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

Odd Colorings of Sparse Graphs

Daniel W. Cranston Department of Computer Science, Virginia Commonwealth University, Richmond, VA, USA; dcranston@vcu.edu
Abstract

A proper coloring of a graph is called odd if every non-isolated vertex has some color that appears an odd number of times on its neighborhood. The smallest number of colors that admits an odd coloring of a graph GG is denoted χo(G)\chi_{o}(G). This notion was introduced by Petruševski and Škrekovski, who proved that if GG is planar then χo(G)9\chi_{o}(G)\leq 9; they also conjectured that χo(G)5\chi_{o}(G)\leq 5. For a positive real number α\alpha, we consider the maximum value of χo(G)\chi_{o}(G) over all graphs GG with maximum average degree less than α\alpha; we denote this value by χo(𝒢α)\chi_{o}(\mathcal{G}_{\alpha}). We note that χo(𝒢α)\chi_{o}(\mathcal{G}_{\alpha}) is undefined for all α4\alpha\geq 4. In contrast, for each α[0,4)\alpha\in[0,4), we give a (nearly sharp) upper bound on χo(𝒢α)\chi_{o}(\mathcal{G}_{\alpha}). Finally, we prove χo(𝒢20/7)=5\chi_{o}(\mathcal{G}_{20/7})=5 and χo(𝒢3)=6\chi_{o}(\mathcal{G}_{3})=6. Both of these results are sharp.


1 Introduction

A proper coloring of a graph is oddmargin: odd coloring if every non-isolated vertex has some color that appears an odd number of times on its neighborhood. The smallest number of colors that admits an odd coloring of a graph GG is denoted χo(G)\chi_{o}(G)margin: χo(G)\chi_{o}(G) . Clearly, χo(G)|V(G)|\chi_{o}(G)\leq|V(G)|, since we can simply color each vertex with its own color. This notion was introduced by Petruševski and Škrekovski [13], who proved that if GG is planar then χo(G)9\chi_{o}(G)\leq 9; they also conjectured that χo(G)5\chi_{o}(G)\leq 5.

Odd coloring is motivated by various types of hypergraph coloring. A hypergraph \mathcal{H} consists of a set 𝒱\mathcal{V} of vertices and a set \mathcal{E} of (hyper)edges, each of which consists of an arbitrary set of vertices in 𝒱\mathcal{V}. Most varieties of hypergraph coloring assign colors (integers in {1,,k}\{1,\ldots,k\}) to the elements of 𝒱\mathcal{V} subject to certain constraints. Standard hypergraph coloring requires only that no edge in \mathcal{E} is monochromatic. Even et al. [6] introduced conflict-free coloring, which requires that each edge in \mathcal{E} has some color that appears exactly once on its vertices. This topic has been widely studied [5, 8, 9, 12, 14]. Cheilaris et al. [4] studied odd coloring of hypergraphs, which requires that each edge in \mathcal{E} has some color that appears an odd number of times on its vertices. Aspects of this problem have been studied in [1, 2, 7]. For graphs, Cheilaris [3] studied conflict-free colorings with respect to open neighborhoods. That is, for each vertex vv some color appears exactly once on N(v)N(v). Finally, Petruševski and Škrekovski [13] studied odd colorings of graphs, which are proper colorings where each vertex vv has some color that appears an odd number of times on N(v)N(v). It is this parameter that we consider in the present short note.

The average degree of a graph HH is 2|E(H)|/|V(H)|2|E(H)|/|V(H)|. The maximum average degree of a graph GG, denoted mad(G)\mathrm{mad}(G)margin: mad(G)\mathrm{mad}(G) , is the maximum, over all non-empty subgraphs HH of GG, of the average degree of HH. That is, mad(G):=maxHG2|E(H)|/|V(H)|\mathrm{mad}(G):=\max_{H\subseteq G}2|E(H)|/|V(H)|. For each positive real number α\alpha, let 𝒢α\mathcal{G}_{\alpha} denote the family of graphs GG with mad(G)<α\mathrm{mad}(G)<\alpha. We denote by χo(𝒢α)\chi_{o}(\mathcal{G}_{\alpha})margin: 𝒢α\mathcal{G}_{\alpha}, χo(𝒢α)\chi_{o}(\mathcal{G}_{\alpha}) the maximum value of χo(G)\chi_{o}(G) over all G𝒢αG\in\mathcal{G}_{\alpha}. The focus of this paper is bounding χo(𝒢α)\chi_{o}(\mathcal{G}_{\alpha}) for various values of α\alpha. We observe that χo(𝒢α)\chi_{o}(\mathcal{G}_{\alpha}) is undefined for all α4\alpha\geq 4. That is, there exists a sequence of graphs GnG_{n} such that χo(Gn)=n\chi_{o}(G_{n})=n and mad(Gn)<4\mathrm{mad}(G_{n})<4 for all nn. In contrast, for each α[0,4)\alpha\in[0,4), we give a (nearly sharp) upper bound on χo(𝒢α)\chi_{o}(\mathcal{G}_{\alpha}). We have two main results.

Theorem 1.

Fix ϵ\epsilon such that 0<ϵ8/50<\epsilon\leq 8/5. If mad(G)4ϵ\mathrm{mad}(G)\leq 4-\epsilon, then χo(G)8/ϵ+2\chi_{o}(G)\leq\lfloor 8/\epsilon\rfloor+2. As ϵ0\epsilon\to 0, infinitely often there exists GϵG_{\epsilon} such that mad(Gϵ)=4ϵ\mathrm{mad}(G_{\epsilon})=4-\epsilon and χo(Gϵ)=8/ϵ1\chi_{o}(G_{\epsilon})=\lfloor 8/\epsilon\rfloor-1.

When ϵ{1,8/7}\epsilon\in\{1,8/7\}, we prove sharper upper bounds on χo(G4ϵ)\chi_{o}(G_{4-\epsilon}).

Theorem 2.

Fix a graph GG. (a) If mad(G)<3\mathrm{mad}(G)<3, then χo(G)6\chi_{o}(G)\leq 6; and (b) if mad(G)<20/7\mathrm{mad}(G)<20/7, then χo(G)5\chi_{o}(G)\leq 5. Furthermore, neither of these upper bounds on χo(G)\chi_{o}(G) can be decreased, and neither of the inequalities can be weakened to allow equality.

For a proper coloring φ\varphi of some subgraph HH of GG, let φo(v)\varphi_{o}(v)margin: φo(v)\varphi_{o}(v) denote the unique color that appears an odd number of times in NH(v)N_{H}(v) if such a color exists; otherwise, φo(v)\varphi_{o}(v) is undefined. Most of the rest of our notation and definitions are standard. But for the reader’s convenience we highlight a few terms. The girthmargin: girth of a graph GG is the length of a shortest cycle in GG (the girth of an acyclic graph is infinite). A kk-vertexmargin: kk/k+k^{+}/kk^{-}-vertex is a vertex of degree kk. A k+k^{+}-vertex (resp. kk^{-}-vertex) is a vertex of degree at least (resp. at most) kk. A kk-neighbormargin: kk/k+k^{+}/kk^{-}-neighbor of a vertex vv is a neighbor of vv that is a kk-vertex. Both k+k^{+}-neighbor and kk^{-}-neighbor are defined analogously.

To close this introduction, we prove three easy results about graphs GG with χo(G)4\chi_{o}(G)\leq 4.

Proposition 1.

A graph GG has χo(G)=1\chi_{o}(G)=1 if and only if GG has vertices but no edges. And GG has χo(G)=2\chi_{o}(G)=2 if and only if GG is bipartite and the degree of each vertex in GG is either 0 or odd.

Proof.

The first statement is obvious, so now we prove the second. If GG is bipartite, then GG has a proper 2-coloring φ\varphi. If the degree of each vertex is either 0 or odd, then φ\varphi is also an odd 2-coloring of GG. If GG is not bipartite, then GG has no proper 2-coloring, so clearly χo(G)>2\chi_{o}(G)>2. Suppose instead GG is bipartite, but some vertex vv has positive even degree. Now the component of GG containing vv has only a single 2-coloring (up to permuting colors), but this 2-coloring is not odd since N(v)N(v) has only a single color, which appears an even number of times. ∎

Proposition 2.

If 3|n3|n, then χo(Cn)=3\chi_{o}(C_{n})=3. If n=5n=5, then χo(Cn)=5\chi_{o}(C_{n})=5. Otherwise, χo(Cn)=4\chi_{o}(C_{n})=4.

Proof.

If 3|n3|n, then we can repeat the colors 1,2,3,1,2,3,1,2,3,1,2,3,\ldots around CnC_{n} to get an odd 3-coloring. Suppose instead that 3n3\nmid n. By the Pigeonhole principle, in every proper 3-coloring φ\varphi some color appears twice on the neighborhood of some vertex vv, so φ\varphi is not odd. Thus, χo(Cn)4\chi_{o}(C_{n})\geq 4. If 3n3\nmid n and n5n\neq 5, then we can begin with 1,2,3,41,2,3,4 or 1,2,3,4,1,2,3,41,2,3,4,1,2,3,4 and continue with 1,2,3,1,2,31,2,3,1,2,3\ldots. Finally, it is easy to check that each proper 4-coloring φ\varphi of C5C_{5} uses the same color on both neighbors of some vertex, so φ\varphi is not an odd coloring. Thus, χo(C5)=5\chi_{o}(C_{5})=5. ∎

It is interesting to note that the class of graphs GG with mad(G)=2\mathrm{mad}(G)=2 and χo(G)=4\chi_{o}(G)=4 is richer than simply cycles CnC_{n} with 3n3\nmid n. Denote the vertices of CnC_{n} by v1,,vnv_{1},\ldots,v_{n}. Starting from such a graph, we can add arbitrarily many leaves at each viv_{i} with 3|i3|i. It is straightforward to check that the resulting graph GG^{\prime} also has mad(G)=2\mathrm{mad}(G^{\prime})=2 and χo(G)=4\chi_{o}(G^{\prime})=4. More generally, we can identify these viv_{i} with vertices of an arbitrary graph and χo\chi_{o} will not decrease.

Proposition 3.

Every tree TT has χo(T)3\chi_{o}(T)\leq 3. Thus, if mad(T)<2\mathrm{mad}(T)<2, then χo(T)3\chi_{o}(T)\leq 3. This bound on mad(T)\mathrm{mad}(T) is sharp.

Proof.

Let TT be a tree. We use induction on |T||T|. The case |T|=1|T|=1 is trivial. Now suppose that |T|2|T|\geq 2. Let vv be a leaf of TT. Let T:=TvT^{\prime}:=T-v and note that TT^{\prime} is a tree. By hypothesis, TT^{\prime} has an odd 3-coloring φ\varphi. Let ww be the neighbor of vv in TT. To extend φ\varphi to an odd 3-coloring of TT, we color vv with a color outside {φ(w),φo(w)}\{\varphi(w),\varphi_{o}(w)\}. This gives the desired odd 3-coloring of TT.

If mad(G)<2\mathrm{mad}(G)<2, then GG is a forest. So the second statement follows from the first. Finally, the third statement follows from Proposition 2. ∎

The following proposition is folklore. But we include a proof for completeness.

Proposition 4.

If GG is a planar graph with girth at least gg, then mad(G)<2g/(g2)\mathrm{mad}(G)<2g/(g-2).

Proof.

Fix a plane embedding of a planar graph GG with girth at least gg. Since each subgraph of GG also has girth at least gg, it suffices to show that 2|E(G)|/|V(G)|<2g/(g2)2|E(G)|/|V(G)|<2g/(g-2). Summing the lengths of all facial walks gives 2|E(G)|g|F|2|E(G)|\geq g|F|, where FF is the set of all faces. To prove the proposition, we substitute this inequality into Euler’s formula and solve for 2|E(G)|/|V(G)|2|E(G)|/|V(G)|. ∎

2 Mad(G)<4\mathrm{Mad}(G)<4

Recall, for each α>0\alpha>0, that 𝒢α\mathcal{G}_{\alpha} denotes the set of all graphs GG with mad(G)<α\mathrm{mad}(G)<\alpha; and we write χo(𝒢α)\chi_{o}(\mathcal{G}_{\alpha}) to denote the maximum value of χo(G)\chi_{o}(G) over all G𝒢αG\in\mathcal{G}_{\alpha}. Proposition 3 shows that χo(G2)=3\chi_{o}(G_{2})=3. In this section, we determine the set of all values α\alpha such that χo(𝒢α)\chi_{o}(\mathcal{G}_{\alpha}) is defined; when it is defined, we prove a (nearly sharp) upper bound on its value.

We denote by KnK_{n}^{*}margin: KnK_{n}^{*} the graph formed from KnK_{n} by subdiving each edge once.

Lemma 5.

We have χo(Kn)=n\chi_{o}(K_{n}^{*})=n and mad(Kn)=48/(n+1)\mathrm{mad}(K_{n}^{*})=4-8/(n+1).

Proof.

For each n1n\geq 1, let ϵn:=8/(n+1)\epsilon_{n}:=8/(n+1). To show that mad(Kn)=4ϵn\mathrm{mad}(K^{*}_{n})=4-\epsilon_{n}, we give a fractional orientation of KnK^{*}_{n} where each vertex has indegree exactly 2ϵn/22-\epsilon_{n}/2. We orient each edge with fraction 1ϵn/41-\epsilon_{n}/4 towards its endpoint of degree 2 and fraction ϵn/4\epsilon_{n}/4 toward its endpoint of degree nn. Each 2-vertex has indegree 2(1ϵn/4)=2ϵn/22(1-\epsilon_{n}/4)=2-\epsilon_{n}/2. Each (n1)(n-1)-vertex has indegree (n1)(ϵn/4)=(n1)(2/(n+1))=(2n2)/(n+1)=24/(n+1)=2ϵn/2(n-1)(\epsilon_{n}/4)=(n-1)(2/(n+1))=(2n-2)/(n+1)=2-4/(n+1)=2-\epsilon_{n}/2. Thus, mad(G)=4ϵn\mathrm{mad}(G)=4-\epsilon_{n}, as desired.

In an odd coloring of KnK^{*}_{n}, all (n1)(n-1)-vertices must get distinct colors. So χo(Kn)n\chi_{o}(K^{*}_{n})\geq n. Given any coloring where all (n1)(n-1)-vertices get distinct colors, it is easy to extend to an odd nn-coloring of KnK^{*}_{n}. Thus, as claimed, χo(Kn)=n\chi_{o}(K^{*}_{n})=n. ∎

Corollary 6.

There exists a sequence ϵ1,ϵ2,\epsilon_{1},\epsilon_{2},\ldots such that ϵn>0\epsilon_{n}>0 for all n1n\geq 1 and limnϵn=0\lim_{n\to\infty}\epsilon_{n}=0 and for each n1n\geq 1 there exists a graph GnG_{n} with mad(Gn)=4ϵn\mathrm{mad}(G_{n})=4-\epsilon_{n} and χo(Gn)=8/ϵn1\chi_{o}(G_{n})=8/\epsilon_{n}-1.

Proof.

Let ϵn:=8/(n+1)\epsilon_{n}:=8/(n+1) and Gn:=KnG_{n}:=K^{*}_{n}. Now χo(Gn)=n=(n+1)1=8/ϵn1\chi_{o}(G_{n})=n=(n+1)-1=8/\epsilon_{n}-1. ∎

Corollary 7.

χo(𝒢α)\chi_{o}(\mathcal{G}_{\alpha}) is undefined whenever α4\alpha\geq 4.

In Lemma 5, we considered KnK^{*}_{n} which is formed by subdividing each edge of KnK_{n}. Applying the same construction to any nn-chromatic graph HH yields a graph HH^{\prime} with χo(H)=n\chi_{o}(H^{\prime})=n and mad(H)<4\mathrm{mad}(H^{\prime})<4. Since there exist graphs HH with both chromatic number and girth arbitrarily large, there also exist graphs HH^{\prime} with χo(H)\chi_{o}(H^{\prime}) and girth arbitrarily large, and with mad(H)<4\mathrm{mad}(H^{\prime})<4. However, every nn-chromatic graph HH that does not contain KnK_{n} as a subgraph gives an upper bound on 4ϵ4-\epsilon that is worse (larger) than that in Corollary 6. Consider an nn-critical (sub)graph HH with aa vertices and bb edges. Recall that δ(H)n1\delta(H)\geq n-1. But HH is not (n1)(n-1)-regular, by Brooks’ Theorem. Thus, b=|E(H)|>(n1)|V(H)|/2=(n1)a/2b=|E(H)|>(n-1)|V(H)|/2=(n-1)a/2. Subdividing each edge of HH gives HH^{\prime} with |V(H)|=a+b|V(H^{\prime})|=a+b and |E(H)|=2b>a(n1)|E(H^{\prime})|=2b>a(n-1). Thus, mad(H)2|E(H)|/|V(H)|=4b/(a+b)>2a(n1)/(a(n1)/2+a)=2a(n1)/(a(n+1)/2)=4(n1)/(n+1)=48/(n+1)\mathrm{mad}(H^{\prime})\geq 2|E(H^{\prime})|/|V(H^{\prime})|=4b/(a+b)>2a(n-1)/(a(n-1)/2+a)=2a(n-1)/(a(n+1)/2)=4(n-1)/(n+1)=4-8/(n+1).

We next show that the construction KnK_{n}^{*} in Lemma 5 is nearly sharp.

Theorem 1.

Fix ϵ\epsilon such that 0<ϵ8/50<\epsilon\leq 8/5. If mad(G)4ϵ\mathrm{mad}(G)\leq 4-\epsilon, then χo(G)8/ϵ+2\chi_{o}(G)\leq\lfloor 8/\epsilon\rfloor+2. As ϵ0\epsilon\to 0, infinitely often there exists GϵG_{\epsilon} such that mad(Gϵ)=4ϵ\mathrm{mad}(G_{\epsilon})=4-\epsilon and χo(Gϵ)=8/ϵ1\chi_{o}(G_{\epsilon})=\lfloor 8/\epsilon\rfloor-1.

Proof.

The second statement follows directly from Corollary 6.

For the first statement, let k:=8/ϵ+2k:=\lfloor 8/\epsilon\rfloor+2margin: kk and note that k7k\geq 7. Our proof is by induction on |V(G)||V(G)|. The base case is when |V(G)|=1|V(G)|=1, so χo(G)=1\chi_{o}(G)=1. Instead assume |V(G)|2|V(G)|\geq 2. If GG contains a 1-vertex vv with neighbor ww, then GvG-v has an odd kk-coloring φ\varphi and we can extend φ\varphi to vv by coloring vv to avoid φ(w)\varphi(w) and φo(w)\varphi_{o}(w). If GG contains a 3-vertex vv, then denote N(v)N(v) by {w1,w2,w3}\{w_{1},w_{2},w_{3}\}. Now GvG-v has an odd kk-coloring φ\varphi and we can extend φ\varphi to vv by coloring vv to avoid {φ(w1),φ(w2),φ(w3),φo(w1),φo(w2),φo(w3)}\{\varphi(w_{1}),\varphi(w_{2}),\varphi(w_{3}),\varphi_{o}(w_{1}),\varphi_{o}(w_{2}),\varphi_{o}(w_{3})\}. So we assume that GG has neither 1-vertices nor 3-vertices. Suppose that GG has adjacent 2-vertices v1v_{1} and v2v_{2}, and denote the remaining neighbors of v1v_{1} and v2v_{2}, respectively, by v0v_{0} and v3v_{3}. Now G{v1,v2}G-\{v_{1},v_{2}\} has an odd kk-coloring φ\varphi. To extend this to v1v_{1} and v2v_{2}, we first color v1v_{1} to avoid {φ(v0),φo(v0),φ(v3)}\{\varphi(v_{0}),\varphi_{o}(v_{0}),\varphi(v_{3})\} and then color v2v_{2} to avoid the new color on v1v_{1} as well as {φ(v0),φ(v3),φo(v3)}\{\varphi(v_{0}),\varphi(v_{3}),\varphi_{o}(v_{3})\}.

Since mad(G)<4\mathrm{mad}(G)<4, we know that δ(G)=2\delta(G)=2. We use discharging to show that some vertex vv with d(v)8/ϵ2d(v)\leq 8/\epsilon-2 has “many” 2-neighbors. Similar to the case of adjacent 2-vertices above, we will delete vv and its 2-neighbors, find an odd kk-coloring φ\varphi for this smaller graph, and extend φ\varphi to all of GG. Let x:=1ϵ/2x:=1-\epsilon/2margin: xx . We give each vertex vv initial charge d(v)d(v) and use the following single discharging rule: Each 2-vertex takes charge xx from each neighbor. Since GG does not have adjacent 2-vertices, each 2-vertex finishes with charge 2+2x=4ϵ2+2x=4-\epsilon. For each 4+4^{+}-vertex vv, let d2(v)d_{2}(v)margin: d2(v)d_{2}(v) denote the number of 2-neighbors of vv. Now vv finishes with charge d(v)xd2(v)d(v)-xd_{2}(v). If d(v)xd2(v)>2+2x=4ϵd(v)-xd_{2}(v)>2+2x=4-\epsilon for each vv, then we contradict the assumption mad(G)4ϵ\mathrm{mad}(G)\leq 4-\epsilon. So there exists vv such that d(v)xd2(v)2+2xd(v)-xd_{2}(v)\leq 2+2x.

Form GG^{\prime} from GG by deleting vv and all of its 2-neighbors. By induction, GG^{\prime} has an odd kk-coloring φ\varphi. To extend φ\varphi to GG, we first color vv to avoid the colors on the colored neighbors (in GG) of its deleted 2-neighbors (in GG^{\prime}) as well as, for each 4+4^{+}-neighbor ww, to avoid φ(w)\varphi(w) and φo(w)\varphi_{o}(w). For convenience, we denote this color used on vv by φ(v)\varphi(v). Then we color each 2-neighbor xx of vv, with other neighbor yy, to avoid {φ(v),φo(v),φ(y),φo(y)}\{\varphi(v),\varphi_{o}(v),\varphi(y),\varphi_{o}(y)\}. This gives an odd coloring φ\varphi^{\prime} of GG. We must only show that φ\varphi^{\prime} uses at most kk colors. In total, vv must avoid at most 2d(v)d2(v)2d(v)-d_{2}(v) colors. So it will suffice to show that k1+(2d(v)d2(v))k\geq 1+(2d(v)-d_{2}(v)). Note that 1+2d(v)d2(v)<1+d(v)+(d(v)xd2(v))1+d(v)+2+2x<1+d(v)+41+2d(v)-d_{2}(v)<1+d(v)+(d(v)-xd_{2}(v))\leq 1+d(v)+2+2x<1+d(v)+4. Since 1+2d(v)d2(v)1+2d(v)-d_{2}(v) is an integer, 1+2d(v)d2(v)d(v)+41+2d(v)-d_{2}(v)\leq d(v)+4. So now we must bound d(v)d(v). Clearly, d(v)d(v)xd(v)d2(v)x2+2xd(v)-d(v)x\leq d(v)-d_{2}(v)x\leq 2+2x. So d(v)(2+2x)/(1x)=(4ϵ)/(ϵ/2)=8/ϵ2d(v)\leq(2+2x)/(1-x)=(4-\epsilon)/(\epsilon/2)=8/\epsilon-2. Since d(v)d(v) is an integer, d(v)8/ϵ2d(v)\leq\lfloor 8/\epsilon\rfloor-2. Thus, d(v)+4(8/ϵ2)+4=8/ϵ+2=kd(v)+4\leq(\lfloor 8/\epsilon\rfloor-2)+4=\lfloor 8/\epsilon\rfloor+2=k. ∎

Remark 8.

In the previous proof we handled adjacent 2-vertices as follows. “Suppose that GG has adjacent 2-vertices v1v_{1} and v2v_{2}, and denote the remaining neighbors of v1v_{1} and v2v_{2}, respectively, by v0v_{0} and v3v_{3}. Now G{v1,v2}G-\{v_{1},v_{2}\} has an odd kk-coloring φ\varphi. To extend this to v1v_{1} and v2v_{2}, we first color v1v_{1} to avoid {φ(v0),φo(v0),φ(v3)}\{\varphi(v_{0}),\varphi_{o}(v_{0}),\varphi(v_{3})\} and then color v2v_{2} to avoid the new color on v1v_{1} as well as {φ(v0),φ(v3),φo(v3)}\{\varphi(v_{0}),\varphi(v_{3}),\varphi_{o}(v_{3})\}.” In this argument we implicitly assumed that v3v0v_{3}\neq v_{0}. Otherwise, our choice for color on v1v_{1} might create a problem for v3v_{3}. Specifically, suppose that v0v_{0} is adjacent to 2-vertices v1v_{1} and v2v_{2}, which are also adjacent to each other, and that colors 1 and 2 each appear once on N(v)N(v) under φ\varphi (and no other color appears an odd number of times); this is possible when v3=v0v_{3}=v_{0}. Now φo(v0)\varphi_{o}(v_{0}) is undefined. But if we are careless and color v1v_{1} with 1 and color v2v_{2} with 2, then we create a problem for v0v_{0}. This obstacle has an easy work-around. Rather than defining φo(v0)\varphi_{o}(v_{0}) relative to the coloring φ\varphi, we define it relative to the current coloring, including any vertices that were deleted but are now already colored. For brevity, we omit mention of this issue throughout the paper, since the solution above always works.

It is natural to consider a list-coloring analogue of odd coloring (although we omit a formal definition). We note that, with minor modifications, the proof of Theorem 1 works equally well in the context of odd list-coloring and even in the context of odd correspondence coloring. The same is true for the proofs of Theorem 2(a) and Theorem 2(b).

We suspect that the construction in Lemma 5 is sharp.

Conjecture 1.

Fix ϵ\epsilon such that 0<ϵ8/50<\epsilon\leq 8/5. If mad(G)4ϵ\mathrm{mad}(G)\leq 4-\epsilon, then χo(G)8/ϵ1\chi_{o}(G)\leq\lfloor 8/\epsilon\rfloor-1.

Conjecture 1 cannot be extended to allow ϵ=2\epsilon=2, since mad(C3s+1)=2\mathrm{mad}(C_{3s+1})=2, but χo(C3s+1)=4\chi_{o}(C_{3s+1})=4, for each positive integer ss. In contrast, when ϵ=1\epsilon=1 and ϵ=8/7\epsilon=8/7, we prove the conjecture in a stronger form. We show that if mad(G)4ϵ\mathrm{mad}(G)\leq 4-\epsilon, then χo(G)8/ϵ2\chi_{o}(G)\leq\lfloor 8/\epsilon\rfloor-2 unless GG contains as a subgraph KsK^{*}_{s}, where s:=8/ϵ1s:=\lfloor 8/\epsilon\rfloor-1. We present these proofs in Sections 3 and 4.

Recall that a graph HH is kk-criticalmargin: kk-critical if χ(H)=k\chi(H)=k and χ(He)<k\chi(H-e)<k for every edge eE(H)e\in E(H). Analogously, a graph HH is odd-kk-criticalmargin: odd-kk-critical if χo(H)=k\chi_{o}(H)=k and χo(He)<k\chi_{o}(H-e)<k for every edge eE(H)e\in E(H). The latter notion is more subtle111An interesting example is C5C_{5}, since χo(C5)=5\chi_{o}(C_{5})=5, but C5C_{5} has no odd-4-critical subgraph., since possibly HH^{\prime} is a subgraph of HH, but χo(H)>χo(H)\chi_{o}(H^{\prime})>\chi_{o}(H). More concretely, if we form HH from KnK^{*}_{n} with n3n\geq 3 by adding a leaf adjacent to each vertex, then χo(H)3\chi_{o}(H)\leq 3, even though χo(Kn)=n\chi_{o}(K^{*}_{n})=n. It is easy to check that subdividing every edge of a kk-critical graph, with k6k\geq 6, yields an odd-kk-critical graph. Kostochka and Yancey [10, 11] showed that the sparsest kk-critical graphs are so-called kk-Ore graphs. For each k6k\geq 6, by subdividing all edges in kk-Ore graphs, we get an infinite family of odd-kk-critical graphs with maximum average degrees slightly larger than 48/(k+3)4-8/(k+3). We suspect that in the limit no infinite family of odd-kk-critical graphs has smaller maximum average degree.

3 Odd 6-Colorings

The goal of this section is to prove the following result.

Theorem 2(a).

If mad(G)<3\mathrm{mad}(G)<3, then χo(G)6\chi_{o}(G)\leq 6. This includes all planar GG of girth at least 6.

Note that Theorem Theorem 2(a) is sharp in two senses. First, mad(K6)=48/(6+1)=2+6/7\mathrm{mad}(K_{6}^{*})=4-8/(6+1)=2+6/7 and χo(K6)=6\chi_{o}(K_{6}^{*})=6, so the upper bound of 6 cannot be decreased. Second, mad(K7)=48/(7+1)=3\mathrm{mad}(K_{7}^{*})=4-8/(7+1)=3 and χo(K7)=7\chi_{o}(K_{7}^{*})=7, so the hypothesis mad(G)<3\mathrm{mad}(G)<3 cannot be weakened at all.

Proof.

The second statement follows from the first by Proposition 4. So we prove the first. Assume the statement is false, and GG is a counterexample with as few vertices as possible. As in the proof of Theorem 1, we assume GG has no 1-vertex and GG has no adjacent 2-vertices. Suppose GG has a 3-vertex vv with neighbors w1,w2,w3w_{1},w_{2},w_{3}, where d(w1)=2d(w_{1})=2. Denote the other neighbor of w1w_{1} by x1x_{1}. Let G:=G{v,w1}G^{\prime}:=G-\{v,w_{1}\}. By minimality, GG^{\prime} has an odd 6-coloring φ\varphi. To extend φ\varphi to GG, we color vv to avoid {φ(w1),φ(w2),φ(w3),φo(w2),φo(w3)}\{\varphi(w_{1}),\varphi(w_{2}),\varphi(w_{3}),\varphi_{o}(w_{2}),\varphi_{o}(w_{3})\} and then color w1w_{1} to avoid the new color on vv and also to avoid {φ(x1),φo(x1)}\{\varphi(x_{1}),\varphi_{o}(x_{1})\}. Thus, no 3-vertex has a 2-neighbor.

Suppose GG has a 4-vertex with neighbors w1,w2,w3,w4w_{1},w_{2},w_{3},w_{4}, where d(w1)=d(w2)=d(w3)=2d(w_{1})=d(w_{2})=d(w_{3})=2. For each i{1,2,3}i\in\{1,2,3\}, denote the other neighbor of wiw_{i} by xix_{i}. Let G:=G{v,w1,w2,w3}G^{\prime}:=G-\{v,w_{1},w_{2},w_{3}\}. By minimality, GG^{\prime} has an odd 6-coloring φ\varphi. To extend φ\varphi to GG, color vv to avoid {φ(x1),φ(x2),φ(x3)\{\varphi(x_{1}),\varphi(x_{2}),\varphi(x_{3}), φ(w4),φo(w4)}\varphi(w_{4}),\varphi_{o}(w_{4})\} and then color each wiw_{i} to avoid the new color on vv and also avoid {φ(xi)\{\varphi(x_{i}), φo(xi),φ(w4)}\varphi_{o}(x_{i}),\varphi(w_{4})\}. Thus, no 4-vertex in GG has three (or more) 2-neighbors.

Suppose GG has a 5-vertex vv with five 2-neighbors w1,w2,w3,w4,w5w_{1},w_{2},w_{3},w_{4},w_{5}. For each i{1,,5}i\in\{1,\ldots,5\}, denote the other neighbor of wiw_{i} by xix_{i}. Let G:=G{v,w1,w2,w3,w4,w5}G^{\prime}:=G-\{v,w_{1},w_{2},w_{3},w_{4},w_{5}\}. By minimality, GG^{\prime} has an odd 6-coloring φ\varphi. To extend φ\varphi to GG, we color vv to avoid {φ(x1),φ(x2),φ(x3),φ(x4),φ(x5)}\{\varphi(x_{1}),\varphi(x_{2}),\varphi(x_{3}),\varphi(x_{4}),\varphi(x_{5})\}. We then color each wiw_{i} to avoid the new color on vv, as well as {φ(xi),φo(xi),φ(v)}\{\varphi(x_{i}),\varphi_{o}(x_{i}),\varphi(v)\}. Thus, no 5-vertex in GG has five 2-neighbors.

Now we use discharging to reach a contradiction, which will finish the proof. We give each vertex vv initial charge d(v)d(v) and use the following single discharging rule: Each 2-vertex takes 1/21/2 from each neighbor. We show that each vertex finishes with charge at least 3, which contradicts the hypothesis mad(G)<3\mathrm{mad}(G)<3. Recall that δ(G)2\delta(G)\geq 2.

If d(v)=2d(v)=2, then vv finishes with 2+2(1/2)=32+2(1/2)=3. If d(v)=3d(v)=3, then vv has no 2-neighbors, so vv finishes with 3. If d(v)=4d(v)=4, then vv has at most two 2-neighbors, so vv finishes with at least 42(1/2)=34-2(1/2)=3. If d(v)=5d(v)=5, then vv has at most four 2-neighbors, so vv finishes with at least 54(1/2)=35-4(1/2)=3. If d(v)6d(v)\geq 6, then vv finishes with at least d(v)d(v)/2=d(v)/23d(v)-d(v)/2=d(v)/2\geq 3, since d(v)6d(v)\geq 6. This finishes the proof. ∎

With a bit more analysis, we can prove the following stronger result.

Corollary 9.

If mad(G)3\mathrm{mad}(G)\leq 3 and GG does not contain K7K^{*}_{7} as a subgraph, then χo(G)6\chi_{o}(G)\leq 6.

Proof.

In the proof of Theorem Theorem 2(a), to contradict the hypothesis mad(G)<3\mathrm{mad}(G)<3, we showed that each vertex finished with charge at least 3. To contradict the weaker hypothesis mad(G)3\mathrm{mad}(G)\leq 3, it suffices to show that also some vertex finishes with charge more than 3. Suppose the contrary.

Now GG has no 7+7^{+}-vertex and each 6-vertex has six 2-neighbors. Each 5-vertex has exactly four 2-neighbors and each 4-vertex has exactly two 2-neighbors. Now it is straightforward to check that GG cannot have any two adjacent 3+3^{+}-vertices vv and ww; in each case, we delete vv, ww, and their 2-neighbors, color the smaller graph and extend. Thus, GG is bipartite, where one part consists of 2-vertices and the other consists of 6-vertices. Form GG^{\prime} from GG by contracting one edge incident to each 2-vertex. Now GG^{\prime} is 6-regular, but no component is K7K_{7}, so χ(G)6\chi(G^{\prime})\leq 6 by Brooks’ Theorem. A 6-coloring of GG^{\prime} induces a 6-coloring of the 6-vertices in GG, which we can easily extend to an odd 6-coloring of GG. ∎

4 Odd 5-Colorings

The goal of this section is to prove the following result.

Theorem 2(b).

If mad(G)<20/7\mathrm{mad}(G)<20/7, then χo(G)5\chi_{o}(G)\leq 5. This includes all planar GG of girth at least 7.

Proof.

The second statement follows from the first by Proposition 4. So we prove the first. Assume the statement is false, and GG is a counterexample with as few vertices as possible. As in the proof of Theorem 1, we assume GG has no 1-vertex and GG has no adjacent 2-vertices. Suppose GG has a 3-vertex vv with neighbors w1,w2,w3w_{1},w_{2},w_{3}, where d(w1)=d(w2)=2d(w_{1})=d(w_{2})=2. Denote the other neighbors of w1w_{1} and w2w_{2} by, respectively, x1x_{1} and x2x_{2}. Let G:=G{v,w1,w2}G^{\prime}:=G-\{v,w_{1},w_{2}\}. By minimality, GG^{\prime} has an odd 5-coloring φ\varphi. To extend φ\varphi to GG, we color vv to avoid {φ(w1),φ(w2),φ(w3),φo(w3)}\{\varphi(w_{1}),\varphi(w_{2}),\varphi(w_{3}),\varphi_{o}(w_{3})\} and color each wiw_{i} to avoid the new color on vv and avoid {φ(xi),φo(xi),φ(w3)}\{\varphi(x_{i}),\varphi_{o}(x_{i}),\varphi(w_{3})\}. Thus, each 3-vertex has at most one 2-neighbor. More generally, each 3-vertex has at most one 33^{-}-neighbor. If not, then we delete vv and any 2-neighbors. We color vv to avoid the color on each 3-neighbor, the color on the colored neighbor of each (deleted) 2-neighbor, as well as {φ(w),φo(w)}\{\varphi(w),\varphi_{o}(w)\}, where ww is the third neighbor of vv. Thus, each 3-vertex has at most one 33^{-}-neighbor.

Suppose GG has a 4-vertex with four 2-neighbors w1,w2,w3,w4w_{1},w_{2},w_{3},w_{4}. For each i{1,2,3,4}i\in\{1,2,3,4\}, denote the other neighbor of wiw_{i} by xix_{i}. Let G:=G{v,w1,w2,w3,w4}G^{\prime}:=G-\{v,w_{1},w_{2},w_{3},w_{4}\}. By minimality, GG^{\prime} has an odd 5-coloring φ\varphi. To extend φ\varphi to GG, color vv to avoid {φ(x1),φ(x2),φ(x3)\{\varphi(x_{1}),\varphi(x_{2}),\varphi(x_{3}), φ(x4)}\varphi(x_{4})\} and then color each wiw_{i}, in succession, to avoid the new color on vv and also avoid {φ(xi)\{\varphi(x_{i}), φo(xi),φo(v)}\varphi_{o}(x_{i}),\varphi_{o}(v)\}. Thus, no 4-vertex in GG has four 2-neighbors. Similarly, if each wiw_{i} has degree at most 3 and at least one wiw_{i} is a 2-vertex, then we form GG^{\prime} from GG by deleting vv and all its 2-neighbors. Again GG^{\prime} has an odd 5-coloring φ\varphi. We extend φ\varphi to vv by avoiding the color on each 3-neighbor and the color on the other neighbor of each 2-neighbor. Finally, we can extend the coloring to all 2-neighbors to get an odd 5-coloring of GG. Thus, GG has no 4-vertex with at least one 2-neighbor and all 33^{-}-neighbors. Similarly, GG does not contain adjacent 4-vertices, vv and ww, each with three 2-neighbors. If so, then G(N[v]N[w])G-(N[v]\cup N[w]) has an odd 5-coloring. We color vv and ww with distinct colors that each differ from the colors on the three colored vertices at distance two in GG. Again, we can extend this coloring to an odd 5-coloring of GG.

Now we use discharging to reach a contradiction, which will finish the proof. We give each vertex vv initial charge d(v)d(v) and use the following two discharging rules.

  1. (R1)

    Each 2-vertex takes 3/73/7 from each neighbor.

  2. (R2)

    Each 3-vertex with a 2-neighbor and each 4-vertex with three 2-neighbors takes 1/71/7 from each 3+3^{+}-neighbor.

We show that each vertex finishes with charge at least 20/7, which contradicts the hypothesis mad(G)<20/7\mathrm{mad}(G)<20/7. Recall that δ(G)2\delta(G)\geq 2. If d(v)=2d(v)=2, then vv finishes with 2+2(3/7)=20/72+2(3/7)=20/7. If d(v)=3d(v)=3, then vv has at most one 2-neighbor. Further, vv does not give away charge by (R2), since no 3-vertex has both a 2-neighbor and a 3-neighbor, and also no 4-vertex has both three 2-neighbors and a 3-neighbor. So, if vv has a 2-neighbor, then it has two 4+4^{+}-neighbors and receives charge 1/71/7 from each. Thus, vv finishes with at least 33/7+2(1/7)=20/73-3/7+2(1/7)=20/7. If vv has no 2-neighbor, then vv starts and finishes with 3.

Let vv be a 4-vertex, and recall that vv has at most three 2-neighbors. If vv has at most two 2-neighbors, then vv finishes with at least 42(3/7)2(1/7)=20/74-2(3/7)-2(1/7)=20/7. If vv has three 2-neighbors, then its fourth neighbor is a 4+4^{+}-neighbor that does not receive charge from vv by (R2) but rather gives vv charge 1/71/7 by (R2). So vv finishes with 43(3/7)+1/7=20/74-3(3/7)+1/7=20/7.

If vv is a 5+5^{+}-vertex, then vv finishes with at least d(v)3d(v)/7=4d(v)/720/7d(v)-3d(v)/7=4d(v)/7\geq 20/7, since d(v)5d(v)\geq 5. This finishes the proof. ∎

With a bit more analysis, we can prove the following stronger result. The proof is similar to that of Corollary 9, so we just provide a sketch.

Corollary 10.

If mad(G)20/7\mathrm{mad}(G)\leq 20/7 and GG does not contain K6K^{*}_{6} as a subgraph, then χo(G)5\chi_{o}(G)\leq 5.

Proof Sketch..

Assume GG is a counterexample. So GG has no 6+6^{+}-vertices and each 5-vertex in GG has five 2-neighbors. Each 3-vertex has a single 2-neighbor. Each 4-vertex either has exactly three 2-neighbors or has exactly two 2-neighbors and gives charge 1/7 to each of it 3+3^{+}-neighbors. In this case, we delete vv, its 4-neighbors and the 2-neighbors of all deleted 4-vertices. This smaller graph has an odd 5-coloring φ\varphi, and it is straightforward to check that we can extend φ\varphi to GG. So GG has no 4-vertices. This implies, in turn, that GG has no 3-vertices, since GG has no 3-vertex with two 33^{-}-neighbors. So GG is bipartite with vertices in one part of degree 2 and those in the other part of degree 5. We contract one edge incident to each 2-vertex, and 5-color the resulting graph by Brooks’ Theorem. Finally, we extend this 5-coloring to an odd 5-coloring of GG. ∎

References

  • [1] D. P. Bunde, K. Milans, D. B. West, and H. Wu. Parity and strong parity edge-coloring of graphs. In Proceedings of the Thirty-Eighth Southeastern International Conference on Combinatorics, Graph Theory and Computing, volume 187, pages 193–213, 2007.
  • [2] D. P. Bunde, K. Milans, D. B. West, and H. Wu. Optimal strong parity edge-coloring of complete graphs. Combinatorica, 28(6):625–632, 2008.
  • [3] P. Cheilaris. Conflict-Free Coloring. 2009. Thesis (Ph.D.)–City University of New York.
  • [4] P. Cheilaris, B. Keszegh, and D. Pálvölgyi. Unique-maximum and conflict-free coloring for hypergraphs and tree graphs. SIAM J. Discrete Math., 27(4):1775–1787, 2013.
  • [5] P. Cheilaris and G. Tóth. Graph unique-maximum and conflict-free colorings. J. Discrete Algorithms, 9(3):241–251, 2011.
  • [6] G. Even, Z. Lotker, D. Ron, and S. Smorodinsky. Conflict-free colorings of simple geometric regions with applications to frequency assignment in cellular networks. SIAM J. Comput., 33(1):94–136, 2003.
  • [7] I. Fabrici and F. Göring. Unique-maximum coloring of plane graphs. Discuss. Math. Graph Theory, 36(1):95–102, 2016.
  • [8] R. Glebov, T. Szabó, and G. Tardos. Conflict-free colouring of graphs. Combin. Probab. Comput., 23(3):434–448, 2014.
  • [9] A. Kostochka, M. Kumbhat, and T. Łuczak. Conflict-free colourings of uniform hypergraphs with few edges. Combin. Probab. Comput., 21(4):611–622, 2012.
  • [10] A. Kostochka and M. Yancey. Ore’s conjecture on color-critical graphs is almost true. J. Combin. Theory Ser. B, 109:73–101, 2014.
  • [11] A. Kostochka and M. Yancey. A Brooks-type result for sparse critical graphs. Combinatorica, 38(4):887–934, 2018.
  • [12] J. Pach and G. Tardos. Conflict-free colourings of graphs and hypergraphs. Combin. Probab. Comput., 18(5):819–834, 2009.
  • [13] M. Petruševski and R. Škrekovski. Colorings with neighborhood parity condition. Dec. 2021, arXiv:2112.13710.
  • [14] S. Smorodinsky. Conflict-free coloring and its applications. In Geometry---intuitive, discrete, and convex, volume 24 of Bolyai Soc. Math. Stud., pages 331--389. János Bolyai Math. Soc., Budapest, 2013.