}(H^{\prime})=k+\left\lfloor\frac{m+2}{2}\right\rfloor and d​(Hβ€²)=⌊m+12βŒ‹π‘‘superscriptπ»β€²π‘š12d(H^{\prime})=\left\lfloor\frac{m+1}{2}\right\rfloor and d¯​(Hβ€²)=⌈m+12βŒ‰Β―π‘‘superscriptπ»β€²π‘š12\overline{d}(H^{\prime})=\left\lceil\frac{m+1}{2}\right\rceil.

Proof.

By Lemma 9 there is a triple system Hβˆ—superscript𝐻H^{*} of order n+1𝑛1n+1 such that c​(Hβˆ—)+c¯​(Hβˆ—)=c​(H)+d​(H)+1+c¯​(H)=k+⌊m+22βŒ‹π‘superscript𝐻¯𝑐superscript𝐻𝑐𝐻𝑑𝐻1Β―π‘π»π‘˜π‘š22c(H^{*})+\overline{c}(H^{*})=c(H)+d(H)+1+\overline{c}(H)=k+\left\lfloor\frac{m+2}{2}\right\rfloor and d​(Hβˆ—)=d​(H)+1=⌊m+22βŒ‹=⌈m+12βŒ‰π‘‘superscript𝐻𝑑𝐻1π‘š22π‘š12d(H^{*})=d(H)+1=\left\lfloor\frac{m+2}{2}\right\rfloor=\left\lceil\frac{m+1}{2}\right\rceil and d¯​(Hβˆ—)=d¯​(H)=⌈m2βŒ‰=⌊m+12βŒ‹Β―π‘‘superscriptπ»Β―π‘‘π»π‘š2π‘š12\overline{d}(H^{*})=\overline{d}(H)=\left\lceil\frac{m}{2}\right\rceil=\left\lfloor\frac{m+1}{2}\right\rfloor. Let Hβ€²superscript𝐻′H^{\prime} be the complement of Hβˆ—superscript𝐻H^{*}. ∎

Theorem 11.

f3​(2,n)β‰€βŒŠ(n+1)24βŒ‹βˆ’2subscript𝑓32𝑛superscript𝑛1242f_{3}(2,n)\leq\left\lfloor\frac{(n+1)^{2}}{4}\right\rfloor-2 for nβ‰₯7𝑛7n\geq 7.

Proof.

We prove by induction that for each nβ‰₯7𝑛7n\geq 7 there is a triple system H𝐻H of order n𝑛n with c​(H)+c¯​(H)=⌊(n+1)24βŒ‹βˆ’2𝑐𝐻¯𝑐𝐻superscript𝑛1242c(H)+\overline{c}(H)=\left\lfloor\frac{(n+1)^{2}}{4}\right\rfloor-2 and d​(H)=⌊n2βŒ‹π‘‘π»π‘›2d(H)=\left\lfloor\frac{n}{2}\right\rfloor and d¯​(H)=⌈n2βŒ‰Β―π‘‘π»π‘›2\overline{d}(H)=\left\lceil\frac{n}{2}\right\rceil. The inductive step follows from Lemma 10 since

⌊(n+1)24βŒ‹βˆ’2+⌊(n+2)2βŒ‹=⌊(n+2)24βŒ‹βˆ’2.superscript𝑛1242𝑛22superscript𝑛2242\left\lfloor\frac{(n+1)^{2}}{4}\right\rfloor-2+\left\lfloor\frac{(n+2)}{2}\right\rfloor=\left\lfloor\frac{(n+2)^{2}}{4}\right\rfloor-2.

For the base case n=7𝑛7n=7, consider the Fano plane as a hypergraph H with the points as vertices and the lines as edges. Then the maximal cliques are the lines, and the maximal anticliques are the complements of lines. Thus c​(H)=c¯​(H)=7𝑐𝐻¯𝑐𝐻7c(H)=\overline{c}(H)=7, and d​(H)=3𝑑𝐻3d(H)=3 and d¯​(H)=4¯𝑑𝐻4\overline{d}(H)=4; each vertex is in exactly three maximal cliques and four maximal anticliques. ∎

In fact f3​(2,7)=14subscript𝑓32714f_{3}(2,7)=14; the proof of f3​(2,7)β‰₯14subscript𝑓32714f_{3}(2,7)\geq 14 is a tedious case analysis which we omit. We do not know if equality holds in Theorem 11 when 8≀n≀148𝑛148\leq n\leq 14. For nβ‰₯15𝑛15n\geq 15 a better bound is given by Theorem 12.

Theorem 12.

f3​(2,n)β‰€βŒŠn24βŒ‹+5subscript𝑓32𝑛superscript𝑛245f_{3}(2,n)\leq\left\lfloor\frac{n^{2}}{4}\right\rfloor+5.

Proof.

The cases with n≀5𝑛5n\leq 5, while easily verified, are of no interest. For nβ‰₯6𝑛6n\geq 6 we prove by induction that there is a triple system H𝐻H of order n𝑛n with c​(H)+c¯​(H)=⌊n24βŒ‹+5𝑐𝐻¯𝑐𝐻superscript𝑛245c(H)+\overline{c}(H)=\left\lfloor\frac{n^{2}}{4}\right\rfloor+5 and d​(H)=⌊nβˆ’12βŒ‹π‘‘π»π‘›12d(H)=\left\lfloor\frac{n-1}{2}\right\rfloor and d¯​(H)=⌈nβˆ’12βŒ‰Β―π‘‘π»π‘›12\overline{d}(H)=\left\lceil\frac{n-1}{2}\right\rceil. The inductive step follows from Lemma 10 since

⌊n24βŒ‹+5+⌊n+12βŒ‹=⌊(n+1)24βŒ‹+5.superscript𝑛245𝑛12superscript𝑛1245\left\lfloor\frac{n^{2}}{4}\right\rfloor+5+\left\lfloor\frac{n+1}{2}\right\rfloor=\left\lfloor\frac{(n+1)^{2}}{4}\right\rfloor+5.

Now consider H=(V,E)𝐻𝑉𝐸H=(V,E) where V=[6]𝑉delimited-[]6V=[6] and E={{1,3,6},{1,4,5},{2,3,5},{2,3,6},{2,4,5},{2,4,6}}𝐸136145235236245246E=\{\{1,3,6\},\\ \{1,4,5\},\{2,3,5\},\{2,3,6\},\{2,4,5\},\{2,4,6\}\}.

We can visualize this as the 666 vertices and 666 of the 888 faces of a regular octahedron, the two missing faces meeting at one point. There are 999 maximal cliques, namely, the 666 edges and the 333 antipodal pairs

{1,2},{3,4},{5,6};123456\{1,2\},\{3,4\},\{5,6\};

and there are 555 maximal anticliques, namely, the missing faces

{1,3,5},{1,4,6},135146\{1,3,5\},\{1,4,6\},

and the complements of the antipodal pairs,

{3,4,5,6},{1,2,5,6},{1,2,3,4}.345612561234\{3,4,5,6\},\{1,2,5,6\},\{1,2,3,4\}.

Thus

c​(H)+c¯​(H)=9+5=14,𝑐𝐻¯𝑐𝐻9514c(H)+\overline{c}(H)=9+5=14,

while

d​(H)=D​(1)=3𝑑𝐻𝐷13d(H)=D(1)=3

and

d¯​(H)=D¯​(2)=2.¯𝑑𝐻¯𝐷22\overline{d}(H)=\overline{D}(2)=2.

The complement of H𝐻H satisfies the requirements of the base case n=6𝑛6n=6. ∎

4 Edge-colored graphs and projective planes

The main results of this section are upper bounds for f2​(t,n)subscript𝑓2𝑑𝑛f_{2}(t,n) from recursive constructions using projective planes. For example, we use projective planes of orders 444 and 555 to show that f2​(5,n)≀n+7subscript𝑓25𝑛𝑛7f_{2}(5,n)\leq n+7 for nβ‰₯37𝑛37n\geq 37 (Theorem 27). From Theorem 20 we see that determining the exact value of f2​(t,n)subscript𝑓2𝑑𝑛f_{2}(t,n) in all cases will be at least as hard as determining all possible orders of finite projective planes.

Theorem 13.

For all integers tβ‰₯2𝑑2t\geq 2 and nβ‰₯1𝑛1n\geq 1,

t​n≀(tβˆ’2)β€‹βŒˆnβŒ‰+⌈4​nβŒ‰β‰€f2​(t,n)≀(tβˆ’1)​n+1𝑑𝑛𝑑2𝑛4𝑛subscript𝑓2𝑑𝑛𝑑1𝑛1t\sqrt{n}\leq(t-2)\left\lceil\sqrt{n}\right\rceil+\left\lceil\sqrt{4n}\hskip 2.84544pt\right\rceil\leq f_{2}(t,n)\leq(t-1)n+1.

Proof.

The upper bound is from Theorem 6. For the lower bound, let f2​(t,n)=c1+c2+β‹―+ctsubscript𝑓2𝑑𝑛subscript𝑐1subscript𝑐2β‹―subscript𝑐𝑑f_{2}(t,n)=c_{1}+c_{2}+\dots+c_{t} where cisubscript𝑐𝑖c_{i} is the number of maximal i𝑖i-cliques in some t𝑑t-coloring of Knsubscript𝐾𝑛K_{n}; we may assume that c1≀c2≀⋯≀ctsubscript𝑐1subscript𝑐2β‹―subscript𝑐𝑑c_{1}\leq c_{2}\leq\dots\leq c_{t}. By Lemma 5 we have c1​c2β‰₯nsubscript𝑐1subscript𝑐2𝑛c_{1}c_{2}\geq n. It follows that ciβ‰₯c2β‰₯⌈nβŒ‰subscript𝑐𝑖subscript𝑐2𝑛c_{i}\geq c_{2}\geq\left\lceil\sqrt{n}\hskip 2.84544pt\right\rceil for i=3,…,t𝑖3…𝑑i=3,\dots,t, and c1+c2β‰₯⌈2​c1​c2βŒ‰β‰₯⌈4​nβŒ‰subscript𝑐1subscript𝑐22subscript𝑐1subscript𝑐24𝑛c_{1}+c_{2}\geq\left\lceil 2\sqrt{c_{1}c_{2}}\hskip 2.84544pt\right\rceil\geq\left\lceil\sqrt{4n}\hskip 2.84544pt\right\rceil, whence f2​(t,n)=c1+c2+β‹―+ctβ‰₯(tβˆ’2)β€‹βŒˆnβŒ‰+⌈4​nβŒ‰β‰₯t​nsubscript𝑓2𝑑𝑛subscript𝑐1subscript𝑐2β‹―subscript𝑐𝑑𝑑2𝑛4𝑛𝑑𝑛f_{2}(t,n)=c_{1}+c_{2}+\dots+c_{t}\geq(t-2)\left\lceil\sqrt{n}\hskip 2.84544pt\right\rceil+\left\lceil\sqrt{4n}\hskip 2.84544pt\right\rceil\geq t\sqrt{n}. ∎

Theorem 14.

f2​(t,n)β‰₯t​nβˆ’(n2)subscript𝑓2𝑑𝑛𝑑𝑛binomial𝑛2f_{2}(t,n)\geq tn-\binom{n}{2}, with equality if and only if n≀2β€‹βŒˆt2βŒ‰π‘›2𝑑2n\leq 2\left\lceil\frac{t}{2}\right\rceil.

Proof.

Let the edges of Knsubscript𝐾𝑛K_{n} be colored with t𝑑t colors. Consider the graph G=G1βˆͺG2βˆͺβ‹―βˆͺGt𝐺subscript𝐺1subscript𝐺2β‹―subscript𝐺𝑑G=G_{1}\cup G_{2}\cup\dots\cup G_{t} where G1,…,Gtsubscript𝐺1…subscript𝐺𝑑G_{1},\dots,G_{t} are vertex-disjoint graphs such that Gisubscript𝐺𝑖G_{i} is isomorphic to the spanning subgraph of Knsubscript𝐾𝑛K_{n} formed by the edges of color i𝑖i. Let kπ‘˜k be the number of components of G𝐺G. Then

βˆ‘i=1tc​(Gi)=c​(G)β‰₯kβ‰₯|V​(G)|βˆ’|E​(G)|=t​nβˆ’(n2).superscriptsubscript𝑖1𝑑𝑐subscriptπΊπ‘–π‘πΊπ‘˜π‘‰πΊπΈπΊπ‘‘π‘›binomial𝑛2\sum_{i=1}^{t}c(G_{i})=c(G)\geq k\geq|V(G)|-|E(G)|=tn-\binom{n}{2}. (2)

For equality to hold in (2) we must have c​(G)=kπ‘πΊπ‘˜c(G)=k, meaning that each component of G𝐺G is a clique, and also k=|V​(G)|βˆ’|E​(G)|π‘˜π‘‰πΊπΈπΊk=|V(G)|-|E(G)|, meaning that G𝐺G is acyclic. So equality holds just in case each component of G𝐺G is K1subscript𝐾1K_{1} or K2subscript𝐾2K_{2}, which means that the given coloring is a proper edge-coloring of Knsubscript𝐾𝑛K_{n}. Of course, Knsubscript𝐾𝑛K_{n} is t𝑑t-edge-colorable if and only if n≀2β€‹βŒˆt2βŒ‰π‘›2𝑑2n\leq 2\left\lceil\frac{t}{2}\right\rceil. ∎

Corollary 15.
(a)f2​(n+1,n)=n​(n+3)2​.(b)Β If ​n​ is odd, then ​f2​(n,n+1)=(n+12)=f2​(n,n)​.(c)Β If ​n​ is even, then ​f2​(n,n+1)>(n+12)=f2​(n,n)​.π‘Žsubscript𝑓2𝑛1𝑛𝑛𝑛32.𝑏 If 𝑛 is odd, thenΒ subscript𝑓2𝑛𝑛1binomial𝑛12subscript𝑓2𝑛𝑛.𝑐 If 𝑛 is even, thenΒ subscript𝑓2𝑛𝑛1binomial𝑛12subscript𝑓2𝑛𝑛.\begin{array}[]{ll}(a)&f_{2}(n+1,n)=\frac{n(n+3)}{2}$.$\\ (b)&$ If $n$ is odd, then $f_{2}(n,n+1)=\binom{n+1}{2}=f_{2}(n,n)$.$\\ (c)&$ If $n$ is even, then $f_{2}(n,n+1)>\binom{n+1}{2}=f_{2}(n,n)$.$\end{array}
Lemma 16.

Let G𝐺G be a t𝑑t-colored complete graph of order mπ‘šm, and let c=c1+β‹―+ct𝑐subscript𝑐1β‹―subscript𝑐𝑑c=c_{1}+\cdots+c_{t} where cisubscript𝑐𝑖c_{i} is the number of maximal i𝑖i-cliques in G𝐺G. If some vertex v𝑣v of G𝐺G is in a unique maximal i𝑖i-clique for each color i𝑖i, then f2​(t,n+mβˆ’1)≀f2​(t,n)+cβˆ’tsubscript𝑓2π‘‘π‘›π‘š1subscript𝑓2𝑑𝑛𝑐𝑑f_{2}(t,n+m-1)\leq f_{2}(t,n)+c-t for all nβˆˆβ„•π‘›β„•n\in\mathbb{N}.

Proof.

Let H𝐻H be an optimally t𝑑t-colored complete graph of order n𝑛n, so that f2​(t,n)=c1H+β‹―+ctHsubscript𝑓2𝑑𝑛superscriptsubscript𝑐1𝐻⋯superscriptsubscript𝑐𝑑𝐻f_{2}(t,n)=c_{1}^{H}+\dots+c_{t}^{H} where ciHsuperscriptsubscript𝑐𝑖𝐻c_{i}^{H} is the number of maximal i𝑖i-cliques in H𝐻H. Replace the vertex v𝑣v of G𝐺G with a copy of H𝐻H. In the resulting t𝑑t-colored complete graph of order n+mβˆ’1π‘›π‘š1n+m-1, the number of maximal i𝑖i-cliques meeting V​(H)𝑉𝐻V(H) is ciHsuperscriptsubscript𝑐𝑖𝐻c_{i}^{H}, while the number of maximal i𝑖i-cliques disjoint from V​(H)𝑉𝐻V(H) is ciβˆ’1subscript𝑐𝑖1c_{i}-1. ∎

Theorem 17.

If there is a projective plane of order qπ‘žq, then f2​(q+1,n+q2βˆ’1)≀f2​(q+1,n)+q2βˆ’1subscript𝑓2π‘ž1𝑛superscriptπ‘ž21subscript𝑓2π‘ž1𝑛superscriptπ‘ž21f_{2}(q+1,n+q^{2}-1)\leq f_{2}(q+1,n)+q^{2}-1 for all nβˆˆβ„•π‘›β„•n\in\mathbb{N}.

Proof.

Let P𝑃P be the point set of a projective plane of order qπ‘žq. Choose a line l0subscript𝑙0l_{0} and let x0,x1,…,xqsubscriptπ‘₯0subscriptπ‘₯1…subscriptπ‘₯π‘žx_{0},x_{1},\dots,x_{q} be the points on l0subscript𝑙0l_{0}. Let G𝐺G be the complete graph of order q2superscriptπ‘ž2q^{2} with vertex set Pβˆ–{x0,x1,…,xq}𝑃subscriptπ‘₯0subscriptπ‘₯1…subscriptπ‘₯π‘žP\setminus\{x_{0},x_{1},\dots,x_{q}\}. Color the edges of G𝐺G with colors 0,1,…,q01β€¦π‘ž0,1,\dots,q by assigning color i𝑖i to an edge u​v𝑒𝑣uv if the points u,v,xi𝑒𝑣subscriptπ‘₯𝑖u,v,x_{i} are collinear. For each color i𝑖i there are qπ‘žq maximal i𝑖i-cliques corresponding to the lines other than l0subscript𝑙0l_{0} through the point xisubscriptπ‘₯𝑖x_{i}, and each vertex of G𝐺G is in just one of them. We obtain the desired inequality by setting t=q+1π‘‘π‘ž1t=q+1, m=q2π‘šsuperscriptπ‘ž2m=q^{2}, c=q2+q𝑐superscriptπ‘ž2π‘žc=q^{2}+q in Lemma 16. ∎

The following unpublished observation by Graham and Van Lint is included by permission.

Theorem 18.

(Ronald L. Graham and Jack van Lint) If there is a projective plane of order qπ‘žq, then f2​(q,n+q2)≀f2​(q,n)+q2subscript𝑓2π‘žπ‘›superscriptπ‘ž2subscript𝑓2π‘žπ‘›superscriptπ‘ž2f_{2}(q,n+q^{2})\leq f_{2}(q,n)+q^{2} for all nβˆˆβ„•π‘›β„•n\in\mathbb{N}.

Proof.

Let P𝑃P be the point set of a projective plane of order qπ‘žq. Choose a line l0subscript𝑙0l_{0}; let x0,x1,…,xqsubscriptπ‘₯0subscriptπ‘₯1…subscriptπ‘₯π‘žx_{0},x_{1},\dots,x_{q} be the points on l0subscript𝑙0l_{0}, and let l1,…,lqsubscript𝑙1…subscriptπ‘™π‘žl_{1},\dots,l_{q} be the other lines through x0subscriptπ‘₯0x_{0}. Let G𝐺G be the complete graph of order q2+1superscriptπ‘ž21q^{2}+1 with vertex set Pβˆ–{x1,…,xq}𝑃subscriptπ‘₯1…subscriptπ‘₯π‘žP\setminus\{x_{1},\dots,x_{q}\}. For any edge u​v𝑒𝑣uv of G𝐺G, let l​(u​v)𝑙𝑒𝑣l(uv) be the line through u𝑒u and v𝑣v. Color the edges of G𝐺G with colors 1,…,q1β€¦π‘ž1,\dots,q by assigning color i𝑖i to an edge u​v𝑒𝑣uv if either l​(u​v)=li𝑙𝑒𝑣subscript𝑙𝑖l(uv)=l_{i} or else l​(u​v)𝑙𝑒𝑣l(uv) meets l0subscript𝑙0l_{0} at the point xisubscriptπ‘₯𝑖x_{i}. For each color i𝑖i there are q+1π‘ž1q+1 maximal i𝑖i-cliques; qπ‘žq of them correspond to the lines other than l0subscript𝑙0l_{0} through xisubscriptπ‘₯𝑖x_{i}; the remaining one, corresponding to the line lisubscript𝑙𝑖l_{i}, is the only maximal i𝑖i-clique containing x0subscriptπ‘₯0x_{0}. We obtain the desired inequality by setting t=qπ‘‘π‘žt=q, m=q2+1π‘šsuperscriptπ‘ž21m=q^{2}+1, c=q2+q𝑐superscriptπ‘ž2π‘žc=q^{2}+q, and v=x0𝑣subscriptπ‘₯0v=x_{0} in Lemma 16. ∎

Theorem 19.

If there is a projective plane of order qπ‘žq, then:

(a)f2​(q+1,n)=q2+qβˆ’1​ whenever ​(qβˆ’1)2<n≀(qβˆ’1)​q​;(b)f2​(q+1,n)=q2+q​ whenever ​(qβˆ’1)​q<n≀q2​.π‘Žsubscript𝑓2π‘ž1𝑛superscriptπ‘ž2π‘ž1Β wheneverΒ superscriptπ‘ž12π‘›π‘ž1π‘ž;𝑏subscript𝑓2π‘ž1𝑛superscriptπ‘ž2π‘žΒ wheneverΒ π‘ž1π‘žπ‘›superscriptπ‘ž2.\begin{array}[]{ll}(a)&f_{2}(q+1,n)=q^{2}+q-1$ whenever $(q-1)^{2}<n\leq(q-1)q$;$\\ (b)&f_{2}(q+1,n)=q^{2}+q$ whenever $(q-1)q<n\leq q^{2}$.$\end{array}
Proof.

If n>(qβˆ’1)2𝑛superscriptπ‘ž12n>(q-1)^{2} then by Theorem 13 we have

f2​(q+1,n)β‰₯(qβˆ’1)β€‹βŒˆ(qβˆ’1)2+1βŒ‰+⌈4​(qβˆ’1)2+4βŒ‰subscript𝑓2π‘ž1π‘›π‘ž1superscriptπ‘ž1214superscriptπ‘ž124\displaystyle f_{2}(q+1,n)\geq(q-1)\left\lceil\sqrt{(q-1)^{2}+1}\right\rceil+\left\lceil\sqrt{4(q-1)^{2}+4}\right\rceil
=(qβˆ’1)​q+(2​qβˆ’1)=q2+qβˆ’1;absentπ‘ž1π‘ž2π‘ž1superscriptπ‘ž2π‘ž1\displaystyle=(q-1)q+(2q-1)=q^{2}+q-1;

moreover, if n>(qβˆ’1)​qπ‘›π‘ž1π‘žn>(q-1)q, then

f2​(q+1,n)β‰₯(qβˆ’1)β€‹βŒˆ(qβˆ’1)​q+1βŒ‰+⌈4​(qβˆ’1)​q+4βŒ‰subscript𝑓2π‘ž1π‘›π‘ž1π‘ž1π‘ž14π‘ž1π‘ž4\displaystyle f_{2}(q+1,n)\geq(q-1)\left\lceil\sqrt{(q-1)q+1}\right\rceil+\left\lceil\sqrt{4(q-1)q+4}\right\rceil (3)
=(qβˆ’1)​q+2​q=q2+q.absentπ‘ž1π‘ž2π‘žsuperscriptπ‘ž2π‘ž\displaystyle=(q-1)q+2q=q^{2}+q. (4)

On the other hand, if there is a projective plane of order qπ‘žq, and if n≀q2𝑛superscriptπ‘ž2n\leq q^{2}, then

f2​(q+1,n)≀f2​(q+1,q2)≀f2​(q+1,1)+q2βˆ’1=q2+qsubscript𝑓2π‘ž1𝑛subscript𝑓2π‘ž1superscriptπ‘ž2subscript𝑓2π‘ž11superscriptπ‘ž21superscriptπ‘ž2π‘žf_{2}(q+1,n)\leq f_{2}(q+1,q^{2})\leq f_{2}(q+1,1)+q^{2}-1=q^{2}+q

by Theorem 17. This completes the proof of (b)𝑏(b).

Take a projective plane of order qπ‘žq. Choose a line l0subscript𝑙0l_{0} with points
x0,x1,…,xqsubscriptπ‘₯0subscriptπ‘₯1…subscriptπ‘₯π‘žx_{0},x_{1},\dots,x_{q} and another line l1subscript𝑙1l_{1} through x0subscriptπ‘₯0x_{0}. Let G𝐺G be the complete graph of order q2βˆ’qsuperscriptπ‘ž2π‘žq^{2}-q whose vertices are the points not on l0subscript𝑙0l_{0} or l1subscript𝑙1l_{1}. Assign the color i𝑖i to an edge u​v𝑒𝑣uv of G𝐺G if the line through u𝑒u and v𝑣v meets l0subscript𝑙0l_{0} at xisubscriptπ‘₯𝑖x_{i}. Now maximal i𝑖i-cliques correspond to lines through xisubscriptπ‘₯𝑖x_{i} other than l0subscript𝑙0l_{0} and l1subscript𝑙1l_{1}, so the number of maximal i𝑖i-cliques is qβˆ’1π‘ž1q-1 if i=0𝑖0i=0 and qπ‘žq if i∈{1,…,q}𝑖1β€¦π‘ži\in\{1,\dots,q\}, for a total of q2+qβˆ’1superscriptπ‘ž2π‘ž1q^{2}+q-1. Hence, for n≀q2βˆ’q𝑛superscriptπ‘ž2π‘žn\leq q^{2}-q, we have

f2​(q+1,n)≀f2​(q+1,q2βˆ’q)≀q2+qβˆ’1.subscript𝑓2π‘ž1𝑛subscript𝑓2π‘ž1superscriptπ‘ž2π‘žsuperscriptπ‘ž2π‘ž1f_{2}(q+1,n)\leq f_{2}(q+1,q^{2}-q)\leq q^{2}+q-1.

This completes the proof of (a)π‘Ž(a). ∎

Theorem 20.

For nβ‰₯2𝑛2n\geq 2 the following statements are equivalent:

(a)f2​(n+1,n2)=n2+n​;(b)f2​(n+1,n2)≀n2+n​;(c)there is a projective plane of order ​n​.π‘Žsubscript𝑓2𝑛1superscript𝑛2superscript𝑛2𝑛;𝑏subscript𝑓2𝑛1superscript𝑛2superscript𝑛2𝑛;𝑐there is a projective plane of order 𝑛.\begin{array}[]{ll}(a)&f_{2}(n+1,n^{2})=n^{2}+n$;$\\ (b)&f_{2}(n+1,n^{2})\leq n^{2}+n$;$\\ (c)&$there is a projective plane of order $n$.$\end{array}
Proof.

We have (c)⟹(a)π‘π‘Ž(c)\implies(a) by Theorem 19(b)𝑏(b), and (a)⟹(b)π‘Žπ‘(a)\implies(b) is trivial; we have to show (b)⟹(c)𝑏𝑐(b)\implies(c).

Assume that nβ‰₯2𝑛2n\geq 2 and f2​(n+1,n2)≀n2+nsubscript𝑓2𝑛1superscript𝑛2superscript𝑛2𝑛f_{2}(n+1,n^{2})\leq n^{2}+n. Consider a factorization Kn2=G0βˆͺG1βˆͺβ‹―βˆͺGnsubscript𝐾superscript𝑛2subscript𝐺0subscript𝐺1β‹―subscript𝐺𝑛K_{n^{2}}=G_{0}\cup G_{1}\cup\dots\cup G_{n} with c0+c1+β‹―+cn≀n2+nsubscript𝑐0subscript𝑐1β‹―subscript𝑐𝑛superscript𝑛2𝑛c_{0}+c_{1}+\dots+c_{n}\leq n^{2}+n, where ci=c​(Gi)subscript𝑐𝑖𝑐subscript𝐺𝑖c_{i}=c(G_{i}). Since ci​cjβ‰₯n2subscript𝑐𝑖subscript𝑐𝑗superscript𝑛2c_{i}c_{j}\geq n^{2} for iβ‰ j𝑖𝑗i\neq j, the geometric mean of c0,c1,…,cnsubscript𝑐0subscript𝑐1…subscript𝑐𝑛c_{0},c_{1},\dots,c_{n} is at least n𝑛n, that is,

n≀(c0​c1​…​cn)1n+1≀c0+c1+β‹―+cnn+1≀n,𝑛superscriptsubscript𝑐0subscript𝑐1…subscript𝑐𝑛1𝑛1subscript𝑐0subscript𝑐1β‹―subscript𝑐𝑛𝑛1𝑛n\leq(c_{0}c_{1}\dots c_{n})^{\frac{1}{n+1}}\leq\frac{c_{0}+c_{1}+\dots+c_{n}}{n+1}\leq n,

whence c0=c1=β‹―=cn=nsubscript𝑐0subscript𝑐1β‹―subscript𝑐𝑛𝑛c_{0}=c_{1}=\dots=c_{n}=n.

For each i𝑖i, there are exactly n𝑛n maximal i𝑖i-cliques, which are pairwise disjoint and contain n𝑛n vertices each. If X𝑋X is a maximal i𝑖i-clique and Yπ‘ŒY a maximal j𝑗j-clique, iβ‰ j𝑖𝑗i\neq j, then |X∩Y|=1π‘‹π‘Œ1|X\cap Y|=1. Hence there is a projective plane of order n𝑛n; the points are the vertices of Kn2subscript𝐾superscript𝑛2K_{n^{2}} and the numbers 0,1,…,n01…𝑛0,1,\dots,n; the lines are the set {0,1,…,n}01…𝑛\{0,1,\dots,n\} and the sets Xβˆͺ{i}𝑋𝑖X\cup\{i\} where X𝑋X is a maximal i𝑖i-clique. ∎

In view of Theorem 20, a computation by Lam, Thiel, and Swiercz [3] shows that f2​(11,100)>110subscript𝑓211100110f_{2}(11,100)>110. On the other hand,

f2​(11,100)≀f2​(11,122)≀f2​(11,1)+121=132subscript𝑓211100subscript𝑓211122subscript𝑓2111121132f_{2}(11,100)\leq f_{2}(11,122)\leq f_{2}(11,1)+121=132

by Theorem 18.

Theorem 21.

(Due to Ronald L. Graham and Jack van Lint in the case of a projective plane of order t𝑑t.) If there is a projective plane of order t𝑑t or tβˆ’1𝑑1t-1, then there is a constant Ctsubscript𝐢𝑑C_{t} such that f2​(t,n)≀n+Ctsubscript𝑓2𝑑𝑛𝑛subscript𝐢𝑑f_{2}(t,n)\leq n+C_{t} for all nβˆˆβ„•π‘›β„•n\in\mathbb{N}.

Proof.

If there is a projective plane of order tβˆ’1𝑑1t-1, the result follows from Theorem 17 with

Ct=max⁑{f2​(t,n)βˆ’n:1≀n≀t2βˆ’2​t};subscript𝐢𝑑:subscript𝑓2𝑑𝑛𝑛1𝑛superscript𝑑22𝑑C_{t}=\max\{f_{2}(t,n)-n:1\leq n\leq t^{2}-2t\};

if there is a projective plane of order t𝑑t, it follows from Theorem 18 with

Ct=max⁑{f2​(t,n)βˆ’n:1≀n≀t2}.subscript𝐢𝑑:subscript𝑓2𝑑𝑛𝑛1𝑛superscript𝑑2C_{t}=\max\{f_{2}(t,n)-n:1\leq n\leq t^{2}\}.

∎

Question 22.

For each integer tβ‰₯2𝑑2t\geq 2, is there a constant Ctsubscript𝐢𝑑C_{t} such that f2​(t,n)≀n+Ctsubscript𝑓2𝑑𝑛𝑛subscript𝐢𝑑f_{2}(t,n)\leq n+C_{t} for all nβˆˆβ„•π‘›β„•n\in\mathbb{N}?

If Ctsubscript𝐢𝑑C_{t} exists then Ctβ‰₯(t2)subscript𝐢𝑑binomial𝑑2C_{t}\geq\binom{t}{2} since, when n=tβˆ’1𝑛𝑑1n=t-1, we have f2​(t,n)=f2​(n+1,n)=n​(n+3)2=n+(n+12)=n+(t2)subscript𝑓2𝑑𝑛subscript𝑓2𝑛1𝑛𝑛𝑛32𝑛binomial𝑛12𝑛binomial𝑑2f_{2}(t,n)=f_{2}(n+1,n)=\frac{n(n+3)}{2}=n+\binom{n+1}{2}=n+\binom{t}{2} by Corollary 15(a)π‘Ž(a). In fact the inequality f2​(t,n)≀n+(t2)subscript𝑓2𝑑𝑛𝑛binomial𝑑2f_{2}(t,n)\leq n+\binom{t}{2} holds for 2≀t≀52𝑑52\leq t\leq 5 (Theorems 1, 23, 25, 26). On the other hand it fails for t=6𝑑6t=6 and t=8𝑑8t=8, since f2​(6,7)=23subscript𝑓26723f_{2}(6,7)=23 and f2​(8,9)=38subscript𝑓28938f_{2}(8,9)=38; we omit the details.

Of course a negative answer to Question 22 would imply a negative answer to Question 4 as well.

Theorem 23.
f2​(3,n)≀{n+2Β ifΒ n≑1(mod3),n+3Β otherwise,f_{2}(3,n)\leq\left\{\begin{array}[]{ll}n+2&$ if $n\equiv 1\pmod{3}$,$\\ n+3&$ otherwise,$\end{array}\right.

with equality at least for n≀10𝑛10n\leq 10.

Proof.

By Theorem 17 with q=2π‘ž2q=2, to verify the inequality it suffices to observe that it holds for n≀3𝑛3n\leq 3. To verify equality for all n≀10𝑛10n\leq 10, it suffices to show that f2​(3,8)β‰₯11subscript𝑓23811f_{2}(3,8)\geq 11 and f2​(3,9)β‰₯12subscript𝑓23912f_{2}(3,9)\geq 12; we omit the details. ∎

Question 24.

Does equality hold in Theorem 23 for all n𝑛n?

Theorem 25.
f2​(4,n)≀{n+3ifΒ n≑1(mod8),n+4ifΒ n≑0(mod8),n+5ifΒ n≑2,6,7(mod8),n+6ifΒ n≑3,4,5(mod8),f_{2}(4,n)\leq\left\{\begin{array}[]{ll}n+3&$if $n\equiv 1\pmod{8}$,$\\ n+4&$if $n\equiv 0\pmod{8}$,$\\ n+5&$if $n\equiv 2,6,7\pmod{8}$,$\\ n+6&$if $n\equiv 3,4,5\pmod{8}$,$\\ \end{array}\right.

with equality at least for n≀10𝑛10n\leq 10.

Proof.

Use q=3π‘ž3q=3 in Theorem 17. ∎

Theorem 26.

f2​(5,n)≀n+10subscript𝑓25𝑛𝑛10f_{2}(5,n)\leq n+10 for all n𝑛n.

Proof.

By Theorem 17 with q=4π‘ž4q=4 we have f2​(5,n+15)≀f2​(5,n)+15subscript𝑓25𝑛15subscript𝑓25𝑛15f_{2}(5,n+15)\leq f_{2}(5,n)+15; hence it will suffice to prove the inequality f2​(5,n)≀n+10subscript𝑓25𝑛𝑛10f_{2}(5,n)\leq n+10 for n≀15𝑛15n\leq 15. By Theorem 14 we have f2​(5,n)=5​nβˆ’(n2)subscript𝑓25𝑛5𝑛binomial𝑛2f_{2}(5,n)=5n-\binom{n}{2} for n≀6𝑛6n\leq 6, and by Theorem 19 with q=4π‘ž4q=4 we have f2​(5,n)=19subscript𝑓25𝑛19f_{2}(5,n)=19 for 10≀n≀1210𝑛1210\leq n\leq 12 and f2​(5,n)=20subscript𝑓25𝑛20f_{2}(5,n)=20 for 13≀n≀1613𝑛1613\leq n\leq 16. We leave it as an exercise for the reader to verify that f2​(5,7)≀17subscript𝑓25717f_{2}(5,7)\leq 17 and f2​(5,8)≀18subscript𝑓25818f_{2}(5,8)\leq 18. (In fact it is easy to show that f2​(5,7)=17subscript𝑓25717f_{2}(5,7)=17 and f2​(5,8)=f2​(5,9)=18subscript𝑓258subscript𝑓25918f_{2}(5,8)=f_{2}(5,9)=18.) ∎

Theorem 27.

f2​(5,n)≀n+7subscript𝑓25𝑛𝑛7f_{2}(5,n)\leq n+7 for nβ‰₯37𝑛37n\geq 37. In fact, if nβ‰₯37𝑛37n\geq 37, then

(a)f2(5,n)≀n+4Β ifΒ n≑1(mod5),(b)f2(5,n)≀n+7Β ifΒ n≑2(mod5),(c)f2(5,n)≀n+7Β ifΒ n≑3(mod5),(d)f2(5,n)≀n+6Β ifΒ n≑4(mod5),(e)f2(5,n)≀n+5Β ifΒ n≑0(mod5).\begin{array}[]{ll}(a)&f_{2}(5,n)\leq n+4$ if $n\equiv 1\pmod{5}$,$\\ (b)&f_{2}(5,n)\leq n+7$ if $n\equiv 2\pmod{5}$,$\\ (c)&f_{2}(5,n)\leq n+7$ if $n\equiv 3\pmod{5}$,$\\ (d)&f_{2}(5,n)\leq n+6$ if $n\equiv 4\pmod{5}$,$\\ (e)&f_{2}(5,n)\leq n+5$ if $n\equiv 0\pmod{5}$.$\\ \end{array}
Proof.

Only (a)π‘Ž(a) and (b)𝑏(b) need proof, as (c)βˆ’(e)𝑐𝑒(c)-(e) follow directly from (a)π‘Ž(a). We will use the facts that f2​(5,1)=5subscript𝑓2515f_{2}(5,1)=5 and f2​(5,2)=9subscript𝑓2529f_{2}(5,2)=9 by Theorem 2 (or Theorem 14), and f2​(5,12)=19subscript𝑓251219f_{2}(5,12)=19 by Theorem 19(a)π‘Ž(a) with q=4π‘ž4q=4.

We have f2​(5,n+15)≀f2​(5,n)+15subscript𝑓25𝑛15subscript𝑓25𝑛15f_{2}(5,n+15)\leq f_{2}(5,n)+15 by Theorem 17 with q=4π‘ž4q=4, and f2​(5,n+25)≀f2​(5,n)+25subscript𝑓25𝑛25subscript𝑓25𝑛25f_{2}(5,n+25)\leq f_{2}(5,n)+25 by Theorem 18 with q=5π‘ž5q=5. It follows that f2​(5,n+5​k)≀f2​(5,n)+5​ksubscript𝑓25𝑛5π‘˜subscript𝑓25𝑛5π‘˜f_{2}(5,n+5k)\leq f_{2}(5,n)+5k for every integer kβ‰₯8π‘˜8k\geq 8.

If n≑1(mod5)𝑛annotated1pmod5n\equiv 1\pmod{5} and n>37𝑛37n>37, then n=1+5​k𝑛15π‘˜n=1+5k for some integer kβ‰₯8π‘˜8k\geq 8, and

f2​(5,n)=f2​(5,1+5​k)≀f2​(5,1)+5​k=5+5​k=n+4.subscript𝑓25𝑛subscript𝑓2515π‘˜subscript𝑓2515π‘˜55π‘˜π‘›4f_{2}(5,n)=f_{2}(5,1+5k)\leq f_{2}(5,1)+5k=5+5k=n+4.

If n≑2(mod5)𝑛annotated2pmod5n\equiv 2\pmod{5} and n>37𝑛37n>37, then n=2+5​k𝑛25π‘˜n=2+5k for some integer kβ‰₯8π‘˜8k\geq 8, and

f2​(5,n)=f2​(5,2+5​k)≀f2​(5,2)+5​k=9+5​k=n+7.subscript𝑓25𝑛subscript𝑓2525π‘˜subscript𝑓2525π‘˜95π‘˜π‘›7f_{2}(5,n)=f_{2}(5,2+5k)\leq f_{2}(5,2)+5k=9+5k=n+7.

Finally, if n=37𝑛37n=37 then n≑2(mod5)𝑛annotated2pmod5n\equiv 2\pmod{5} and

f2​(5,n)=f2​(5,12+25)≀f2​(5,12)+25=19+25=n+7.subscript𝑓25𝑛subscript𝑓251225subscript𝑓2512251925𝑛7f_{2}(5,n)=f_{2}(5,12+25)\leq f_{2}(5,12)+25=19+25=n+7.

∎

5 When c​(G)+c¯​(G)=n​(G)+2𝑐𝐺¯𝑐𝐺𝑛𝐺2c(G)+\overline{c}(G)=n(G)+2

For a graph G𝐺G we define τ​(G)=c​(G)+c¯​(G)βˆ’n​(G)πœπΊπ‘πΊΒ―π‘πΊπ‘›πΊ\tau(G)=c(G)+\overline{c}(G)-n(G). In this section we characterize the graphs G𝐺G with τ​(G)=2𝜏𝐺2\tau(G)=2. In doing so we also prove the characterization of graphs with τ​(G)=1𝜏𝐺1\tau(G)=1 which was proved more simply in [2]. We need the following improved version of the result of [2] that τ​(G)β‰₯1𝜏𝐺1\tau(G)\geq 1 for every graph G𝐺G.

Theorem 28.

If G𝐺G is a graph of order n𝑛n, then

c​(G)+c¯​(G)β‰₯n+d​(G)+d¯​(G)βˆ’1β‰₯n+1𝑐𝐺¯𝑐𝐺𝑛𝑑𝐺¯𝑑𝐺1𝑛1c(G)+\overline{c}(G)\geq n+d(G)+\overline{d}(G)-1\geq n+1.

Proof.

We use induction on n𝑛n. Let G𝐺G be a graph of order n𝑛n; let c=c​(G)𝑐𝑐𝐺c=c(G), cΒ―=c¯​(G)¯𝑐¯𝑐𝐺\overline{c}=\overline{c}(G), d=d​(G)𝑑𝑑𝐺d=d(G), dΒ―=d¯​(G)¯𝑑¯𝑑𝐺\overline{d}=\overline{d}(G). We may assume that there is a vertex v𝑣v such that N​(v)β‰ βˆ…β‰ N¯​(v)𝑁𝑣¯𝑁𝑣N(v)\neq\varnothing\neq\overline{N}(v); otherwise G𝐺G or G¯¯𝐺\overline{G} is a complete graph and the result is clear. Let G1=G​[N​(v)]subscript𝐺1𝐺delimited-[]𝑁𝑣G_{1}=G[N(v)] and G2=G​[N¯​(v)]subscript𝐺2𝐺delimited-[]¯𝑁𝑣G_{2}=G[\overline{N}(v)]; Let ni=n​(Gi)subscript𝑛𝑖𝑛subscript𝐺𝑖n_{i}=n(G_{i}), ci=c​(Gi)subscript𝑐𝑖𝑐subscript𝐺𝑖c_{i}=c(G_{i}), cΒ―i=c¯​(Gi)subscript¯𝑐𝑖¯𝑐subscript𝐺𝑖\overline{c}_{i}=\overline{c}(G_{i}), di=d​(Gi)subscript𝑑𝑖𝑑subscript𝐺𝑖d_{i}=d(G_{i}), dΒ―i=d¯​(Gi)subscript¯𝑑𝑖¯𝑑subscript𝐺𝑖\overline{d}_{i}=\overline{d}(G_{i}). By the inductive hypothesis, ci+cΒ―iβ‰₯ni+di+dΒ―iβˆ’1subscript𝑐𝑖subscript¯𝑐𝑖subscript𝑛𝑖subscript𝑑𝑖subscript¯𝑑𝑖1c_{i}+\overline{c}_{i}\geq n_{i}+d_{i}+\overline{d}_{i}-1.

Choose a vertex w𝑀w of G2subscript𝐺2G_{2} which is in exactly d2subscript𝑑2d_{2} maximal cliques of G2subscript𝐺2G_{2}. The number of maximal cliques of G𝐺G containing v𝑣v is equal to c1subscript𝑐1c_{1}; the number of maximal cliques containing w𝑀w is at least d𝑑d; the number of maximal cliques containing neither v𝑣v nor w𝑀w is at least c2βˆ’d2subscript𝑐2subscript𝑑2c_{2}-d_{2}. Since no clique contains both v𝑣v and w𝑀w, cβ‰₯c1+d+c2βˆ’d2.𝑐subscript𝑐1𝑑subscript𝑐2subscript𝑑2c\geq c_{1}+d+c_{2}-d_{2}. Similarly, cΒ―β‰₯cΒ―2+dΒ―+cΒ―1βˆ’dΒ―1¯𝑐subscript¯𝑐2¯𝑑subscript¯𝑐1subscript¯𝑑1\overline{c}\geq\overline{c}_{2}+\overline{d}+\overline{c}_{1}-\overline{d}_{1}.

Adding these two inequalities we get

c+cΒ―β‰₯(c1+cΒ―1)+(c2+cΒ―2)+d+dΒ―βˆ’d2βˆ’dΒ―1β‰₯(n1+d1+dΒ―1βˆ’1)+(n2+d2+dΒ―2βˆ’1)+d+dΒ―βˆ’d2βˆ’dΒ―1=n1+n2+(d1βˆ’1)+(dΒ―2βˆ’1)+d+dΒ―β‰₯n1+n2+d+dΒ―=n+d+dΒ―βˆ’1.𝑐¯𝑐subscript𝑐1subscript¯𝑐1subscript𝑐2subscript¯𝑐2𝑑¯𝑑subscript𝑑2subscript¯𝑑1missing-subexpressionabsentsubscript𝑛1subscript𝑑1subscript¯𝑑11subscript𝑛2subscript𝑑2subscript¯𝑑21𝑑¯𝑑subscript𝑑2subscript¯𝑑1missing-subexpressionabsentsubscript𝑛1subscript𝑛2subscript𝑑11subscript¯𝑑21𝑑¯𝑑missing-subexpressionabsentsubscript𝑛1subscript𝑛2𝑑¯𝑑𝑛𝑑¯𝑑1missing-subexpression\begin{array}[]{ll}c+\overline{c}\geq(c_{1}+\overline{c}_{1})+(c_{2}+\overline{c}_{2})+d+\overline{d}-d_{2}-\overline{d}_{1}\\ \geq(n_{1}+d_{1}+\overline{d}_{1}-1)+(n_{2}+d_{2}+\overline{d}_{2}-1)+d+\overline{d}-d_{2}-\overline{d}_{1}\\ =n_{1}+n_{2}+(d_{1}-1)+(\overline{d}_{2}-1)+d+\overline{d}\\ \geq n_{1}+n_{2}+d+\overline{d}=n+d+\overline{d}-1.\end{array}

∎

We define four classes of graphs, unimaginatively named after their smallest members. As motivation note that K1subscript𝐾1K_{1} is the smallest graph G𝐺G with τ​(G)=1𝜏𝐺1\tau(G)=1, while P4subscript𝑃4P_{4}, C4subscript𝐢4C_{4}, and CΒ―4subscript¯𝐢4\overline{C}_{4} are the smallest graphs with τ​(G)=2𝜏𝐺2\tau(G)=2. Recall that G𝐺G is a split graph if V​(G)𝑉𝐺V(G) is the union of a clique and an anticlique. We begin by defining two classes of split graphs.

A graph G𝐺G is K1subscript𝐾1K_{1}-like if V​(G)=XβˆͺYπ‘‰πΊπ‘‹π‘ŒV(G)=X\cup Y where X𝑋X is a clique, Yπ‘ŒY is an anticlique, and X∩Yβ‰ βˆ…π‘‹π‘ŒX\cap Y\neq\varnothing; equivalently, if ω​(G)+α​(G)=n​(G)+1πœ”πΊπ›ΌπΊπ‘›πΊ1\omega(G)+\alpha(G)=n(G)+1.

A graph G𝐺G is P4subscript𝑃4P_{4}-like if V​(G)=XβˆͺYπ‘‰πΊπ‘‹π‘ŒV(G)=X\cup Y where X𝑋X is a maximal clique, Yπ‘ŒY is a maximal anticlique, and X∩Y=βˆ…π‘‹π‘ŒX\cap Y=\varnothing.

Lemma 29.

If G𝐺G is a split graph then just one of the following statements holds:

(a)ω​(G)+α​(G)=n​(G)+1​, ​G​ is ​K1​-like, and ​τ​(G)=1;(b)ω​(G)+α​(G)=n​(G)​, ​G​ is ​P4​-like, and ​τ​(G)=2.π‘Žπœ”πΊπ›ΌπΊπ‘›πΊ1, 𝐺 isΒ subscript𝐾1-like, and 𝜏𝐺1π‘πœ”πΊπ›ΌπΊπ‘›πΊ, 𝐺 isΒ subscript𝑃4-like, and 𝜏𝐺2\begin{array}[]{ll}(a)&\omega(G)+\alpha(G)=n(G)+1$, $G$ is $K_{1}$-like, and $\tau(G)=1;\\ (b)&\omega(G)+\alpha(G)=n(G)$, $G$ is $P_{4}$-like, and $\tau(G)=2.\end{array}

A graph G𝐺G is C4subscript𝐢4C_{4}-like if there is a 444-element set U={p,q,r,s}βŠ†V​(G)π‘ˆπ‘π‘žπ‘Ÿπ‘ π‘‰πΊU=\{p,q,r,s\}\subseteq V(G) such that p​q,q​r,r​s,p​s∈E​(G)π‘π‘žπ‘žπ‘Ÿπ‘Ÿπ‘ π‘π‘ πΈπΊpq,qr,rs,ps\in E(G) while p​r,q​sβˆ‰E​(G)π‘π‘Ÿπ‘žπ‘ πΈπΊpr,qs\notin E(G), and V​(G)βˆ–U=XβˆͺYπ‘‰πΊπ‘ˆπ‘‹π‘ŒV(G)\setminus U=X\cup Y where X𝑋X is a clique and Yπ‘ŒY is an anticlique, each vertex in X𝑋X is joined to p𝑝p and qπ‘žq and at least one more vertex in Uπ‘ˆU, and no vertex in Yπ‘ŒY is joined to any vertex in Uπ‘ˆU.

A graph G𝐺G is C¯4subscript¯𝐢4\overline{C}_{4}-like if G¯¯𝐺\overline{G} is C4subscript𝐢4C_{4}-like.

Lemma 30.

If a graph G𝐺G is P4subscript𝑃4P_{4}-like or C4subscript𝐢4C_{4}-like or CΒ―4subscript¯𝐢4\overline{C}_{4}-like, then τ​(G)=2𝜏𝐺2\tau(G)=2 and ω​(G)+α​(G)=n​(G)πœ”πΊπ›ΌπΊπ‘›πΊ\omega(G)+\alpha(G)=n(G).

Lemma 31.

Let G𝐺G be a graph. If τ​(G)≀2𝜏𝐺2\tau(G)\leq 2 and ω​(G)+α​(G)β‰₯n​(G)πœ”πΊπ›ΌπΊπ‘›πΊ\omega(G)+\alpha(G)\geq n(G), then G𝐺G is C4subscript𝐢4C_{4}-like or CΒ―4subscript¯𝐢4\overline{C}_{4}-like or a split graph.

Proof.

Let G=(V,E)𝐺𝑉𝐸G=(V,E) and n=|V|𝑛𝑉n=|V|. Let X𝑋X be a clique and Yπ‘ŒY an anticlique with |X|+|Y|=nπ‘‹π‘Œπ‘›|X|+|Y|=n. We may assume that XβˆͺYβ‰ Vπ‘‹π‘Œπ‘‰X\cup Y\neq V; let Vβˆ–(XβˆͺY)={s}π‘‰π‘‹π‘Œπ‘ V\setminus(X\cup Y)=\{s\}, X∩Y={p}π‘‹π‘Œπ‘X\cap Y=\{p\}, X0=Xβˆ–{p}subscript𝑋0𝑋𝑝X_{0}=X\setminus\{p\}, Y0=Yβˆ–{p}subscriptπ‘Œ0π‘Œπ‘Y_{0}=Y\setminus\{p\}. Since the statement of the lemma is invariant under complementation, we may assume that p𝑝p is joined to s𝑠s. We may assume that some vertex q∈X0π‘žsubscript𝑋0q\in X_{0} is not joined to s𝑠s, and some vertex r∈Y0π‘Ÿsubscriptπ‘Œ0r\in Y_{0} is joined to s𝑠s, as otherwise G𝐺G is a split graph. Then D​(p)β‰₯2𝐷𝑝2D(p)\geq 2, since {p,q}π‘π‘ž\{p,q\} and {p,s}𝑝𝑠\{p,s\} are contained in different maximal cliques.

Then D¯​(q)β‰₯2Β―π·π‘ž2\overline{D}(q)\geq 2. From this and D​(p)β‰₯2𝐷𝑝2D(p)\geq 2 we have

n+2β‰₯c​(G)+c¯​(G)β‰₯βˆ‘y∈YD​(y)+βˆ‘x∈XD¯​(x)β‰₯(|Y|+1)+(|X|+1)=n+2.𝑛2𝑐𝐺¯𝑐𝐺subscriptπ‘¦π‘Œπ·π‘¦subscriptπ‘₯𝑋¯𝐷π‘₯π‘Œ1𝑋1𝑛2n+2\geq c(G)+\overline{c}(G)\geq\sum_{y\in Y}D(y)+\sum_{x\in X}\overline{D}(x)\geq(|Y|+1)+(|X|+1)=n+2. (5)

Since equality holds in (5), we have D​(y)=1𝐷𝑦1D(y)=1 for each y∈Y0𝑦subscriptπ‘Œ0y\in Y_{0}, D¯​(x)=1¯𝐷π‘₯1\overline{D}(x)=1 for each x∈Xβˆ–{q}π‘₯π‘‹π‘žx\in X\setminus\{q\}, and every maximal anticlique meets X𝑋X.

If some vertex x∈X0βˆ–{q}π‘₯subscript𝑋0π‘žx\in X_{0}\setminus\{q\} is not joined to s𝑠s, then, according as xπ‘₯x is joined to rπ‘Ÿr or not, either D​(r)β‰₯2π·π‘Ÿ2D(r)\geq 2 or D¯​(x)β‰₯2¯𝐷π‘₯2\overline{D}(x)\geq 2, neither of which is possible. Hence (Xβˆ–{q})βˆͺ{s}π‘‹π‘žπ‘ (X\setminus\{q\})\cup\{s\} is a clique.

Now suppose qπ‘žq is joined to some vertex y∈Y0𝑦subscriptπ‘Œ0y\in Y_{0}. If y𝑦y is joined to s𝑠s, then D​(y)β‰₯2𝐷𝑦2D(y)\geq 2, which is impossible; otherwise {y,s}𝑦𝑠\{y,s\} is contained in a maximal anticlique disjoint from X𝑋X, also impossible. Hence qπ‘žq is joined to no vertex in Y0subscriptπ‘Œ0Y_{0}, so Y0βˆͺ{q}subscriptπ‘Œ0π‘žY_{0}\cup\{q\} is an anticlique and G𝐺G is a split graph.

Then D​(r)β‰₯2π·π‘Ÿ2D(r)\geq 2. From this and D​(p)β‰₯2𝐷𝑝2D(p)\geq 2 we have

n+2β‰₯c​(G)+c¯​(G)β‰₯βˆ‘y∈YD​(y)+βˆ‘x∈XD¯​(x)β‰₯(|Y|+2)+|X|=n+2.𝑛2𝑐𝐺¯𝑐𝐺subscriptπ‘¦π‘Œπ·π‘¦subscriptπ‘₯𝑋¯𝐷π‘₯π‘Œ2𝑋𝑛2n+2\geq c(G)+\overline{c}(G)\geq\sum_{y\in Y}D(y)+\sum_{x\in X}\overline{D}(x)\geq(|Y|+2)+|X|=n+2. (6)

Since equality holds in (6), we have D​(y)=1𝐷𝑦1D(y)=1 for each y∈Y0βˆ–{r}𝑦subscriptπ‘Œ0π‘Ÿy\in Y_{0}\setminus\{r\}, D¯​(x)=1¯𝐷π‘₯1\overline{D}(x)=1 for each x∈Xπ‘₯𝑋x\in X, and every maximal anticlique meets X𝑋X.

Each vertex x∈Xπ‘₯𝑋x\in X is joined to rπ‘Ÿr or to s𝑠s, as otherwise we would have D¯​(x)β‰₯2¯𝐷π‘₯2\overline{D}(x)\geq 2.

If s𝑠s were joined to some vertex y∈Y0βˆ–{r}𝑦subscriptπ‘Œ0π‘Ÿy\in Y_{0}\setminus\{r\}, then we would have D​(y)β‰₯2𝐷𝑦2D(y)\geq 2 or D¯​(q)β‰₯2Β―π·π‘ž2\overline{D}(q)\geq 2 according as y𝑦y is joined to qπ‘žq or not. Hence s𝑠s is not joined to any vertex in Y0βˆ–{r}subscriptπ‘Œ0π‘ŸY_{0}\setminus\{r\}.

Now the anticlique (Y0βˆ–{r})βˆͺ{s}subscriptπ‘Œ0π‘Ÿπ‘ (Y_{0}\setminus\{r\})\cup\{s\} is contained in a maximal anticlique, which must meet X𝑋X but cannot contain p𝑝p. Hence there is a vertex qβ€²βˆˆX0superscriptπ‘žβ€²subscript𝑋0q^{\prime}\in X_{0} which is not joined to s𝑠s or to any vertex in Y0βˆ–{r}subscriptπ‘Œ0π‘ŸY_{0}\setminus\{r\}. Of course qβ€²superscriptπ‘žβ€²q^{\prime} is joined to rπ‘Ÿr, since D¯​(qβ€²)=1¯𝐷superscriptπ‘žβ€²1\overline{D}(q^{\prime})=1.

It is easy to check that G𝐺G is C4subscript𝐢4C_{4}-like, with p,qβ€²,r,s,X0βˆ–{qβ€²},Y0βˆ–{r}𝑝superscriptπ‘žβ€²π‘Ÿπ‘ subscript𝑋0superscriptπ‘žβ€²subscriptπ‘Œ0π‘Ÿp,q^{\prime},r,s,X_{0}\setminus\{q^{\prime}\},Y_{0}\setminus\{r\} playing the roles of p,q,r,s,X,Yπ‘π‘žπ‘Ÿπ‘ π‘‹π‘Œp,q,r,s,X,Y in the definition. ∎

Lemma 32.

If G𝐺G is a graph and H=G​[N¯​(v)]𝐻𝐺delimited-[]¯𝑁𝑣H=G[\overline{N}(v)], where v𝑣v is a vertex such that N​(v)𝑁𝑣N(v) is a clique and N¯​(v)β‰ βˆ…Β―π‘π‘£\overline{N}(v)\neq\varnothing, then τ​(H)≀τ​(G)𝜏𝐻𝜏𝐺\tau(H)\leq\tau(G).

Proof.

Let X=N​(v)𝑋𝑁𝑣X=N(v) and Y=N¯​(v)=V​(H)π‘ŒΒ―π‘π‘£π‘‰π»Y=\overline{N}(v)=V(H). Then

τ​(G)=c​(G)+c¯​(G)βˆ’n​(G)β‰₯(1+c​(H))+(c¯​(H)+|X|)βˆ’(|X|+|Y|+1)=c​(H)+c¯​(H)βˆ’|Y|=τ​(H).𝜏𝐺absent𝑐𝐺¯𝑐𝐺𝑛𝐺missing-subexpressionabsent1π‘π»Β―π‘π»π‘‹π‘‹π‘Œ1missing-subexpressionabsentπ‘π»Β―π‘π»π‘Œπœπ»\begin{array}[]{ll}\tau(G)&=c(G)+\overline{c}(G)-n(G)\\ &\geq(1+c(H))+(\overline{c}(H)+|X|)-(|X|+|Y|+1)\\ &=c(H)+\overline{c}(H)-|Y|=\tau(H).\end{array}

∎

Lemma 33.

If τ​(G)≀2𝜏𝐺2\tau(G)\leq 2 then ω​(G)+α​(G)β‰₯n​(G)πœ”πΊπ›ΌπΊπ‘›πΊ\omega(G)+\alpha(G)\geq n(G).

Proof.

We use induction on n​(G)𝑛𝐺n(G). Let G𝐺G be a graph, n​(G)=n𝑛𝐺𝑛n(G)=n, τ​(G)≀2𝜏𝐺2\tau(G)\leq 2, so c​(G)+c¯​(G)≀n+2𝑐𝐺¯𝑐𝐺𝑛2c(G)+\overline{c}(G)\leq n+2. By Theorem 28, d​(G)+d¯​(G)≀3𝑑𝐺¯𝑑𝐺3d(G)+\overline{d}(G)\leq 3, so d​(G)=1𝑑𝐺1d(G)=1 or d¯​(G)=1¯𝑑𝐺1\overline{d}(G)=1. We may assume that d​(G)=1𝑑𝐺1d(G)=1, i.e., some vertex v𝑣v is in a unique maximal clique; i.e., N​(v)𝑁𝑣N(v) is a clique. We may assume that N​(v)β‰ βˆ…β‰ N¯​(v)𝑁𝑣¯𝑁𝑣N(v)\neq\varnothing\neq\overline{N}(v). Let Z=N​(v)𝑍𝑁𝑣Z=N(v), W=N¯​(v)π‘ŠΒ―π‘π‘£W=\overline{N}(v); then n=|W|+|Z|+1π‘›π‘Šπ‘1n=|W|+|Z|+1. Let H=G​[W]𝐻𝐺delimited-[]π‘ŠH=G[W]; Then τ​(H)≀2𝜏𝐻2\tau(H)\leq 2 by Lemma 32. By the inductive hypothesis, ω​(H)+α​(H)β‰₯|W|πœ”π»π›Όπ»π‘Š\omega(H)+\alpha(H)\geq|W|. By Lemmas 31 and 29, H𝐻H is K1subscript𝐾1K_{1}-like or P4subscript𝑃4P_{4}-like or C4subscript𝐢4C_{4}-like or CΒ―4subscript¯𝐢4\overline{C}_{4}-like. Since A=Zβˆͺ{v}𝐴𝑍𝑣A=Z\cup\{v\} is a clique which meets every maximal anticlique, and since D¯​(v)=c¯​(H)¯𝐷𝑣¯𝑐𝐻\overline{D}(v)=\overline{c}(H), we have

c¯​(G)=βˆ‘a∈AD¯​(a)=c¯​(H)+|Z|+βˆ‘z∈Z(D¯​(z)βˆ’1).¯𝑐𝐺subscriptπ‘Žπ΄Β―π·π‘ŽΒ―π‘π»π‘subscript𝑧𝑍¯𝐷𝑧1\overline{c}(G)=\sum_{a\in A}\overline{D}(a)=\overline{c}(H)+|Z|+\sum_{z\in Z}(\overline{D}(z)-1). (7)

So W=XβˆͺYπ‘Šπ‘‹π‘ŒW=X\cup Y where X𝑋X is a clique, Yπ‘ŒY is an anticlique, and X∩Y={p}π‘‹π‘Œπ‘X\cap Y=\{p\}. Then n=|X|+|Y|+|Z|π‘›π‘‹π‘Œπ‘n=|X|+|Y|+|Z| and c¯​(H)=|X|¯𝑐𝐻𝑋\overline{c}(H)=|X|, so

c¯​(G)=|X|+|Z|+βˆ‘z∈Z(D¯​(z)βˆ’1)=nβˆ’|Y|+βˆ‘z∈Z(D¯​(z)βˆ’1).¯𝑐𝐺𝑋𝑍subscript𝑧𝑍¯𝐷𝑧1π‘›π‘Œsubscript𝑧𝑍¯𝐷𝑧1\overline{c}(G)=|X|+|Z|+\sum_{z\in Z}(\overline{D}(z)-1)=n-|Y|+\sum_{z\in Z}(\overline{D}(z)-1). (8)

Let B=Yβˆͺ{v}π΅π‘Œπ‘£B=Y\cup\{v\} and let mπ‘šm be the number of maximal cliques disjoint from B𝐡B. Since B𝐡B is an anticlique,

c​(G)=m+βˆ‘b∈BD​(b)β‰₯m+1+|Y|+(D​(p)βˆ’1).π‘πΊπ‘šsubscriptπ‘π΅π·π‘π‘š1π‘Œπ·π‘1c(G)=m+\sum_{b\in B}D(b)\geq m+1+|Y|+(D(p)-1). (9)

Adding 8 and 9 we get

c​(G)+c¯​(G)β‰₯n+1+m+(D​(p)βˆ’1)+βˆ‘z∈Z(D¯​(z)βˆ’1).𝑐𝐺¯𝑐𝐺𝑛1π‘šπ·π‘1subscript𝑧𝑍¯𝐷𝑧1c(G)+\overline{c}(G)\geq n+1+m+(D(p)-1)+\sum_{z\in Z}(\overline{D}(z)-1).

Since c​(G)+c¯​(G)≀n+2𝑐𝐺¯𝑐𝐺𝑛2c(G)+\overline{c}(G)\leq n+2, it follows that

m+(D​(p)βˆ’1)+βˆ‘z∈Z(D¯​(z)βˆ’1)≀1.π‘šπ·π‘1subscript𝑧𝑍¯𝐷𝑧11m+(D(p)-1)+\sum_{z\in Z}(\overline{D}(z)-1)\leq 1. (10)

Since α​(G)=α​(H)+1=|Y|+1𝛼𝐺𝛼𝐻1π‘Œ1\alpha(G)=\alpha(H)+1=|Y|+1, we have to show that ω​(G)β‰₯|X|+|Z|βˆ’1πœ”πΊπ‘‹π‘1\omega(G)\geq|X|+|Z|-1, i.e., that ω​(G)β‰₯|Z|+|X0|πœ”πΊπ‘subscript𝑋0\omega(G)\geq|Z|+|X_{0}| where X0=Xβˆ–{p}subscript𝑋0𝑋𝑝X_{0}=X\setminus\{p\}. If ZβˆͺX0𝑍subscript𝑋0Z\cup X_{0} is a clique we’re done, so we may assume that some vertex z0∈Zsubscript𝑧0𝑍z_{0}\in Z is not joined to some vertex x0∈X0subscriptπ‘₯0subscript𝑋0x_{0}\in X_{0}. Then either D​(p)β‰₯2𝐷𝑝2D(p)\geq 2 (if z0subscript𝑧0z_{0} is joined to p𝑝p) or D¯​(z0)β‰₯2¯𝐷subscript𝑧02\overline{D}(z_{0})\geq 2 (otherwise), so m=0π‘š0m=0, i.e., every maximal clique meets B𝐡B. Hence it will suffice to show that Zβˆͺ(X0βˆ–{x0})𝑍subscript𝑋0subscriptπ‘₯0Z\cup(X_{0}\setminus\{x_{0}\}) is a clique, since, being disjoint from B𝐡B, it’s not a maximal clique.

So D​(p)=2𝐷𝑝2D(p)=2, and D¯​(z)=1¯𝐷𝑧1\overline{D}(z)=1 for all z∈Z𝑧𝑍z\in Z. Then z0subscript𝑧0z_{0} is joined to every vertex in X0βˆ–{x0}subscript𝑋0subscriptπ‘₯0X_{0}\setminus\{x_{0}\}, since z0subscript𝑧0z_{0} is not joined to x0subscriptπ‘₯0x_{0}, and D¯​(z0)=1¯𝐷subscript𝑧01\overline{D}(z_{0})=1. Likewise, if z∈Z𝑧𝑍z\in Z and z𝑧z is not joined to p𝑝p, then z𝑧z is joined to every vertex in X0subscript𝑋0X_{0}, since D¯​(z)=1¯𝐷𝑧1\overline{D}(z)=1. Finally, suppose there are vertices z1∈Zβˆ–{z0}subscript𝑧1𝑍subscript𝑧0z_{1}\in Z\setminus\{z_{0}\} and x1∈X0βˆ–{x0}subscriptπ‘₯1subscript𝑋0subscriptπ‘₯0x_{1}\in X_{0}\setminus\{x_{0}\} such that z1subscript𝑧1z_{1} is joined to p𝑝p and not to x1subscriptπ‘₯1x_{1}. Then the cliques {p,z0,z1}𝑝subscript𝑧0subscript𝑧1\{p,z_{0},z_{1}\}, {p,z0,x1}𝑝subscript𝑧0subscriptπ‘₯1\{p,z_{0},x_{1}\}, and X𝑋X extend to three different maximal cliques, contradicting D​(p)=2𝐷𝑝2D(p)=2.

So D¯​(z0)=2¯𝐷subscript𝑧02\overline{D}(z_{0})=2, D¯​(z)=1¯𝐷𝑧1\overline{D}(z)=1 for all z∈Zβˆ–{z0}𝑧𝑍subscript𝑧0z\in Z\setminus\{z_{0}\}, and D​(p)=1𝐷𝑝1D(p)=1. Suppose z∈Zβˆ–{z0}𝑧𝑍subscript𝑧0z\in Z\setminus\{z_{0}\}; if z𝑧z is joined to p𝑝p then z𝑧z is joined to every vertex in X0subscript𝑋0X_{0} because D​(p)=1𝐷𝑝1D(p)=1; if z𝑧z is not joined to p𝑝p, then z𝑧z is joined to every vertex in X0subscript𝑋0X_{0} because D¯​(z)=1¯𝐷𝑧1\overline{D}(z)=1. Finally, z0subscript𝑧0z_{0} is joined to every vertex in X0βˆ–{x0}subscript𝑋0subscriptπ‘₯0X_{0}\setminus\{x_{0}\} because otherwise we would have D¯​(z0)β‰₯3¯𝐷subscript𝑧03\overline{D}(z_{0})\geq 3.

In Cases 2 – 4 we have ω​(H)+α​(H)=n​(H)=|W|πœ”π»π›Όπ»π‘›π»π‘Š\omega(H)+\alpha(H)=n(H)=|W| by Lemma 30, and we have to show that ω​(G)+α​(G)β‰₯nπœ”πΊπ›ΌπΊπ‘›\omega(G)+\alpha(G)\geq n. Since n=|Z|+|W|+1=|Z|+ω​(H)+α​(H)+1=|Z|+ω​(H)+α​(G)π‘›π‘π‘Š1π‘πœ”π»π›Όπ»1π‘πœ”π»π›ΌπΊn=|Z|+|W|+1=|Z|+\omega(H)+\alpha(H)+1=|Z|+\omega(H)+\alpha(G), we have to show that ω​(G)β‰₯|Z|+ω​(H)πœ”πΊπ‘πœ”π»\omega(G)\geq|Z|+\omega(H).

So W=XβˆͺYπ‘Šπ‘‹π‘ŒW=X\cup Y where X𝑋X is a maximal clique in H𝐻H, Yπ‘ŒY is a maximal anticlique in H𝐻H, X∩Y=βˆ…π‘‹π‘ŒX\cap Y=\varnothing, and n=|X|+|Y|+|Z|+1π‘›π‘‹π‘Œπ‘1n=|X|+|Y|+|Z|+1. Then c¯​(H)=|X|+1¯𝑐𝐻𝑋1\overline{c}(H)=|X|+1, so (7) becomes

c¯​(G)=|X|+1+|Z|+βˆ‘z∈Z(D¯​(z)βˆ’1).¯𝑐𝐺𝑋1𝑍subscript𝑧𝑍¯𝐷𝑧1\overline{c}(G)=|X|+1+|Z|+\sum_{z\in Z}(\overline{D}(z)-1). (11)

The set B=Yβˆͺ{v}π΅π‘Œπ‘£B=Y\cup\{v\} is an anticlique. Let mπ‘šm be the number of maximal cliques disjoint from B𝐡B; mβ‰₯1π‘š1m\geq 1, since X𝑋X is contained in a maximal clique which is disjoint from B𝐡B. Then we have

c​(G)=m+βˆ‘b∈BD​(b)β‰₯2+|Y|+βˆ‘y∈Y(D​(y)βˆ’1).π‘πΊπ‘šsubscript𝑏𝐡𝐷𝑏2π‘Œsubscriptπ‘¦π‘Œπ·π‘¦1c(G)=m+\sum_{b\in B}D(b)\geq 2+|Y|+\sum_{y\in Y}(D(y)-1). (12)

Adding (11) and (12) and recalling that n=|X|+|Y|+|Z|+1π‘›π‘‹π‘Œπ‘1n=|X|+|Y|+|Z|+1, we get

c​(G)+c¯​(G)β‰₯n+2+βˆ‘y∈Y(D​(y)βˆ’1)+βˆ‘z∈Z(D¯​(z)βˆ’1).𝑐𝐺¯𝑐𝐺𝑛2subscriptπ‘¦π‘Œπ·π‘¦1subscript𝑧𝑍¯𝐷𝑧1c(G)+\overline{c}(G)\geq n+2+\sum_{y\in Y}(D(y)-1)+\sum_{z\in Z}(\overline{D}(z)-1).

Since c​(G)+c¯​(G)≀n+2𝑐𝐺¯𝑐𝐺𝑛2c(G)+\overline{c}(G)\leq n+2, it follows that D​(y)=1𝐷𝑦1D(y)=1 for all y∈Yπ‘¦π‘Œy\in Y, and D¯​(z)=1¯𝐷𝑧1\overline{D}(z)=1 for all z∈Z𝑧𝑍z\in Z. We have to show that ω​(G)β‰₯|Z|+ω​(H)πœ”πΊπ‘πœ”π»\omega(G)\geq|Z|+\omega(H), i.e., ω​(G)β‰₯|Z|+|X|πœ”πΊπ‘π‘‹\omega(G)\geq|Z|+|X|. It will suffice to show that ZβˆͺX𝑍𝑋Z\cup X is a clique. Assume for a contradiction that some vertex z0∈Zsubscript𝑧0𝑍z_{0}\in Z is not joined to some vertex x0∈Xsubscriptπ‘₯0𝑋x_{0}\in X. Since Yπ‘ŒY is a maximal anticlique in H𝐻H, x0subscriptπ‘₯0x_{0} is joined to some vertex y0∈Ysubscript𝑦0π‘Œy_{0}\in Y. But then either D​(y0)β‰₯2𝐷subscript𝑦02D(y_{0})\geq 2 (if y0subscript𝑦0y_{0} is joined to z0subscript𝑧0z_{0}) or else D¯​(z0)β‰₯2¯𝐷subscript𝑧02\overline{D}(z_{0})\geq 2 (otherwise); either way we have a contradiction.

So there is a 444-element set U={p,q,r,s}βŠ†Wπ‘ˆπ‘π‘žπ‘Ÿπ‘ π‘ŠU=\{p,q,r,s\}\subseteq W such that p​q,q​r,r​s,p​s∈E​(G)π‘π‘žπ‘žπ‘Ÿπ‘Ÿπ‘ π‘π‘ πΈπΊpq,qr,rs,ps\in E(G) while p​r,q​sβˆ‰E​(G)π‘π‘Ÿπ‘žπ‘ πΈπΊpr,qs\notin E(G); and Wβˆ–U=XβˆͺYπ‘Šπ‘ˆπ‘‹π‘ŒW\setminus U=X\cup Y where X𝑋X is a clique and Yπ‘ŒY is an anticlique, each vertex in X𝑋X is joined to p𝑝p and qπ‘žq and at least one more vertex in Uπ‘ˆU, and no vertex in Yπ‘ŒY is joined to any vertex in Uπ‘ˆU. Then n=|X|+|Y|+|Z|+5π‘›π‘‹π‘Œπ‘5n=|X|+|Y|+|Z|+5, c¯​(H)=ω​(H)=|X|+2Β―π‘π»πœ”π»π‘‹2\overline{c}(H)=\omega(H)=|X|+2, and (7) becomes

c¯​(G)=|X|+2+|Z|+βˆ‘z∈Z(D¯​(z)βˆ’1).¯𝑐𝐺𝑋2𝑍subscript𝑧𝑍¯𝐷𝑧1\overline{c}(G)=|X|+2+|Z|+\sum_{z\in Z}(\overline{D}(z)-1). (13)

Since B=Yβˆͺ{p,r,v}π΅π‘Œπ‘π‘Ÿπ‘£B=Y\cup\{p,r,v\} is an anticlique and D​(r)β‰₯2π·π‘Ÿ2D(r)\geq 2,

c​(G)β‰₯βˆ‘b∈BD​(b)β‰₯|Y|+D​(p)+3.𝑐𝐺subscriptπ‘π΅π·π‘π‘Œπ·π‘3c(G)\geq\sum_{b\in B}D(b)\geq|Y|+D(p)+3. (14)

Adding (13) and (14), with n=|X|+|Y|+|Z|+5π‘›π‘‹π‘Œπ‘5n=|X|+|Y|+|Z|+5, we get

c​(G)+c¯​(G)β‰₯n+D​(p)+βˆ‘z∈Z(D¯​(z)βˆ’1).𝑐𝐺¯𝑐𝐺𝑛𝐷𝑝subscript𝑧𝑍¯𝐷𝑧1c(G)+\overline{c}(G)\geq n+D(p)+\sum_{z\in Z}(\overline{D}(z)-1).

Since c​(G)+c¯​(G)≀n+2𝑐𝐺¯𝑐𝐺𝑛2c(G)+\overline{c}(G)\leq n+2, we must have D​(p)=2𝐷𝑝2D(p)=2 and D¯​(z)=1¯𝐷𝑧1\overline{D}(z)=1 for all z∈Z𝑧𝑍z\in Z. A similar argument, using the anticlique Bβ€²=Yβˆͺ{q,s,v}superscriptπ΅β€²π‘Œπ‘žπ‘ π‘£B^{\prime}=Y\cup\{q,s,v\} instead of B𝐡B, shows that D​(q)=2π·π‘ž2D(q)=2. We have to show that ω​(G)β‰₯|Z|+ω​(H)πœ”πΊπ‘πœ”π»\omega(G)\geq|Z|+\omega(H), i.e., ω​(G)β‰₯|Z|+|X|+2πœ”πΊπ‘π‘‹2\omega(G)\geq|Z|+|X|+2.

Proof.

Suppose z∈Z𝑧𝑍z\in Z, x∈Xπ‘₯𝑋x\in X, z𝑧z not joined to xπ‘₯x. Then z𝑧z must be joined to p𝑝p and qπ‘žq, since D¯​(z)=1¯𝐷𝑧1\overline{D}(z)=1. But then the cliques {p,q,z}π‘π‘žπ‘§\{p,q,z\}, {p,q,x}π‘π‘žπ‘₯\{p,q,x\}, {p,s}𝑝𝑠\{p,s\} extend to three different maximal cliques, contradicting D​(p)=2𝐷𝑝2D(p)=2. ∎

Proof.

If neither ZβˆͺXβˆͺ{q}π‘π‘‹π‘žZ\cup X\cup\{q\} nor ZβˆͺXβˆͺ{s}𝑍𝑋𝑠Z\cup X\cup\{s\} is a clique, then there are vertices t1,t2∈ZβˆͺXsubscript𝑑1subscript𝑑2𝑍𝑋t_{1},t_{2}\in Z\cup X (not necessarily distinct) such that t1subscript𝑑1t_{1} is not joined to qπ‘žq and t2subscript𝑑2t_{2} is not joined to s𝑠s. Then t1subscript𝑑1t_{1} and t2subscript𝑑2t_{2} are joined to p𝑝p, because every vertex in X𝑋X is joined to p𝑝p, and if ti∈Zsubscript𝑑𝑖𝑍t_{i}\in Z then tisubscript𝑑𝑖t_{i} must be joined to p𝑝p because D¯​(ti)=1¯𝐷subscript𝑑𝑖1\overline{D}(t_{i})=1. But then {p,q}π‘π‘ž\{p,q\}, {p,s}𝑝𝑠\{p,s\}, and {p,t1,t2}𝑝subscript𝑑1subscript𝑑2\{p,t_{1},t_{2}\} are cliques extending to three different maximal cliques, contradicting D​(p)=2𝐷𝑝2D(p)=2. ∎

Proof.

Similar to Claim 2, using D​(q)=2π·π‘ž2D(q)=2. ∎

It follows from Claims 2 and 3 that there are two adjacent vertices u,w∈Uπ‘’π‘€π‘ˆu,w\in U such that ZβˆͺXβˆͺ{u,w}𝑍𝑋𝑒𝑀Z\cup X\cup\{u,w\} is a clique, so ω​(G)β‰₯|Z|+|X|+2πœ”πΊπ‘π‘‹2\omega(G)\geq|Z|+|X|+2.

So there is a 444-element set U={p,q,r,s}βŠ†Wπ‘ˆπ‘π‘žπ‘Ÿπ‘ π‘ŠU=\{p,q,r,s\}\subseteq W such that p​r,q​s∈E​(G)π‘π‘Ÿπ‘žπ‘ πΈπΊpr,qs\in E(G) while p​q,q​r,r​s,p​sβˆ‰E​(G)π‘π‘žπ‘žπ‘Ÿπ‘Ÿπ‘ π‘π‘ πΈπΊpq,qr,rs,ps\notin E(G); and Wβˆ–U=XβˆͺYπ‘Šπ‘ˆπ‘‹π‘ŒW\setminus U=X\cup Y where X𝑋X is an anticlique and Yπ‘ŒY is a clique, no vertex of X𝑋X is joined to p𝑝p or qπ‘žq or to more than one vertex in Uπ‘ˆU, and every vertex in Yπ‘ŒY is joined to every vertex in Uπ‘ˆU. Then n=|X|+|Y|+|Z|+5π‘›π‘‹π‘Œπ‘5n=|X|+|Y|+|Z|+5, and c¯​(H)=|Y|+4Β―π‘π»π‘Œ4\overline{c}(H)=|Y|+4 so (7) becomes

c¯​(G)=|Y|+4+|Z|+βˆ‘z∈Z(D¯​(z)βˆ’1).Β―π‘πΊπ‘Œ4𝑍subscript𝑧𝑍¯𝐷𝑧1\overline{c}(G)=|Y|+4+|Z|+\sum_{z\in Z}(\overline{D}(z)-1). (15)

We have to show that ω​(G)β‰₯|Z|+ω​(H)πœ”πΊπ‘πœ”π»\omega(G)\geq|Z|+\omega(H), i.e., that ω​(G)β‰₯|Z|+|Y|+2πœ”πΊπ‘π‘Œ2\omega(G)\geq|Z|+|Y|+2. The set B=Xβˆͺ{p,q,v}π΅π‘‹π‘π‘žπ‘£B=X\cup\{p,q,v\} is an anticlique. Let mπ‘šm be the number of maximal cliques disjoint from B𝐡B. Then

c​(G)=m+βˆ‘b∈BD​(b)β‰₯m+|X|+D​(p)+2.π‘πΊπ‘šsubscriptπ‘π΅π·π‘π‘šπ‘‹π·π‘2c(G)=m+\sum_{b\in B}D(b)\geq m+|X|+D(p)+2. (16)

Adding (15) and (16), since n=|X|+|Y|+|Z|+5π‘›π‘‹π‘Œπ‘5n=|X|+|Y|+|Z|+5, we get

c​(G)+c¯​(G)β‰₯n+1+D​(p)+m+βˆ‘z∈Z(D¯​(z)βˆ’1).𝑐𝐺¯𝑐𝐺𝑛1π·π‘π‘šsubscript𝑧𝑍¯𝐷𝑧1c(G)+\overline{c}(G)\geq n+1+D(p)+m+\sum_{z\in Z}(\overline{D}(z)-1).

But c​(G)+c¯​(G)≀n+2𝑐𝐺¯𝑐𝐺𝑛2c(G)+\overline{c}(G)\leq n+2, so D​(p)=1𝐷𝑝1D(p)=1, and D¯​(z)=1¯𝐷𝑧1\overline{D}(z)=1 for all z∈Z𝑧𝑍z\in Z, and m=0π‘š0m=0, i.e., every maximal clique meets B𝐡B.

Now ZβˆͺYβˆͺ{r}π‘π‘Œπ‘ŸZ\cup Y\cup\{r\} is a clique, since Z𝑍Z and Yβˆͺ{r}π‘Œπ‘ŸY\cup\{r\} are cliques, and if some vertex z∈Z𝑧𝑍z\in Z were not joined to some vertex w∈Yβˆͺ{r}π‘€π‘Œπ‘Ÿw\in Y\cup\{r\}, then, since p𝑝p is joined to w𝑀w, we would have either D​(p)β‰₯2𝐷𝑝2D(p)\geq 2 (if z𝑧z is joined to p𝑝p) or else D¯​(z)β‰₯2¯𝐷𝑧2\overline{D}(z)\geq 2 (otherwise). So ZβˆͺYβˆͺ{r}π‘π‘Œπ‘ŸZ\cup Y\cup\{r\} is a clique of size |Z|+|Y|+1π‘π‘Œ1|Z|+|Y|+1, but it can’t be a maximal clique since it’s disjoint from B𝐡B, so ω​(G)β‰₯|Z|+|Y|+2πœ”πΊπ‘π‘Œ2\omega(G)\geq|Z|+|Y|+2. ∎

Theorem 34.

Let G𝐺G be a graph of order n𝑛n.

(a)c​(G)+c¯​(G)=n+1​ if and only if ​G​ is ​K1​-like.(b)c​(G)+c¯​(G)=n+2​ if and only if ​G​ is ​P4​-like, ​C4​-like or ​CΒ―4​-like.π‘Žπ‘πΊΒ―π‘πΊπ‘›1Β if and only if 𝐺 isΒ subscript𝐾1-like.𝑏𝑐𝐺¯𝑐𝐺𝑛2Β if and only if 𝐺 isΒ subscript𝑃4-like,Β subscript𝐢4-like orΒ subscript¯𝐢4-like.\begin{array}[]{ll}(a)&c(G)+\overline{c}(G)=n+1$ if and only if $G$ is $K_{1}$-like.$\\ (b)&c(G)+\overline{c}(G)=n+2$ if and only if $G$ is $P_{4}$-like, $C_{4}$-like or $\overline{C}_{4}$-like.$\\ \end{array}
Proof.

By Lemmas 29 and 30, c​(G)+c¯​(G)=n+1𝑐𝐺¯𝑐𝐺𝑛1c(G)+\overline{c}(G)=n+1 if G𝐺G is K1subscript𝐾1K_{1}-like, while c​(G)+c¯​(G)=n+2𝑐𝐺¯𝑐𝐺𝑛2c(G)+\overline{c}(G)=n+2 if G𝐺G is P4subscript𝑃4P_{4}-like or C4subscript𝐢4C_{4}-like or CΒ―4subscript¯𝐢4\overline{C}_{4}-like. On the other hand, if c​(G)+c¯​(G)≀n+2𝑐𝐺¯𝑐𝐺𝑛2c(G)+\overline{c}(G)\leq n+2, then by Lemmas 33 and 31 and 29, G𝐺G is K1subscript𝐾1K_{1}-like or P4subscript𝑃4P_{4}-like or C4subscript𝐢4C_{4}-like or CΒ―4subscript¯𝐢4\overline{C}_{4}-like. ∎

Only part (b) of Theorem 34 is new; part (a) was proved in [2].

Corollary 35.

Let G𝐺G be a graph of order n𝑛n.

(a)c​(G)+c¯​(G)β‰₯n+1​.(b)If ​ω​(G)+α​(G)≀n​ then ​c​(G)+c¯​(G)β‰₯n+2​.(c)If ​ω​(G)+α​(G)<n​ then ​c​(G)+c¯​(G)β‰₯n+3​.π‘Žπ‘πΊΒ―π‘πΊπ‘›1.𝑏IfΒ πœ”πΊπ›ΌπΊπ‘›Β then 𝑐𝐺¯𝑐𝐺𝑛2.𝑐IfΒ πœ”πΊπ›ΌπΊπ‘›Β then 𝑐𝐺¯𝑐𝐺𝑛3.\begin{array}[]{ll}(a)&c(G)+\overline{c}(G)\geq n+1$.$\\ (b)&$If $\omega(G)+\alpha(G)\leq n$ then $c(G)+\overline{c}(G)\geq n+2$.$\\ (c)&$If $\omega(G)+\alpha(G)<n$ then $c(G)+\overline{c}(G)\geq n+3$.$\\ \end{array}
Proof.

This follows from Theorem 28, Theorem 34(a), and Lemma 33. ∎

Only part (c) of Corollary 35 is new; parts (a) and (b) were proved in [2].

Acknowledgement

We would like to thank Ronald L. Graham and Jack van Lint for permission to include their Theorem 18 and the corresponding portion of Theorem 21. Theorem 18 is also used in the proof of Theorem 27.

References

  • [1] P. ErdΕ‘s, F. Galvin, and M. Krieger, Factoring complete hypergraphs into factors with few cliques, Abstract 750-A23, Notices. Amer. Math. Soc. 24 (1977), A-629.
  • [2] F. Galvin and M. Krieger, The minimum number of cliques in a graph and its complement, Proc. Second Louisiana Conf. on Combinatorics, Graph Theory, and Computing (Louisiana State University, Baton Rouge, La., 1971), pp. 345–352.
  • [3] C. W. H. Lam, L. Thiel, and S. Swiercz, The non-existence of finite projective planes of order 10, Canad. J. Math. 41 (1989), 1117–1123.
  • [4] D. B. West, Introduction to Graph Theory, 2nd edition, Prentice–Hall, Inc., Upper Saddle River, NJ, 2001.