This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

\newdate

date31092023

Duality and χ\chi-Boundedness of Ordered Graphs

Michal Čertík, Jaroslav Nešetřil
Computer Science Institute, Faculty of Mathematics and Physics
Charles University
Prague, Czech Republic
(\displaydatedate)
Abstract

We shall show that there exists only one duality pair for ordered graphs. We will also define a corresponding definition of χ\chi-boundedness for ordered graphs and show that all ordered graphs are χ\chi-bounded and prove an analogy of Gyarfás-Sumner conjecture for ordered graphs. We shall conclude with showing that these results hold for directed ordered graphs, but they do not hold for oriented ordered graphs.

1 Introduction

An Ordered Graph is an undirected graph whose vertices are totally ordered. Thus, ordered graph GG is a triple G=(V,E,G)G=(V,E,\leq_{G}) (see figure 1).

Ordered graphs are interesting objects which have been studied in the context of extremal problems ( [9]), Ramsey theory ( [2]) and structural graph theory ( [13]). They also provide an important class of objects in model theory (NIP classes which are not stable, see e.g. [10]).

Ordered structures are also a key object in structural Ramsey theory (see e.g. [6][7]).

In this paper we consider homomorphisms of ordered graphs. These are defined as edge- and order-preserving mappings and they are naturally related to ordered chromatic number (which in turn naturally relates to extremal results, see e.g. [9]).

We characterize homomorphism dualities and we also consider Gyarfás-Sumner type problems (χ\chi-boundedness) and in the context of ordered graphs we fully characterize χ\chi-boundedness.

2 Preliminaries and Statement of Results

For ordered graphs G=(V,E,G)G=(V,E,\leq_{G}) and G=(V,E,G)G^{\prime}=(V^{\prime},E^{\prime},\leq_{G^{\prime}}), an Ordered Homomorphism is a mapping f:VVf:V\to V^{\prime} preserving both edges and orderings. Explicitly, ff satisfies

  1. 1.

    f(u)f(v)Ef(u)f(v)\in E^{\prime} for all uvEuv\in E,

  2. 2.

    f(u)Gf(v)f(u)\leq_{G^{\prime}}f(v) whenever uGvu\leq_{G}v.

The existence of ordered homomorphism will be denoted by GGG\to G^{\prime} (see figure 1).

Ordered homomorphisms compose and thus most of the categorical definitions can be considered without any changes, see e.g. [5].

Particularly we can define a core of a graph GG as the smallest subgraph HH of GG such that GHG\to H. (Equivalently, this is the smallest retract of GG.)

Let us now define an independent interval as a set of independent vertices, explicitly, if G\leq_{G} is given as v1,v2,,vnv_{1},v_{2},\ldots,v_{n}, then an [i,j][i,j] independent interval in GG is the set {vi,vi+1,,vj}\{v_{i},v_{i+1},\ldots,v_{j}\} which does not contain any edge of GG (see figure 1). Further on, we may call independent interval simply, an interval.

Refer to caption
Figure 1: Ordered Homomorphism ff and Independent Intervals.

The (ordered) chromatic number χ(G)\chi(G) is then the minimum kk such that V(G)V(G) can be partitioned to kk disjoint intervals. Notice that for ordered graphs this is the size of the smallest homomorphic image and, alternatively, the minimum kk such that GKkG\to K_{k}. (KkK_{k} is of course the complete graph with a fixed linear ordering.) We shall call χ(G)\chi(G) also a colouring of GG.

As opposed to the chromatic number of (unordered) graphs the ordered chromatic number can be determined by a simple greedy algorithm. This is formulated below as Proposition 2.1 (which may be a folklore).

Greedy Algorithm is a natural one: process the vertices in the given order and colour each vertex by the smallest available colour which fits the rule. What is the rule? For graph GG with ordering G\leq_{G} each colour has to be an interval in G\leq_{G}.

We have the following:

Proposition 2.1.

For every ordered graph GG the greedy algorithm finds χ(G)\chi(G).

Proof.

Put χ(G)=k\chi(G)=k. Obviously, by the algorithm, the greedy algorithm finds a kk^{\prime}-colouring for kkk^{\prime}\geq k.

We prove k=kk^{\prime}=k by an induction on kk. For k=1k=1 the graph is just a single interval.

In the induction step let χ(G)=k+1\chi(G)=k+1 and let I1,I2,,IkI_{1},I_{2},\ldots,I_{k} be an optimal colouring. Consider the interval I1I^{\prime}_{1} given by the greedy algorithm. Clearly I1I1I_{1}\subseteq I^{\prime}_{1} and thus G=GI1G^{\prime}=G-I^{\prime}_{1} (with the vertex set I1I^{\prime}_{1} deleted) satisfies χ(G)=k\chi(G^{\prime})=k and the greedy algorithm produces also a kk-colouring. Thus also k+1k+1 is the result of the greedy algorithm. ∎

Following the standard definition of the complexity class 𝒫\mathcal{P} as in [1], the following result then follows.

Corollary 2.2.

Let GG be an ordered graph. Then determining χ(G)\chi(G) is in 𝒫\mathcal{P}.

Proof.

As the Greedy algorithm goes over all vertices of GG and at each step it checks whether the vertex is not connected to some of the vertices before, the complexity of an algorithm is at most 𝒪(|G|2)\mathcal{O}(|G|^{2}). ∎

For the purposes of further explorations we define a double as a pair of consecutive vertices in an ordered graphs connected by an edge. Then MkM_{k} will be an ordered graph that has kk doubles, kk edges, and 2k2k vertices and we will call MkM_{k} an ordered matching.

Let us now define a (Singleton) Homomorphism Duality as a pair of graphs F,DF,D satisfying

F↛G if and only if GDF\not\to G\text{ if and only if }G\to D

for every graph GG.

For graphs and relational structures the dualities are characterized in [8]. In Theorem 3.1 we provide result in ordered setting. The key property is played by ordered matching. Note that ordered matching plays role also in Ramsey context ( [3]).

In Section 4 we deal with questions which subgraphs are unavoidable in large chromatic number.

Formulating it dually, we ask when is chromatic number of a graph bounded as a function of its subgraphs (for not ordered graphs this amounts to bound chromatic number as a function of the clique number, which leads to χ\chi-bounded classes and Gyarfás-Sumner conjecture).

For ordered graphs is the situation easier and we prove the analogous statement, Theorem 4.2, with the following definition of some of such unavoidable graphs.

Definition 2.1.

Ordered Matching MnM_{n} has points ai,bi,i=1,,na_{i},b_{i},i=1,\ldots,n, with ordering a1<b1<a2<b2<<an<bna_{1}<b_{1}<a_{2}<b_{2}<\ldots<a_{n}<b_{n} and edges {ai,bi},i=1,,n\{a_{i},b_{i}\},i=1,\ldots,n. aia_{i} are left vertices, bib_{i} are right vertices.

  • MnLRM^{LR}_{n} is MnM_{n} together with all edges {ai,bj},i<j\{a_{i},b_{j}\},i<j.

  • MnRLM^{RL}_{n} is MnM_{n} together with all edges {bi,aj},i<j\{b_{i},a_{j}\},i<j.

  • Mn+M^{+}_{n} is just MnLRMnRLM^{LR}_{n}\cup M^{RL}_{n}.

Refer to caption
Figure 2: Mn,MnLR,MnRLM_{n},M^{LR}_{n},M^{RL}_{n} and Mn+M^{+}_{n}

In Section 5 we extend our results to oriented and directed graphs. We show that the general pattern for directed graphs is similar, while for the oriented graphs it does not hold.

To conclude the preliminaries, let us also define PmP_{m} as an ordered graph on mm vertices and E(Pm)={vivi+1|i=1,,m1;viV(Pm)}E(P_{m})=\{v_{i}v_{i+1}|i=1,\ldots,m-1;v_{i}\in V(P_{m})\}. We will call PmP_{m} a directed path. And we define an edge set EE^{\prime} in an ordered graph as non-intersecting if for every two distinct edges e1={v1,v2},v1<v2e_{1}=\{v_{1},v_{2}\},v_{1}<v_{2} and e2={v3,v4},v3<v4e_{2}=\{v_{3},v_{4}\},v_{3}<v_{4} in EE^{\prime}, either v2v3v_{2}\leq v_{3} or v4v1v_{4}\leq v_{1} (we again use the intuition of (in this case open) intervals).

Observation 2.3.

Let GG be an ordered graph with kk non-intersecting edges. Then if there exists an ordered homomorphism from GG to HH, HH must also contain at least kk non-intersecting edges.

Proof.

By definition, the ordered homomorphism from GG to HH must preserve an ordering <G<_{G} as well as edges of GG, therefore the observation follows. ∎

3 Dualities of Ordered Graphs

We shall show now that the following duality is the only duality of ordered graphs.

Theorem 3.1.

Let GG be any ordered core. Then MkM_{k} and KkK_{k} is the only pair of ordered cores satisfying Mk↛GM_{k}\not\to G if and only if GKkG\to K_{k}.

Proof.

We will first show that MkM_{k} and KkK_{k} are a singleton. From Observation 2.3, we see that Mk↛KkM_{k}\not\to K_{k}, as KkK_{k} has at most k1k-1 non-intersecting edges and MkM_{k} would need at least kk. Therefore if MkGM_{k}\to G then G↛KkG\not\to K_{k}.

On the other hand, if Mk↛GM_{k}\not\to G, then we use the greedy algorithm on GG and Proposition 2.1 to get ordered graph AgA_{g} on minimum number of gg vertices. AgA_{g} will of course contain a directed path PgP_{g} (as we took all the partition classes maximal). It is clear that g1<kg-1<k, as otherwise MkAgM_{k}\to A_{g}, and therefore also to GG, as MkAgM_{k}\to A_{g} maps non-intersecting edges of MkM_{k} to non-intersecting edges of AgA_{g}, therefore MkM_{k} could map to these edges in GG. Hence as GAgKkG\to A_{g}\to K_{k} (as AgA_{g} is on kk or less vertices), we get GKkG\to K_{k}.

Now we shall prove that there does not exist any other duality pair. Let FF and HH be ordered cores satisfying F↛GF\not\to G if and only if GHG\to H and FF and HH are not equal to MkM_{k} and KkK_{k}, resp.

Let GG be any ordered matching. We see that if FGF\to G, then FF must contain partitions that map to an ordered matching. But then GG and FF must contain an ordered matching MkM_{k}, for which the ordered homomorphism FF to MkM_{k} is onto. But then MkM_{k} is the core of FF.

Now, if there is no homomorphism from FF to any ordered matching, then there does not exist any ordered graph HH, for which there exists ordered homomorphism from ordered matching to HH, as ordered matchings can require a homomorphic image of any size. E.g. already for FF being a P3P_{3} or K3K_{3}, with GG being an ordered matching, we can see that there is no ordered homomorphism from FF to GG, yet we can get χ(G)\chi(G) of any size by increasing nn in MnM_{n}.

We also see that if we forbid an ordered homomorphism from MkM_{k} to GG, then KkK_{k} is the only graph satisfying Mk↛GM_{k}\not\to G if and only if GKkG\to K_{k}. ∎

4 χ\chi-boundedness of Ordered Graphs

For the ordinary graphs, we recall that χ\chi-bounded family \mathcal{F} of graphs is one for which there is a function ff such that, with ω(G)\omega(G) being a size of maximum clique in GG, for every ω(G),G\omega(G),G\in\mathcal{F}, χ(G)\chi(G) coloring of GG is at most f(ω(G))f(\omega(G)). We see that we could adopt this notion simply by replacing ordinary graphs with ordered graphs and homomorphism with an ordered homomorphism.

We of course see that not all ordered graphs are χ\chi-bounded, as already the matching itself can have arbitrarily large chromatic number and maximum clique number of two. The standard notion of χ\chi-boundedness will therefore not work.

Adopting the notion of χ\chi-bounded graphs from the ordinary graphs nonetheless, we shall try to replace the maximum clique size in the definition of χ\chi-boundedness with the size nn of maximum ordered matching MnM_{n}. We can then define the analogy of χ\chi-boundedness for ordered graphs as follows.

χ<\chi^{<}-bounded family \mathcal{F} of ordered graphs is one for which there is a function ff such that, for every nn, nn being a size of maximum ordered matching subgraph MnM_{n} of GG (maximum in the sense that there is no greater k>nk>n such that MkM_{k} is an ordered matching subgraph of GG), χ(G)\chi(G) is at most f(n)f(n). Without any risk of confusion, we shall denote χ<\chi^{<}-bounded simply as χ\chi-bounded further on.

We will now show that all ordered graphs are χ\chi-bounded.

Theorem 4.1.

Let GG be an ordered graph. Let MkM_{k} be maximum ordered matching subgraph of GG. Then χ(G)2k+1\chi(G)\leq 2k+1.

Proof.

Let us choose a subgraph MkM_{k} of GG, where kk is maximum, and run the Greedy Algorithm on GG. As a result of this procedure we will get an ordered graph AgA_{g}, where |Ag|=χ(G)|A_{g}|=\chi(G) from 2.1. As in 3.1, AgA_{g} will contain a directed path PgP_{g}. We also see from Observation 2.3 that the MkM_{k} edges will map to non-intersecting edges in AgA_{g}, as they were non-intersecting in GG (where in GG they did not share a common vertex).

Now let us assume that χ(G)>2k+1\chi(G)>2k+1. Then AgA_{g} will have at least 2k+22k+2 vertices and it will contain P2k+2P_{2k+2}. But then P2k+2P_{2k+2} contains an ordered matching of size k+1k+1, which should have been chosen in GG initially. ∎

Let us now try to prove a stronger (induced) version of the statement, again borrowing an idea from the ordinary graphs - the famous Gyárfás–Sumner conjecture. The conjecture states that for every tree TT and complete graph KK, the graphs with neither TT nor KK as induced subgraphs can be properly colored using only a constant number of colors.

We shall replace the tree with previously introduced forbidden structures and prove the following statement.

Theorem 4.2.

There exists a function f:f:\mathbb{N}\to\mathbb{N} with the following property:

Let GG be an ordered graph not containing any of the following graphs as induced subgraphs:

Kn,Mn,MnRL,Mn+.K_{n},M_{n},M^{RL}_{n},M^{+}_{n}.

Then χ(G)f(n)\chi(G)\leq f(n).

Advancing the proof, let us first define a set of ordered graphs as non-induced, if for every two ordered graphs within this set, one ordered graph is not an induced graph of the other, and vice versa. We prove that Mn,Km,MkRL,Ml+,n,k2,m,l3M_{n},K_{m},M^{RL}_{k},M^{+}_{l},n,k\geq 2,m,l\geq 3 are non-induced ordered graphs and we determine a coloring of them.

Proposition 4.3.

Let k,l,m,n,n,k2,m,l3k,l,m,n\in\mathbb{N},n,k\geq 2,m,l\geq 3. Then Mn,Km,MkRL,Ml+M_{n},K_{m},M^{RL}_{k},M^{+}_{l} are non-induced ordered graphs and χ(Mn)=n+1,χ(Km)=m,χ(MkRL)=2k,χ(Ml+)=2l\chi(M_{n})=n+1,\chi(K_{m})=m,\chi(M^{RL}_{k})=2k,\chi(M^{+}_{l})=2l.

Proof.

Let us first define a size of an edge {vi,vi+j}\{v_{i},v_{i+j}\} as jj and a distance in between vertices viv_{i} and vi+jv_{i+j} as jj as well.

We shall start with an easy observation that none of the ordered graphs Km,MkRL,Ml+K_{m},M^{RL}_{k},M^{+}_{l} can be an induced subgraph of MnM_{n}. This is due to the fact that there are always more edges in these graphs than in any induced subgraph of MnM_{n} that has more than 33 vertices (as Km,MkRL,Ml+,m2,m,l3K_{m},M^{RL}_{k},M^{+}_{l},m\geq 2,m,l\geq 3 have more than three vertices).

We can also see that none of Mn,MkRL,Ml+M_{n},M^{RL}_{k},M^{+}_{l} can be an induced subgraph of KmK_{m}, as Mn,MkRL,Ml+M_{n},M^{RL}_{k},M^{+}_{l} are at least on four vertices and their first four vertices do not form a complete graph, which is always the case for any induced subgraph of KmK_{m}.

Let us show that MkRLM^{RL}_{k} does not contain Mn,Km,Ml+M_{n},K_{m},M^{+}_{l} as induced subgraphs.

We notice that MkRLM^{RL}_{k} contains 2k2k vertices and directed path P2kP_{2k} on all of these vertices. We will call the set of edges in P2kP_{2k} the path edges and we will call all the other edges (of size longer than one) long edges. We shall notice that all of the path edges are of size one and all the long edges are of odd size of at least three (from definition). Therefore there are no edges of even size.

We will first show that MkRLM^{RL}_{k} does not contain a triangle. As there are no edges of even size in MkRLM^{RL}_{k}, every edge will connect even and odd vertex. As every triangle needs three vertices, it will contain at least two even or two odd vertices. But these vertices would need an even edge to connect them.

Let us now show that MkRLM^{RL}_{k} does not contain induced ordered matching of size two. There are three possible cases of such a matching. Having both path edges, both long edges or having one path edge and one long edge of MkRLM^{RL}_{k}, all pairs of edges of course non-intersecting (without common vertex).

Let us first introduce a notation for a path edge as ei={vi,vi+1}e_{i}=\{v_{i},v_{i+1}\}. In case the induced matching of size two would be containing both path edges, these edges of course could not be consecutive (eie_{i} and ei+1e_{i+1}) and they could not be edges eie_{i} and ei+2e_{i+2} as they would have another path edge ei+1e_{i+1} in between them. Therefore, these edges would need to be eie_{i} and ei+je_{i+j} with j>2j>2. But all of these path edges are connected by a long edge.

If we consider an induced matching of size two consisting of two long edges, we notice that these will of course need to have each left vertex of each of these edges at different even vertices. But we see that every two non-intersecting edges with different even left vertex will be connected by an edge connecting the first vertex of the first edge and the second vertex of the second edge (as their distance is odd and greater than two).

The last case of induced matching of size two consists of one path edge rr and one long edge ss. We notice that there are two case. One where path edge rr is on the left and one where it is on the right side of the long edge ss.

In case path edge rr is on the left, long edges coming from the even vertex of path edge rr (whether it is the first or the second vertex of it) will connect rr with the second vertex of the long edge ss, again as long edges in MkRLM^{RL}_{k} connect even vertex on the left and odd vertex on the right, if the distance in between them is longer than two.

If path edge rr is on the right, we will notice that its odd vertex will connect it with left vertex of every long edge ss on the left side of the path edge rr, as this vertex is even and distance in between them is again longer than two.

Let us show that MkRLM^{RL}_{k} does not contain Ml+M^{+}_{l} as an induced subgraph. We see that by taking vertices b1,a2,b2,a3b_{1},a_{2},b_{2},a_{3} in MkRL,k3M^{RL}_{k},k\geq 3 we get M2+M^{+}_{2}. Therefore Ml+M^{+}_{l} needs to have l3l\geq 3.

We can also observe that in MkRL,k2M^{RL}_{k},k\geq 2 every edge (path or long) connects two vertices, out of which one vertex can have long edges connecting this vertex to the vertices on the right and another vertex does not have any long edges connecting it to the vertices on the right (by definition). Another observation is that when choosing an induced subgraph GG of MkRL,k2M^{RL}_{k},k\geq 2, this subgraph will need to contain a directed path P|G|P_{|G|}, as Ml+,l3M^{+}_{l},l\geq 3 always contains a directed path P|Ml+|P_{|M^{+}_{l}|}. We then observe that Ml+M^{+}_{l} always contains edges (a1,b1),(a1,bn),(an,bn),(b1,an)(a_{1},b_{1}),(a_{1},b_{n}),(a_{n},b_{n}),(b_{1},a_{n}), where (a1,bn),(b1,an)(a_{1},b_{n}),(b_{1},a_{n}) are long edges connecting the first vertex with the last one and second vertex with the one vertex before the last one.

But then whichever induced subgraph GG we take in MkRL,k2M^{RL}_{k},k\geq 2, the first two vertices will need to be connected, as there needs to be a path P|G|P_{|G|} within GG. GG will also need to contain at least 66 vertices (because of the minimum order of Ml+,l3M^{+}_{l},l\geq 3), and therefore first two vertices of GG will need to be connected to the last two vertices by long edges of size |G|1|G|-1 and |G|3|G|-3. But as we have shown, every edge in MkRLM^{RL}_{k} contains a vertex, which is not connected to the vertices on the right by a long edge. Therefore as the first and second vertex in GG are connected by a path edge, one of the first two vertices in GG is not connected to neither of the last two vertices.

Let us show that Ml+,l3M^{+}_{l},l\geq 3 does not contain Mn,Km,MkRL,n,k2,m3M_{n},K_{m},M^{RL}_{k},n,k\geq 2,m\geq 3 as induced subgraphs. The proof of the first two ordered induced subgraphs Mn,KmM_{n},K_{m} goes essentially the same as before.

We notice that Ml+M^{+}_{l} contains 2l2l vertices and directed path P2lP_{2l}. Again, all of the path edges are of size one and all the long edges are of odd size of at least three (by definition). Therefore, there are no edges of even size also in Ml+M^{+}_{l}.

We will first show that Ml+M^{+}_{l} does not contain a triangle. As observed before, as there are no edges of even size in Ml+M^{+}_{l} either, every edge will connect even and odd vertex. As every triangle needs 33 vertices, it will again contain at least two even or two odd vertices and these vertices would need an even edge to connect them.

Let us now show that Ml+M^{+}_{l} does not contain induced matching of size 22. There are again three possible cases of such an induced matching - having both path edges, both long edges or having one path edge and one long edge in Ml+M^{+}_{l}, all the three cases again non-intersecting without common vertex in Ml+M^{+}_{l}.

In case the induced matching of size 22 would be containing both path edges, we notice that these edges could not be consecutive again (eie_{i} and ei+1e_{i+1}) and they could not be edges eie_{i} and ei+2e_{i+2} as they would have another path edge ei+1e_{i+1} in between them. Therefore, these edges would need to be eie_{i} and ei+je_{i+j} with j>2j>2. But all of these path edges are connected by a long edge.

If we consider the induced matching of size 22 consisting of two long edges, then there are 3 cases.

  • Left vertex of each of the long edges is at different odd vertices.

  • Left vertex of each of the long edges is at different even vertices.

  • Left vertex of one of the long edges is at odd vertex and left vertex of another long edge is at even vertex.

For the first case, we see that every two long edges with different odd left vertex will be connected by an edge in between the first vertex of the first edge and second vertex of the second edge (and also by an edge in between second vertex of the first edge and first vertex of the second edge).

For the second case, we see that every two long edges with different even left vertex will be also connected by first vertex of the first edge and second vertex of the second edge (and also by an edge in between second vertex of the first edge and first vertex of the second edge).

In the last case, these edges will be connected by an edge in between the first vertex of the first edge and the first vertex of the second edge (and also by an edge in between the second vertex of the first edge and the second vertex of the second edge).

The last case of induced matching of size 22 consists of one path edge rr and one long edge ss. We notice that there are two case. One where path edge rr is on the left and one where it is on the right side of the long edge ss.

In case path edge rr is on the left, we can see that vertices of rr are connecting this edge to all the vertices on the right from rr (by definition), therefore rr must be connected to ll by an edge in Ml+M^{+}_{l}.

If path edge rr is on the right, we will notice that ll will have one odd and one even vertex. But then also ll is connected to all the vertices on the right from ll (again by definition), so ll must be connected to rr by an edge in Ml+M^{+}_{l}.

For the proof that Ml+,l3M^{+}_{l},l\geq 3 does not contain MkRL,k2M^{RL}_{k},k\geq 2 as induced subgraph, we first notice that first vertex of the MkRLM^{RL}_{k} is not connected to other vertices on the right by a long edge. We also notice that MkRLM^{RL}_{k} has at least four vertices and that MkRLM^{RL}_{k} always contains a directed path P2kP_{2k}.

Therefore any induced ordered subgraph GG of Ml+M^{+}_{l} will also always need to contain a path P4P_{4} on the first 44 vertices. We also notice that for every vertex viv_{i} in Ml+M^{+}_{l}, vertex viv_{i} is connected to all the vertices, with odd distance that is greater than 22 (from definition), on the right from viv_{i}.

This means that as the edges in Ml+M^{+}_{l} are all of the odd size, whichever edges we pick in Ml+M^{+}_{l} to form the directed path P4P_{4} on the first four vertices of MkRLM^{RL}_{k}, the sum of the sizes of these edges in Ml+M^{+}_{l} will be odd and greater than two, as the sum of three odd numbers is always odd and greater than two. But then the first four vertices of GG must have the first and the fourth vertex connected, and as these can be connected only by long edge in GG, we observed that this cannot be the case for MkRLM^{RL}_{k}, as MkRLM^{RL}_{k} does not have the first vertex connected to any vertices on the right by long edge.

The last thing to show is the coloring of these ordered graphs.

But this is clear, as for Mn,n2M_{n},n\geq 2 we map all the pairs of vertices bi,ai+1,1in1b_{i},a_{i+1},1\leq i\leq n-1 to one interval (this is equivalent to running the greedy algorithm on MnM_{n}) and get a directed path Pn+1P_{n+1} as a minimum homomorphic image of Mn,n2M_{n},n\geq 2.

For MkRL,Ml+,k2,l3M^{RL}_{k},M^{+}_{l},k\geq 2,l\geq 3 this is also straightforward, as both of these ordered graphs contain a directed path P2kP_{2k} and P2lP_{2l}, respectively, and coloring of Km,m3K_{m},m\geq 3 is mm.

Proof of Theorem 4.2.

We see from the Theorem 4.1 that forbidding ordered matchings as subgraphs can bound the coloring of the ordered graphs and from Theorem 4.1 and Theorem 3.1, that as the coloring of our input ordered graph GG grows, so must grow our ordered matching subgraph MkM_{k} of GG and there are no other ordered graphs with this property.

Let us now take this ordered matching subgraph MkM_{k} of GG and take ordered matching induced subgraph M~k\tilde{M}_{k} of GG on the vertices of MkM_{k}.

We see that there can be four different edges in between two doubles of M~k\tilde{M}_{k} and we denote them as follows:

  • LRLR edge, if it connects left vertex of the first double with the right vertex of the second double.

  • RLRL edge, if it connects right vertex of the first double with the left vertex of the second double.

  • LLLL edge, if it connects left vertex of the first double with the left vertex of the second double.

  • RRRR edge, if it connects right vertex of the first double with the right vertex of the second double.

We therefore have 24=162^{4}=16 possible ways how two doubles can be connected (with one option being no edges in between them).

Let us denote a set of edges connecting two doubles as e.g. {LR,RL,RR}\{LR,RL,RR\}, if these two doubles are connected by edges LR,RLLR,RL and RRRR, and other sets of edges connecting two doubles analogously.

Let us take an ordered graph where all the doubles are connected by the same set of edges (one of 1616). Then these 1616 different ordered graphs correspond to the following five induced graphs:

  • If nn doubles are not connected by any edge {}\{\}, this results in an induced ordered matching MnM_{n}.

  • If nn doubles are all connected to each other by an edge {LR}\{LR\}, we get an ordered graph MnLRM^{LR}_{n}.

  • If nn doubles are all connected to each other by an edge {RL}\{RL\}, we get an ordered graph MnRLM^{RL}_{n}.

  • If nn doubles are all connected to each other by edges {LR,RL}\{LR,RL\}, we get an ordered graph Mn+M^{+}_{n}.

  • The rest of the cases is if nn doubles are all connected to each other by edges {LL}\{LL\} or {RR}\{RR\} or any other sets of edges containing these edges.

    {LL,RR},{LL,RL},{LL,LR},{RR,RL},{RR,LR},{LL,RR,LR},\displaystyle\{LL,RR\},\{LL,RL\},\{LL,LR\},\{RR,RL\},\{RR,LR\},\{LL,RR,LR\},
    {LL,RR,LR},{LL,RL,LR},{RR,RL,LR},{LL,RR,LR,RL}\displaystyle\{LL,RR,LR\},\{LL,RL,LR\},\{RR,RL,LR\},\{LL,RR,LR,RL\}

    Then we get a complete ordered graph KnK_{n} (or bigger complete ordered graph, e.g. K2nK_{2n} on {LL,RR,LR,RL}\{LL,RR,LR,RL\} edge set). We can easily see this, as if all the doubles are connected by the vertex on the same side of the double, these vertices will be connected in all doubles, and therefore form a complete graph.

The Ramsey’s Theorem tells us that for any given finite number of colours, cc, and any given integers n1,,ncn_{1},\ldots,n_{c}, there is a number, R(n1,,nc)R(n_{1},\ldots,n_{c}), such that if the edges of a complete graph of order R(n1,,nc)R(n_{1},\ldots,n_{c}) are coloured with cc different colours, then for some ii between 11 and cc, it must contain a complete subgraph of order nin_{i} whose edges are all of colour ii.

We notice that the ordered matching induced subgraph M~k\tilde{M}_{k} of GG can be thought of as such a complete graph JJ, where each double can be thought of as a vertex of JJ, and 24=162^{4}=16 possible edge connections can be though of as 1616 possible edge colors of JJ.

Then using the Ramsey’s Theorem we see that for any jj\in\mathbb{N}, there exists sufficiently large ordered graph JJ, such that JJ must contain at least one of 1616 complete graphs of order jj with all edges of the same color. This complete graph with all edges of the same color will then correspond to jj doubles in M~k\tilde{M}_{k} that are all connected to each other by the same edge set, which will result in one of the five induced ordered subgraphs of M~k\tilde{M}_{k}. Therefore with growing coloring of GG and consequent growing of M~k\tilde{M}_{k} (as we have shown before), also at least one of the five resulting ordered graphs listed above will grow.

We showed in Proposition 4.3 that all of the ordered graphs Mn,Km,MkRL,Ml+,n,k2,m,l3M_{n},K_{m},M^{RL}_{k},M^{+}_{l},n,k\geq 2,m,l\geq 3 are non-induced. We have also shown that their coloring grows with their size.

The last thing to show is that set of ordered graphs Mn,Km,MkRL,Ml+M_{n},K_{m},M^{RL}_{k},M^{+}_{l} together with an ordered graph MrLRM^{LR}_{r} is not non-induced. But this can be seen by taking the vertices b1,a2,b3,a4,b5,a6,b_{1},a_{2},b_{3},a_{4},b_{5},a_{6},\ldots of MkRLM^{RL}_{k} and seeing that resulting induced ordered subgraph is MrLRM^{LR}_{r}. Therefore the set of induced ordered graphs Mn,Kn,MnRL,Mn+M_{n},K_{n},M^{RL}_{n},M^{+}_{n} of GG is indeed minimal and sufficient to limit the size of χ(G)\chi(G).

As mentioned before, this then answers the analogy of the Gyaŕfaś–Sumner conjecture for χ\chi-boundedness of ordered graphs, with replacing the forbidden structures (of a tree and clique) in the original conjecture (with our four graph classes).

5 χ\chi-boundedness of Oriented and Directed Ordered Graphs

Let us now have a look at an analogy of previous results for the following structure. Let us allow single and double orientation of edges in between two vertices of an ordered graph GG (single in a sense that there is only one arrow in between two vertices and double when there are two arrows in both directions), then we shall call this a directed ordered graph and denote G\overleftrightarrow{G}. Let us then also define a Kn\overleftrightarrow{K}_{n} as a complete ordered graph with each edge having double orientation.

We then define a directed colouring χ(G)\overleftrightarrow{\chi}(\overleftrightarrow{G}) as the minimum nn such that there exists ordered homomorpshism from G\overleftrightarrow{G} to Kn\overleftrightarrow{K}_{n}. We notice that this notion leads to the similar results as the coloring of ordered graphs.

Theorem 5.1.

Let GG be an ordered graph and G\overleftrightarrow{G} a corresponding directed ordered graph with any orientation of the edges of GG. Then χ(G)=χ(G)\chi(G)=\overleftrightarrow{\chi}(\overleftrightarrow{G}).

Proof.

Let us assume an ordered homomorphism from GG to KnK_{n}. We can see, that if we impose left, right or double orientation on edge of GG, this edge can always map to the same edge in Kn\overleftrightarrow{K}_{n} as it did in KnK_{n}. We can also easily see that if there is an ordered homomorphism G\overleftrightarrow{G} to Kn\overleftrightarrow{K}_{n}, then we can simply forget the directions and replace them by simple edges and we get the same homomorphism from GG to KnK_{n}. ∎

We might therefore denote χ\overleftrightarrow{\chi} simply as χ\chi.

We can then of course adopt similar concept for Mk\overleftrightarrow{M_{k}} as we did for the ordered matching MkM_{k} by imposing an orientation on MkM_{k} and defining corresponding χ\chi-boundedness. We shall call Mk\overleftrightarrow{M_{k}} an Directed Ordered Matching.

Let us then define the following ordered graphs.

Right Oriented Ordered Matching MkRM^{R}_{k} has vertices ai,bi,i=1,,ka_{i},b_{i},i=1,\ldots,k, with ordering a1<b1<a2<b2<<ak<bka_{1}<b_{1}<a_{2}<b_{2}<\ldots<a_{k}<b_{k} and edges (ai,bi),i=1,,k(a_{i},b_{i}),i=1,\ldots,k. Left Oriented Ordered Matching MkLM^{L}_{k} has points ai,bi,i=1,,ka_{i},b_{i},i=1,\ldots,k, with ordering a1<b1<a2<b2<<ak<bka_{1}<b_{1}<a_{2}<b_{2}<\ldots<a_{k}<b_{k} and edges (bi,ai),i=1,,k(b_{i},a_{i}),i=1,\ldots,k. Directed Ordered Matching MkDM^{D}_{k} has points ai,bi,i=1,,ka_{i},b_{i},i=1,\ldots,k, with ordering a1<b1<a2<b2<<ak<bka_{1}<b_{1}<a_{2}<b_{2}<\ldots<a_{k}<b_{k} and edges (ai,bi)(a_{i},b_{i}) and (bi,ai),i=1,,k(b_{i},a_{i}),i=1,\ldots,k.

Then the following holds.

Corollary 5.2.

Let 𝒞k\mathcal{C}_{k} be a class of directed ordered graphs where for every G𝒞k\overleftrightarrow{G}\in\mathcal{C}_{k}, G\overleftrightarrow{G} has the biggest left, right oriented and directed ordered matching subgraph smaller or equal to MaRM^{R}_{a}, MbLM^{L}_{b} and McDM^{D}_{c}, resp., where a+b+cka+b+c\leq k. Then χ(G)2(a+b+c)+1\chi(\overleftrightarrow{G})\leq 2(a+b+c)+1 for every G𝒞k\overleftrightarrow{G}\in\mathcal{C}_{k}.

Proof.

As every directed ordered graph Mk\overleftrightarrow{M_{k}} consists of MaRM^{R}_{a}, MbLM^{L}_{b} and McDM^{D}_{c}, where (a+b+c)=k(a+b+c)=k, the statement follows from Theorem 4.1 and 5.1. ∎

Let us now consider an orientation of an ordered graph GG, where we allow only left or right orientation of edges (and not both). We shall call this an oriented ordered graph and denote G\overrightarrow{G}. The topic has been studied for standard graphs in [4].

One might hope that the similar approach as we applied for the directed graphs might work for limiting the coloring of the classes of oriented ordered graphs. Perhaps surprisingly, a similar statement does not hold for the oriented ordered graphs.

We shall notice first that the coloring of oriented ordered graphs using the intervals will not work. This is due to the fact that oriented ordered graphs do not allow edges with both directions. This will give us cases such as V(G)=(v1,v2,v3),E(G)={(v1,v2),(v3,v1)}V(\overrightarrow{G})=(v_{1},v_{2},v_{3}),E(\overrightarrow{G})=\{(v_{1},v_{2}),(v_{3},v_{1})\}, where v2v_{2} and v3v_{3} form an interval, but they cannot map into the same vertex, as we would create an edge with both directions.

We therefore define an oriented ordered coloring χ(G)\overrightarrow{\chi}(\overrightarrow{G}) as the order of minimum homomorphism image of oriented ordered graph G\overrightarrow{G} (which is in line with the definition of coloring of ordered graphs).

Let oriented ordered graph M2\overrightarrow{M}_{2} be an ordered matching M2M_{2} with any orientation and analogously an oriented ordered graph P3\overrightarrow{P}_{3} be an ordered path P3P_{3} with any orientation.

Proposition 5.3.

For every nn\in\mathbb{N} there exists an oriented ordered graph G\overrightarrow{G} that has each vertex of degree 11, order of G\overrightarrow{G} is 2n2n, G\overrightarrow{G} does not contain an oriented ordered subgraph M2\overrightarrow{M}_{2} or P3\overrightarrow{P}_{3}, and has χ(G)=n+1\chi(\overrightarrow{G})=n+1.

Proof.

We shall prove this statement by construction.

Let G\overrightarrow{G} be an oriented ordered graph on 2n2n vertices v1,v2,,v2nv_{1},v_{2},\ldots,v_{2n} with the following edge set (see Figure 3).

E(G)={(v1,v2n),(v2n1,v2),(v3,v2n2),}E(\overrightarrow{G})=\{(v_{1},v_{2n}),(v_{2n-1},v_{2}),(v_{3},v_{2n-2}),\ldots\}

We see that G\overrightarrow{G} does not contain neither M2\overrightarrow{M}_{2} nor P3\overrightarrow{P}_{3} and that every vertex of G\overrightarrow{G} has degree 11.

We notice that every time we map vertices vi1v_{i-1} and vi,2inv_{i},2\leq i\leq n in ordered homomorphism to one vertex, we cannot map vertices vn+i1v_{n+i-1} and vn+iv_{n+i} to one vertex in the same ordered homomorphism anymore, and vice versa. Therefore we have maximum n1n-1 options to map two vertices into one in ordered homomorphism (as we cannot map vertices vnv_{n} and vn+1v_{n+1} to one vertex). Therefore the order of minimum homomorphism image of G\overrightarrow{G} is 2n(n1)=n+12n-(n-1)=n+1. ∎

Refer to caption
Figure 3: G\overrightarrow{G} from the Proposition 5.3

Natural next step would be to investigate a complexity of determining χ(G)\overrightarrow{\chi}(\overrightarrow{G}), but we shall not address it in this paper.

We might only mention that greedy algorithm will not help us in this case, as the inability to create an edge with both directions will affect the rule within the greedy algorithm, as we will need to check if by mapping the vertices in an interval into one vertex we shall not create an edge with both directions in our homomorphism image. This will then not always yield an oriented ordered graph, that has number of vertices equal to the coloring of the input oriented ordered graph. An example of oriented ordered graph that would not yield the same homomorphism image after running the (adjusted) greedy algorithm from left or right is a graph V(G)=(v1,v2,v3,v4,v5),E(G)={(v3,v1),(v5,v1),(v2,v4),(v3,v5)}V(\overrightarrow{G})=(v_{1},v_{2},v_{3},v_{4},v_{5}),E(\overrightarrow{G})=\{(v_{3},v_{1}),(v_{5},v_{1}),(v_{2},v_{4}),(v_{3},v_{5})\}, where starting the greedy algorithm with the adjusted rule from left would yield a χ(G)=4\overrightarrow{\chi}(\overrightarrow{G})=4, while starting it from the right would give us a χ(G)=3\overrightarrow{\chi}(\overrightarrow{G})=3.

This makes a theory of ordered graphs yet even more fascinating.

We focus on the complexity of finding ordered homomorphims and cores of ordered graphs in prepared papers [12] and [11], resp.

References

  • [1] S. Arora and B. Barak. Computational Complexity - A Modern Approach. Cambridge, 2009.
  • [2] Martin Balko, Vít Jelínek, and Pavel Valtr. On ordered ramsey numbers of bounded-degree graphs. Journal of Combinatorial Theory, Series B, 134:179–202, jan 2019.
  • [3] Martin Balko and Marian Poljak. On ordered ramsey numbers of matchings versus triangles, 2023.
  • [4] O.V. Borodin, A.V. Kostochka, J. Nešetřil, A. Raspaud, and E. Sopena. On the maximum average degree and the oriented chromatic number of a graph. Discrete Mathematics, 206(1):77–89, 1999.
  • [5] P. Hell and J. Nešetřil. Graphs and Homomorphisms. Oxford University Press, 2004.
  • [6] Jan Hubička and Jaroslav Nešetřil. All those ramsey classes (ramsey classes with closures and forbidden homomorphisms). Advances in Mathematics, 356:106791, nov 2019.
  • [7] Jaroslav Nešetřil and Vojtěch Rödl. Statistics of orderings. Abhandlungen aus dem Mathematischen Seminar der Universität Hamburg, 87, 01 2017.
  • [8] Jaroslav Nešetřil and Claude Tardif. Duality theorems for finite structures (characterising gaps and good characterisations). Journal of Combinatorial Theory, Series B, 80(1):80–97, 2000.
  • [9] János Pach and Gábor Tardos. Forbidden paths and cycles in ordered graphs and matrices. Israel Journal of Mathematics, 155(1):359–380, Dec 2006.
  • [10] Pierre Simon. A guide to nip theories, 2014.
  • [11] Michal Čertík, Andreas Emil Feldmann, Jaroslav Nešetřil, and Paweł Rzążewski. Complexity of finding core of ordered graphs. In Preparation.
  • [12] Michal Čertík, Andreas Emil Feldmann, Jaroslav Nešetřil, and Paweł Rzążewski. Complexity of finding homomorphisms of ordered graphs. In Preparation.
  • [13] Édouard Bonnet, Eun Jung Kim, Stéphan Thomassé, and Rémi Watrigant. Twin-width i: tractable fo model checking, 2021.