This paper investigates the star chromatic index of subcubic multigraphs, proving NP-completeness for certain cases, establishing structural properties, and confirming the conjecture that such graphs are star 6-edge-colorable under specific maximum average degree conditions.
Contribution
It proves NP-completeness for determining star 3-edge-colorability, develops structural results for subcubic multigraphs, and confirms the star 6-edge-colorability conjecture under certain maximum average degree bounds.
Findings
01
NP-complete to decide if '_s(G) for any graph G
02
Structural properties of subcubic multigraphs with '_s(G) > k
03
Subcubic multigraphs with mad(G)<5/2 are star 6-edge-colorable
Abstract
The star chromatic index of a multigraph G, denoted χs′(G), is the minimum number of colors needed to properly color the edges of G such that no path or cycle of length four is bi-colored. A multigraph G is star k-edge-colorable if χs′(G)≤k. Dvo\v{r}\'ak, Mohar and \v{S}\'amal [Star chromatic index, J. Graph Theory 72 (2013), 313--326] proved that every subcubic multigraph is star 7-edge-colorable. They conjectured in the same paper that every subcubic multigraph should be star 6-edge-colorable. In this paper, we first prove that it is NP-complete to determine whether χs′(G)≤3 for an arbitrary graph G. This answers a question of Mohar. We then establish some structure results on subcubic multigraphs G with δ(G)≤2 such that χs′(G)>k but χs′(G−v)≤k for any v∈V(G), where k∈{5,6}. We finally apply the structure…
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsAdvanced Graph Theory Research · Limits and Structures in Graph Theory · Graph Labeling and Dimension Problems
Full text
Star chromatic index of subcubic multigraphs
Hui Lei11footnotemark: 1 and Yongtang Shi
Center for Combinatorics and LPMC
Nankai University
Tianjin 300071, China
Zi-Xia Song
Department of Mathematics
University of Central Florida
Orlando, FL 32816, USA
Partially supported by the National
Natural Science Foundation of China and Natural Science Foundation of Tianjin (No.
17JCQNJC00300).Corresponding author. Email: [email protected]
Abstract
The star chromatic index of a multigraph G, denoted
χs′(G), is the minimum number of colors needed to properly
color the edges of G such that no path or cycle of length four is
bi-colored. A multigraph G is star k-edge-colorable if
χs′(G)≤k. Dvořák, Mohar and Šámal [Star
chromatic index, J. Graph Theory72 (2013), 313–326]
proved that every subcubic multigraph is star 7-edge-colorable.
They conjectured in the same paper that every subcubic multigraph
should be star 6-edge-colorable. In this paper, we first prove that
it is NP-complete to determine whether χs′(G)≤3 for an arbitrary graph G.
This answers a question of Mohar. We then establish some
structure results on subcubic multigraphs G with δ(G)≤2 such that χs′(G)>k
but χs′(G−v)≤k for any v∈V(G), where k∈{5,6}.
We finally apply the structure results, along with a simple
discharging method, to prove that every subcubic multigraph G is
star 6-edge-colorable if mad(G)<5/2, and star 5-edge-colorable
if mad(G)<24/11, respectively, where mad(G) is
the maximum average degree of a multigraph G. This partially confirms the conjecture of Dvořák, Mohar and Šámal.
Keywords: star edge-coloring; subcubic multigraphs; maximum average degree
1 Introduction
All multigraphs in this paper are finite and loopless; and all graphs are finite and without loops or multiple edges.
Given a multigraph G, let c:E(G)→[k] be a proper
edge-coloring of G, where k≥1 is an integer and [k]:={1,2,…,k}. We say that
c is a star k-edge-coloring of G if no path or cycle of
length four in G is bi-colored under the coloring c; and G is
star k-edge-colorable if G admits a star
k-edge-coloring. The star chromatic index of G, denoted
by χs′(G), is the smallest integer k such that G is
star k-edge-colorable. The chromatic index of G is denoted by
χ′(G). As pointed out in [6], the definition of
star edge-coloring of a graph G is equivalent to the star
vertex-coloring of its line graph L(G). Star edge-coloring of a
graph was initiated by Liu and Deng [10], motivated by the
vertex version (see [1, 4, 5, 9, 11]). Given a multigraph G, we use ∣G∣ to denote the number of vertices, e(G) the number of edges, δ(G) the minimum degree, and Δ(G) the maximum degree of G, respectively. For any v∈V(G), let dG(v) and NG(v) denote the degree and neighborhood of v in G, respectively. For any subsets A,B⊆V(G), let NG(A):=⋃a∈ANG(a), and let A\B:=A−B. If B={b}, we simply write A\b instead of A\B. We use Kn and Pn to denote the complete graph and the path on n vertices, respectively.
It is well-known [13] that the chromatic index of a graph with maximum degree Δ is either Δ or Δ+1. However, it is NP-complete [7] to determine whether the chromatic index of an
arbitrary graph with maximum degree Δ is Δ or Δ+1. The problem remains NP-complete even for cubic graphs. A multigraph G is subcubic if the maximum degree
of G is at most three. Mohar (private communication with the second author) proposed that it is NP-complete to determine whether χs′(G)≤3 for an arbitrary graph G. We first answer this question in the positive.
Theorem 1.1
It is NP-complete to determine whether χs′(G)≤3 for an arbitrary graph G.
We prove Theorem 1.1 in Section 2. Theorem 1.2 below is a result of Dvořák, Mohar and Šámal [6], which gives an upper bound and a lower bound for complete graphs.
The star chromatic index of the complete graph Kn satisfies
[TABLE]
In particular, for every ϵ>0, there exists a constant c such that χs′(Kn)≤cn1+ϵ for every integer n≥1.
The true order of magnitude of χs′(Kn) is still unknown. From Theorem 1.2, an upper bound in terms of the maximum degree for general graphs is also derived in [6], i.e.,
χs′(G)≤Δ⋅2O(1)logΔ for any graph G with maximum degree Δ. In the same paper, Dvořák, Mohar and Šámal [6] also considered the star chromatic index of subcubic multigraphs. To state their result, we need to introduce one notation. A graph G covers a graph H if there is a mapping f:V(G)→V(H) such that for any uv∈E(G), f(u)f(v)∈E(H), and for any u∈V(G), f is a bijection between NG(u) and NH(f(u)). They proved the following.
*(b) *If G is cubic and has no multiple edges, then χs′(G)≥4 and the equality holds if and only if G covers the graph of 3-cube.
As observed in [6], χs′(K3,3)=6 and the Heawood graph is star 6-edge-colorable. No subcubic multigraphs with star chromatic index seven are known. Dvořák, Mohar and Šámal [6] proposed the following conjecture.
Conjecture 1.4
Let G be a subcubic multigraph. Then χs′(G)≤6.
As far as we know, not much progress has been made yet towards Conjecture 1.4. It was recently shown in [2] that every subcubic outerplanar graph is star 5-edge-colorable.
A tight upper bound for trees was also obtained in [2]. We summarize the main results in [2] as follows.
(a) χs′(G)≤⌊23Δ(G)⌋ if G is a tree. Moreover, the bound is tight.
(b) χs′(G)≤5 if Δ(G)≤3.
(c) χs′(G)≤⌊23Δ(G)⌋+12 if Δ(G)≥4.
The maximum average degree of a multigraph G, denoted mad(G), is defined as the maximum of 2e(H)/∣H∣ taken over all the subgraphs H of G.
We want to point out here that there is an error in the proof of
Theorem 2.3 in a recent published paper by Pradeep and Vijayalakshmi
[Star chromatic index of subcubic graphs, Electronic Notes in Discrete
Mathematics53 (2016), 155–164]. Theorem 2.3 in [12] claims that if
G is a subcubic graph with mad(G)<11/5, then χs′(G)≤5.
The error in the proof of
Theorem 2.3 arises from ambiguity in the
statement of Claim 3 in their paper. From its proof given in [12] (on page 158),
Claim 3 should be stated as “H does not
contain a path uvw, where either all of u,v,w are 2-vertices or all of u,v,w are light 3-vertices". This new statement of Claim 3
does not imply that “a 2-vertex must be adjacent to a heavy
3-vertex" in Case 2 of the proof of Theorem 2.3 (on page 162). It seems nontrivial to fix this error in their proof. If Claim 3 in their paper is true, using the technique we developed in the proof of Theorem 1.6(b), one can obtain a stronger result that every subcubic multigraph with mad(G)<7/3 is star 5-edge-colorable.
In this paper, we prove two main results, namely Theorem 1.1 mentioned above and Theorem 1.6 below.
Theorem 1.6
Let G be a subcubic multigraph.
*(a) *If mad(G)<2, then χs′(G)≤4 and the bound is tight.
*(b) *If mad(G)<24/11, then χs′(G)≤5.
*(c) *If mad(G)<5/2, then χs′(G)≤6.
The rest of this paper is organized as follows. We prove Theorem 1.1 in Section 2. Before we prove Theorem 1.6 in Section 4, we establish in Section 3 some structure results on subcubic multigraphs G with δ(G)≤2 such that χs′(G)>k and χs′(G−v)≤k for any v∈V(G), where k∈{5,6}. We believe that our structure results can be used to solve Conjecture 1.4.
First let us denote by SEC the problem stated in Theorem 1.1, and we denote by 3EC the following well-known NP-complete problem of Holyer [7]:
Given a cubic graph G, is G3-edge-colorable?
Proof of Theorem 1.1: Clearly, SEC is in the class NP. We shall reduce 3EC to SEC.
Let H be an instance of 3EC. We construct a graph G from H by replacing each edge e=uw∈E(H) with a copy of graph Hab, identifying u with a and w with b, where Hab is depicted in Figure 1.
The size of G is clearly polynomial in the size of H, and Δ(G)=3.
It suffices to show that χ′(H)≤3 if and only if χs′(G)≤3. Assume that χ′(H)≤3. Let c:E(H)→{1,2,3}
be a proper 3-edge-coloring of H. Let c∗ be an edge coloring of G obtained from c as follows:
for each edge e=uw∈E(H), let c∗(av1e)=c∗(v3ev4e)=c∗(v6eb)=c(uw),
c∗(v1ev2e)=c∗(v3ev7e)=c∗(v4ev8e)=c∗(v5ev6e)=c(uw)+1,
and c∗(v2ev3e)=c∗(v4ev5e)=c(uw)+2, where all colors here and henceforth are done modulo 3.
Notice that c∗ is a proper 3-edge-coloring of G. Furthermore, it can be easily checked that
G has no bi-colored path or cycle of length four under the coloring c∗. Thus c∗ is a star 3-edge-coloring of G and so χs′(G)≤3.
Conversely, assume that χs′(G)≤3. Let c∗:E(G)→{1,2,3}
be a star 3-edge-coloring of G. Let c be an edge-coloring of H obtained from c∗ by letting c(e)=c∗(av1e) for any e=uw∈E(H). Clearly, c is a proper 3-edge-coloring of H if for any edge e=uw in G, c∗(av1e)=c∗(v6eb). We prove this next. Let e=uw be an edge of H. We consider the following two cases.
Case 1:c∗(v3ev7e)=c∗(v4ev8e).
In this case, let c∗(v3ev7e)=c∗(v4ev8e)=α, where α∈{1,2,3}. We may further assume that c∗(v3ev4e)=β and c∗(v2ev3e)=c∗(v4ev5e)=γ, where {β,γ}={1,2,3}\α. This is possible because dG∗(v3e)=dG∗(v4e)=3 and c∗ is a proper 3-edge-coloring of G∗. Since c∗ is a star edge-coloring of G∗, we see that c∗(v1ev2e)=c∗(v5ev6e)=α and so c∗(av1e)=c∗(v6eb)=β.
Case 2:c∗(v3ev7e)=c∗(v4ev8e).
In this case, let c∗(v3ev7e)=α, c∗(v4ev8e)=β, c∗(v3ev4e)=γ, where {α,β,γ}={1,2,3}. This is possible because α=β by assumption.
Since c∗ is a proper edge-coloring of G∗, we see that c∗(v2ev3e)=β and c∗(v4ev5e)=α. One can easily check now that c∗(v1ev2e)=α and c∗(v5ev6e)=β, and so c∗(av1e)=c∗(v6eb)=γ, because c∗ is a star edge-coloring of G∗.
In both cases we see that c∗(av1e)=c∗(v6eb). Therefore c is a proper 3-edge-coloring of H and so χ′(H)≤3. This completes the proof of Theorem 1.1.
3 Properties of star k-critical subcubic multigraphs
In this section, we establish some structure results on subcubic multigraphs G with δ(G)≤2 such that χs′(G)>k and χs′(G−v)≤k for any v∈V(G), where k∈{5,6}. For simplicity, we say that a multigraph G is star k-critical if χs′(G)>k and χs′(G−v)≤k for any v∈V(G), where k∈{5,6}.
Clearly, every star k-critical graph must be connected.
Throughout the remainder of this section, let G be a star k-critical subcubic multigraph with δ(G)≤2, and let N(v) and d(v) denote the neighborhood and degree of a vertex v in G, respectively. Since every multigraph with maximum degree two is star 4-edge-colorable, we see that Δ(G)=3 and ∣G∣≥3. Let x∈V(G) with d(x)≤2. Let H=G−x and let c:E(H)→[k] be a star k-edge-coloring of H, where k∈{5,6}.
For any u∈V(H), let c(u) denote the set of all colors such that each is used to color an edge incident with u under the coloring c.
For any A⊆V(H), let c(A):=⋃a∈Ac(a). By abusing the notation we use c(uv) to denote the set of all colors on the edges between u and v under the coloring c if uv∈E(H) is a multiple edge.
Observation If d(x)=2, then ∣N(x)∣=2.
Proof. Suppose that ∣N(x)∣=1. Let N(x)={z}. Since G is connected, we see that d(z)=3. Let N(z)={x,z∗}. We obtain a star k-edge-coloring of G by coloring the two edges between x and z by two distinct colors in [k]\c(z∗), a contradiction.
Lemma 3.1
Assume that d(x)=1. Let N(x)={y}. The following are true.
(a) c(NH(y))=[k] and ∣N(y)∣=3.
(b) N(y) is an independent set in G, d(y1)=3 and d(y2)≥k−3, where N(y)={x,y1,y2} with d(y1)≥d(y2).
*(c) *If d(y2)=k−3, then for any i∈{1,2} and any v∈NH(yi)\y, c(yyi)∈c(v), ∣N(v)∣≥2, ∣N(y1)∣=3, ∣N(y2)∣=k−3, and N[y1]∩N[y2]={y}.
*(d) *If d(y2)=2, then k=5 and d(w1)=3, where w1 is the other neighbor of y2 in G.
*(e) *If k=6 and for some i∈{1,2}, N(yi)\y has a vertex v with d(v)=2, then vv′∈/E(G), N(v′) is an independent set in G, and d(u)=3 for any u∈N(v)∪N[v′], where N(yi)={y,v,v′}.
*(f) *If k=6 and for some i∈{1,2}, each vertex of N(yi)\y has degree three in G, then either d(v)≥2 for any v∈N(yi1) or d(v)≥2 for any v∈N(yi2), where N(yi)={y,yi1,yi2}.
*(g) *If k=5 and d(y2)=3, then either d(v)≥2 for any v∈N(y1) or d(v)≥2 for any v∈N(y2).
Proof. To prove Lemma 3.1(a), suppose that c(NH(y))=[k]. Then coloring the edge xy by a color in [k]\c(NH(y)), we obtain a star k-edge-coloring of G, a contradiction. Thus c(NH(y))=[k] and so ∣N(y)∣=3.
Next let N(y)={x,y1,y2} with d(y1)≥d(y2) by Lemma 3.1(a). Suppose that y1y2∈E(G). Then ∣N(y1)∣=∣N(y2)∣=3 and all the edges incident to y1 or y2 are colored with distinct colors because c(NH(y))=[k]. Now coloring the edge xy by color c(y1y2) yields a star k-edge-coloring of G, a contradiction. Thus y1y2∈/E(G). Since ∣c(NH(y))∣=k≥5, we see that d(y1)=3 and d(y2)≥k−3≥2. This proves Lemma 3.1(b).
To prove Lemma 3.1(c), since d(y2)=k−3, we see that all the edges incident to y1 or y2 are colored with distinct colors because c(NH(y))=[k]. Suppose that for some i∈{1,2}, there exists a vertex v∈NH(yi)\y such that c(yyi)∈/c(v). Then we obtain a star k-edge-coloring of G by coloring the edge xy by a color in c(yiv), a contradiction. Thus for any i∈{1,2} and any v∈NH(yi)\y, c(yyi)∈c(v). Hence ∣N(v)∣≥2. Since Δ(G)=3, we see that N[y1]∩N[y2]={y}. We next show that
∣N(y1)∣=3 and ∣N(y2)∣=k−3.
Suppose that ∣N(y1)∣<3. Then ∣N(y1)∣=2 because d(y1)=3. Let N(y1)={y,y1∗}. Then d(y1∗)=3 by Observation. Let u be the other neighbor of y1∗ in G.
Then c(y1∗u)=c(yy1). Let α,β be two distinct colors on the parallel edges y1y1∗. Then α,β∈/c(u) because c is a star k-edge-coloring of H.
Let e∗ be the edge between y1 and y1∗ with color α. If c(y2)\c(u)=∅, then we obtain a star k-edge-coloring of G by recoloring the edge e∗ by a color in c(y2)\c(u), yy1 by color α, and coloring the edge xy by color c(yy1), a contradiction. Thus c(y2)⊂c(u) and so k=5. Clearly, ∣N(y2)∣=k−3=2. Let y2∗ be the other neighbor of y2 in G. Then c(yy2)∈c(y2∗). We obtain a star 5-edge-coloring of G by recoloring the edge yy2 by a color in {α,β}\c(y2∗), yy1 by color c(yy2), and coloring the edge xy by color c(yy1), a contradiction. Thus ∣N(y1)∣=3. By symmetry, ∣N(y2)∣=3 if k=6. Clearly, ∣N(y2)∣=k−3=2 if k=5. This proves Lemma 3.1(c).
It remains to prove Lemma 3.1(d), (e), (f) and (g). Notice that if d(y2)=2, then by Lemma 3.1(b), d(y2)=2≥k−3. Thus k=5 and so d(y2)=k−3. For each proof of Lemma 3.1(d), (e), and (f), let d(y2)=k−3. By Lemma 3.1(c), we may assume that N(y1)={y,z1,z2} and N(y2)={y,w1} if k=5 (and N(y2)={y,w1,w2} if k=6), where N[y1]∩N[y2]={y}. By Lemma 3.1(a), we may further assume that c(yy1)=1, c(yy2)=2, c(y1z1)=3, c(y1z2)=4, c(y2w1)=5 (and c(y2w2)=6 when k=6).
By Lemma 3.1(c), 1∈c(z1)∩c(z2) and 2∈c(w1) if k=5 (and 2∈c(w1)∩c(w2) if k=6).
We next prove Lemma 3.1(d). Clearly, k=5. Suppose that d(w1)≤2. By Lemma 3.1(c), d(w1)=2 and c(w1)={2,5}. Let w∗∈N(w1) with c(w1w∗)=2. Notice that w∗ is not necessarily different from z1 or z2. Since c is a star edge-coloring of H, 5∈/c(w∗). If 3∈/c(w∗), then we obtain a star 5-edge-coloring of G by coloring the edge xy by color 5 and recoloring the edge y2w1 by color 3, a contradiction. Thus 3∈c(w∗), and similarly, 4∈c(w∗). Hence w∗∈/{z1,z2} because Δ(G)=3. We obtain a star 5-edge-coloring of G by coloring the edge xy by color 2, recoloring the edge yy2 by color 5, and y2w1 by color 1, a contradiction.
To prove Lemma 3.1(e), since k=6, we see that c(y1)={1,3,4} and c(y2)={2,5,6}. By symmetry, we may assume that i=1, v=z1, and v′=z2. Then 1∈c(z1)∩c(z2). Suppose that z1z2∈E(G). Then c(z1z2)=1. Now recoloring the edge y1z1 by a color in {5,6}\c(z2) and coloring the edge xy by color 3, we obtain a star 6-edge-coloring of G, a contradiction. Thus z1z2∈/E(G). Let N(z1)={y1,z1∗}. Then c(z1z1∗)=1. Since z1z2∈/E(G), we see that z1∗=z2. If {2,5,6}\(c(z1∗)∪c(z2))=∅, then recoloring the edge y1z1 by a color in {2,5,6}\(c(z1∗)∪c(z2)), y1y by color 3, and coloring xy by color 1 yields a star 6-edge-coloring of G, a contradiction. Thus {2,5,6}⊂c(z1∗)∪c(z2) and so d(z1∗)=d(z2)=3. Clearly, z1∗z2∈/E(G) because Δ(G)=3. Let c(z2)={1,4,α}, where α∈{2,5,6}.
Suppose that c(N(z2)\y1)=[6]. Then we obtain a star 6-edge coloring of G by recoloring the edge y1z2 by a color, say β, in [6]\c(N(z2)\y1), y1z1 by color α, yy1 by color 3 if β=3 or color 4 if β=3,
and finally coloring xy by color 1, a contradiction.
Thus c(N(z2)\y1)=[6] and so ∣N(z2)∣=3. Let N(z2)={y1,z21,z22}. Clearly, z21z22∈/E(G) because c(z21)∪c(z22)=[6], and d(z21)=d(z22)=3, as desired. This proves Lemma 3.1(e).
To prove Lemma 3.1(f), we may assume that i=1, z1=y11 and z2=y12. Suppose that there exist vertices z11∈N(z1) and z21∈N(z2) such that d(z11)=d(z21)=1. Then ∣N(z1)∣=∣N(z2)∣=3 by Lemma 3.1(a). Let N(z1)={y1,z11,z12} and N(z2)={y1,z21,z22}. Then d(z12)=d(z22)=3 by Lemma 3.1(a). Assume first that c(z1z11)=1. If {2,5,6}\c(z12)=∅, then recoloring the edge z1z11 by a color in {2,5,6}\c(z12), we obtain a star 6-edge-coloring of H with 1∈/c(z1), contrary to Lemma 3.1(c). Thus c(z12)={2,5,6}. Clearly, 3∈c(z2), otherwise recoloring the edge z1z11 by color 4 yields a star 6-edge coloring of H with 1∈/c(z1), contrary to Lemma 3.1(c). Thus c(z2)={1,3,4}. Then
{2,5,6}\c(z22)=∅. Now recoloring the edge z2z21 by a color in {2,5,6}\c(z22), we obtain a star 6-edge-coloring of H with c(z2)={1,3,4},
a contradiction. Thus c(z1z11)=1. By symmetry, c(z2z21)=1. By Lemma 3.1(c), c(z1z12)=c(z2z22)=1. Clearly, 3∈/c(z12)
because H has no bi-colored path of length four. If {5,6}\c(z12)=∅, then we obtain a star 6-edge-coloring of G by recoloring the edge z1z11 by color 3, z1y1 by a color in {5,6}\c(z12), and coloring xy by color 3, a contradiction. Thus c(z12)={1,5,6}. Similarly, c(z22)={1,5,6}. Now recoloring the edges z1z11 by color 3, z1y1 by color 4, z21z2 by color 4, y1z2 by color 2, yy1 by color 3, and finally coloring the edge xy by color 1, we obtain a star 6-edge-coloring of G, a contradiction. This proves Lemma 3.1(f).
It remains to prove Lemma 3.1(g). Suppose that there exist vertices y11∈N(y1) and y21∈N(y2) such that d(y11)=d(y21)=1. Then ∣N(y1)∣=∣N(y2)∣=3 by Lemma 3.1(a). Let N(y1)={y,y11,y12} and N(y2)={y,y21,y22}. By Lemma 3.1 (a), c(y1)∪c(y2)=[5]. If c(y1y11)=c(y2y21), then we obtain a star 5-edge-coloring of G by coloring the edge xy by color c(y1y11) because c(y1)∪c(y2)=[5], a contradiction. Thus c(y1y11)=c(y2y21). Since c(y1)∪c(y2)=[5], we see that either c(y1y11)∈/c(y2) or c(y2y21)∈/c(y1). We may assume that c(y1y11)∈/c(y2). But then coloring the edge xy by color c(y1y11) yields a star 5-edge-coloring of G, a contradiction.
Assume that d(x)=2. Let N(x)={z,w} with ∣N(z)∣≤∣N(w)∣.
*(a) *If zw∈E(G), then k=5, ∣N(z)∣=∣N(w)∣=3 and d(v)≥2 for any v∈N(z)∪N(w).
*(b) *If zw∈/E(G), then ∣N(w)∣=3 or k=5, ∣N(w)∣=∣N(z)∣=2, and d(w)=d(z)=3.
*(c) *If d(z)=2 and z∗w∈E(G), then k=5, ∣N(z∗)∣=∣N(w)∣=3, and d(u)=3 for any u∈(N[w]∪N[z∗])\{x,z}, where z∗ is the other neighbor of z in G.
*(d) *If k=6 and d(z)=2, then z∗w∈/E(G), ∣N(z∗)∣=∣N(w)∣=3, and for any v∈(N(w)∪N(z∗))\{x,z}, d(v)=3 and d(u)≥2 for any u∈N(v), where N(z)={x,z∗}.
*(e) *If k=5 and d(z)=2, then ∣N(z∗)∣=∣N(w)∣=3, and ∣N(v)∣≥2 for any v∈N(w)∪N(z∗), where N(z)={x,z∗}.
Proof. Assume that zw∈E(G). Since G is connected, we see that ∣N(w)∣=3. Let N(w)={x,z,w∗}. We first show that ∣N(z)∣=3 and N(z)∩N(w)={x}. Suppose that
∣N(z)∣=2 or ∣N(z)∣=3 and zw∗∈E(G). Then ∣c(w)∪c(w∗)∣≤4 when c(zw)∈/c(w∗) and ∣c(w)∪c(w∗)∣≤3 when c(zw)∈c(w∗). We obtain a star k-edge-coloring of G by coloring the edge xw by a color, say α, in [k]\(c(w)∪c(w∗)) and then coloring xz by color c(ww∗) if c(zw)∈/c(w∗) or a color in [k]\(c(w)∪c(w∗)∪{α}) if c(zw)∈c(w∗), a contradiction. Thus ∣N(z)∣=3 and z∗=w∗, where N(z)={x,w,z∗}. We next show that k=5. Suppose that k=6. Then c(zw)∈c(z∗), otherwise we obtain a star 6-edge-coloring of G by coloring the edge xz by a color, say α, in [6]\(c(z∗)∪c(w)) and xw by a color in [6]\(c(w)∪c(w∗)∪{α}). We then obtain a star 6-edge-coloring of G by coloring the edge xw by a color, say β, in [6]\(c(w∗)∪c(z)) and xz by a color in [6]\(c(z∗)∪{c(ww∗),β}), a contradiction. Hence k=5. By Lemma 3.1(b), we see that d(v)≥2 for any v∈N(z)∪N(w). This proves Lemma 3.2(a).
To prove Lemma 3.2(b), by Lemma 3.1(a), ∣N(w)∣≥∣N(z)∣≥2. We are done if ∣N(w)∣=3. So we may assume that ∣N(w)∣=2. Then ∣N(z)∣=2 because ∣N(z)∣≤∣N(w)∣. Let z∗ and w∗ be the other neighbor of z and w, respectively. If ww∗ or zz∗ is not a multiple edge (say the former) or k=6, then we obtain a star k-edge-coloring of G by coloring the edge xz by a color, say α, in [k]\(c(z∗)∪{c(ww∗)}) and xw by a color in [k]\(c(w∗)∪{α}), a contradiction. Thus ∣c(ww∗)∣=∣c(zz∗)∣=2 and k=5. We see that d(w)=d(z)=3.
We next prove Lemma 3.2(c). Since d(z)=2, we see that zw∈/E(G) by Lemma 3.2(a). Then ∣N(w)∣=3 by Lemma 3.2(b). Since xz∗∈/E(G), by Lemma 3.2(b) again, ∣N(z∗)∣=3. Let N(w)={x,z∗,w∗} and N(z∗)={z,z1∗,w}.
We first show that k=5. Suppose that k=6.
Then c can be extended to be a star 6-edge-coloring of G by coloring the edge xw by color c(zz∗) if c(zz∗)∈/c(w∗) or a color α in [6]\(c(z∗)∪c(w∗)) if c(zz∗)∈c(w∗), and then coloring the edge xz by a color in [6]\(c(z∗)∪{α,c(ww∗)}), a contradiction. Thus k=5. We next show that d(w∗)=3. Suppose that d(w∗)≤2. If c(zz∗)∈/c(w∗), then we obtain a star 5-edge-coloring of G by coloring the edge xw by color c(zz∗) and xz by a color in [5]\(c(z∗)∪c(w)), a contradiction. Thus c(zz∗)∈c(w∗) and so ∣c(w∗)∪c(z∗)∣≤4. We obtain a star 5-edge coloring of G by coloring the edge xw by a color, say α, in [5]\(c(z∗)∪c(w∗)) and xz by a color in [5]\(c(z∗)∪{α}), a contradiction. By symmetry, d(z1∗)=3. This proves Lemma 3.2(c).
To prove Lemma 3.2(d), since k=6, by Lemma 3.2(a,c), we see that wz,wz∗∈/E(G). By Lemma 3.2(b), ∣N(z∗)∣=∣N(w)∣=3. Let N(w)={x,w1,w2} and N(z∗)={z,z1∗,z2∗}. Then c(zz∗)∈c(w1)∪c(w2), otherwise we obtain a star 6-edge-coloring of G by coloring the edge xw by color c(zz∗) and xz by a color in [6]\(c(z∗)∪c(w)), a contradiction. We next show that c(w1)∪c(w2)=[6]. Suppose that c(w1)∪c(w2)=[6]. Now coloring the edge xw by a color, say α, in [6]\(c(w1)∪c(w2)), and then coloring xz by color c(ww1) if c(ww1)∈/c(z∗) or a color in [6]\(c(z∗)∪c(w)∪{α}) if c(ww1)∈c(z∗), we obtain a star 6-edge-coloring of G, a contradiction. Thus c(w1)∪c(w2)=[6] and so d(w1)=d(w2)=3. By symmetry, d(z1∗)=d(z2∗)=3. Finally, for any u∈N(z1∗)∪N(z2∗)∪N(w1)∪N(w2), by Lemma 3.1(e), d(u)≥2. This proves Lemma 3.2(d).
It remains to prove Lemma 3.2(e). By Lemma 3.2(a), zw∈/E(G). We may assume that wz∗∈/E(G) by Lemma 3.2(c). By Lemma 3.2(b), ∣N(z∗)∣=∣N(w)∣=3. Suppose that there exists a vertex v∈N(w)∪N(z∗) with d(v)=1. We may assume that v∈N(w). Let N(w)={x,v,w∗}. Then c(zz∗)∈c(v)∪c(w∗), otherwise we recolor the edge wv by color c(zz∗). Now we obtain a star 5-edge-coloring of G by coloring the edge xw by a color, say α, in [5]\(c(v)∪c(w∗)) and xz by a color in [5]\(c(z∗)∪{α}), a contradiction. Thus ∣N(v)∣,∣N(w∗)∣≥2. By symmetry, ∣N(u)∣≥2 for any u∈N(z∗).
Let G be a subcubic multigraph that is star 6-critical. Let v∈V(G) be a vertex with N(v)={v1,v2,v3} and d(v1)≥d(v2)≥d(v3)=2. The following are true.
(a) v1,v2,v3 are pairwise distinct, d(v1)=3 and v3∗∈/{v1,v2}, where N(v3)={v,v3∗}.
*(b) *If d(v3∗)=2, then d(v2)=3, and d(u)≥2 for each u∈N(v1)∪N(v2),
*(c) *If d(v2)=2, then every vertex of N(v2)∪N(v3) has degree three in G, and for any u∈N(v1)\v, d(u)≥2.
*(d) *If d(v2)=3 and there exists a vertex vi∗∈N(vi) with d(vi∗)=1 for some i∈{1,2}, then for any u∈N(v3−i), d(u)=3.
Proof. To prove Corollary 3.3(a), we first show that v1,v2,v3 are pairwise distinct. By Observation, v3 is distinct from v1,v2. Suppose that v1=v2. Let v1∗ be the other neighbor of v1, where v1∗,v3∗ are not necessarily distinct. Let c:E(G\v)→[6] be a star edge-coloring of G\v. We obtain a star 6-edge-coloring of G by coloring the edges vv1 by two distinct colors, say α,β, in [6]\(c(v1∗)∪{c(v3v3∗)}) and vv3 by a color in [6]\(c(v3∗)∪{α,β}), a contradiction. This proves that v1,v2,v3 are pairwise distinct. By Lemma 3.2(a), v3∗∈/{v1,v2}. We next show that d(v1)=3. Suppose that d(v1)=3. Then d(vi)=2 for all i∈[3]. By Lemma 3.2(a), we see that v1v2∈/E(G). Let vi∗ be the other neighbor of vi for all i∈{1,2}, where v1∗,v2∗,v3∗ are not necessarily distinct. Let c:E(G\v)→[6] be a star edge-coloring of G\v. We obtain a star 6-edge-coloring of G by coloring the edge vv1 by a color, say α, in [6]\(c(v1∗)∪{c(v2v2∗),c(v3v3∗)}), vv2 by a color, say β, in [6]\(c(v2∗)∪{α,c(v3v3∗)}), and vv3 by a color in [6]\(c(v3∗)∪{α,β}), a contradiction. Thus d(v1)=3. This proves Corollary 3.3(a).
By Lemma 3.2(d), Corollary 3.3(b) is true. By Lemma 3.2(d) and Lemma 3.1(e), Corollary 3.3(c) is true. Finally, by Lemma 3.1(b,e) applied to vi∗, Corollary 3.3(d) is true.
To prove Theorem 1.6(a), let G be a subcubic multigraph with mad(G)<2. Then G must be a simple graph. Notice that a simple graph G has mad(G)<2
if and only if G is a forest. Now applying Theorem 1.5(a) to
every component of G, we see that χs′(G)≤4. This bound
is sharp in the sense that there exist graphs G with mad(G)=2 and χs′(G)>4. One such
example from [2] is depicted in Figure 2.
We next proceed the proof of Theorem 1.6(c) by contradiction. Suppose the assertion is false. Let G be a subcubic multigraph with mad(G)<5/2 and χs′(G)>6. Among all counterexamples we choose G so that
∣G∣ is minimum. By the choice of G, G is connected, star 6-critical, and mad(G)<5/2. For all i∈[3], let Ai={v∈V(G):dG(v)=i} and ni=∣Ai∣ for all i∈[3].
By Lemma 3.1(a), A1 is an independent set in G and NG(A1)⊆A3. Let G∗=G\A1. Then mad(G∗)<5/2. We see that 2e(G∗)=2e(G)−2n1=2n2+3n3−n1<5(n2+n3)/2 and so n3<n2+2n1. Thus A1∪A2=∅.
By Lemma 3.1(b), δ(G∗)≥2. We say that a vertex v∈V(G∗) with dG∗(v)=2 is good if dG(v)=3; bad if dG(v)=2 and v is adjacent to another vertex of degree two in G; and fair if dG(v)=2 and v is not bad. We shall apply the discharging method below to obtain a contradiction.
For each vertex v∈V(G∗), let ω(v):=dG∗(v)−25 be the initial charge of v. Then ∑v∈V(G∗)ω(v)=2e(G∗)−25∣G∗∣=(2n2+3n3−n1)−25(n2+n3)=(n3−n2−2n1)/2<0, because n3<n2+2n1. Notice that for each v∈V(G∗), ω(v)=2−5/2=−1/2 if dG∗(v)=2, and ω(v)=3−5/2=1/2 if dG∗(v)=3. Let x∈V(G∗) be a vertex with dG∗(x)=3 such that x is adjacent to exactly t≥1 vertices of degree two in G∗. We claim that t≤2 and t=1 when NG∗(x) has a bad vertex. Clearly, NG∗(x) has at most two good vertices by Lemma 3.1(f), and at most two fair vertices by Corollary 3.3(a). By Corollary 3.3(b), NG∗(x) has at most one bad vertex, and if such a bad vertex exists, then NG∗(x) has no good or fair vertex. Finally, NG∗(x) has at most one fair vertex and one good vertex simultaneously by Corollary 3.3(c,d). Thus t≤2 and t=1 when NG∗(x) has a bad vertex, as claimed. We will redistribute the charges of x according to the following discharging rule:
(R): For each x∈V(G∗) with dG∗(x)=3 and exactly t≥1 neighbors of degree two in G∗, x sends 2t1≥41 charges to each of its neighbors of degree two in G∗.
Let ω∗ be the new charge of G∗ after applying the above discharging rule. We see that for any v∈V(G∗) with dG∗(v)=3, ω∗(v)≥0. We next show that for any v∈V(G∗) with dG∗(v)=2,
ω∗(v)≥0.
Let v∈V(G∗) be a vertex with dG∗(v)=2. By Observation and Lemma 3.1(a), ∣NG∗(v)∣=2. Let NG∗(v)={u,w}. If v is good, then dG∗(u)=dG∗(w)=3 by Lemma 3.1(c). Thus ω∗(v)≥ω(v)+1/4+1/4=0.
Next, suppose that v is bad. We may assume that u is bad. By Lemma 3.2(d), dG∗(w)=3. By the above claim, v is the only (bad) vertex of degree two of NG∗(w).
By the discharging rule (R), ω∗(v)≥ω(v)+1/2=0. Finally, suppose that v is a fair vertex. Then dG(u)=dG(w)=3. By Lemma 3.1(b), neither u nor w is good. Thus dG∗(u)=dG∗(w)=3. By the discharging rule (R),
ω∗(v)≥ω(v)+1/4+1/4=0. This proves that ω∗(v)≥0 for any v∈V(G∗) with dG∗(v)=2. Thus ∑v∈V(G∗)ω∗(v)≥0, contrary to the fact that ∑v∈V(G∗)ω∗(v)=∑v∈V(G∗)ω(v)<0.
The proof of Theorem 1.6(b) is similar to the proof Theorem 1.6(c). For its completeness, we include its proof here because the discharging part is different and more involved. Suppose the assertion is false. Let G be a subcubic multigraph with mad(G)<24/11 and G is not star 5-edge-colorable. Among all counterexamples we choose G so that
∣G∣ is minimum. By the choice of G, G is connected and star 5-critical. Clearly, mad(G)<24/11. For all i∈[3], let Ai={v∈V(G):dG(v)=i} and let ni=∣Ai∣ for all i∈[3].
By Lemma 3.1(a), A1 is an independent set in G and NG(A1)⊆A3. Let G∗=G\A1. Then mad(G∗)<24/11. We see that 2e(G∗)=2e(G)−2n1=2n2+3n3−n1<24(n2+n3)/11 and so 9n3<2n2+11n1. Thus A1∪A2=∅.
By Lemma 3.1(b), δ(G∗)≥2. We say that a vertex v∈V(G∗) with dG∗(v)=2 is good if dG(v)=3; bad if dG(v)=2 and v is adjacent to another vertex of degree two in G; and fair if dG(v)=2 and v is not bad. Let B:={v∈V(G∗):dG∗(v)=2}. We next claim that every component of G∗[B] is isomorphic to K1, K2 or P3.
Suppose not. Let P be a longest path in G∗[B] with vertices x1,x2,…,xp in order, where p≥4 or p=3 and x1x3∈E(G) or p=2 and x1x2 is a multiple edge. Clearly, p=2 by Observation and Lemma 3.1 (a). Suppose that p=3. By Lemma 3.2(a), none of x1,x2,x3 is a fair or bad vertex. Thus all of x1,x2,x3 must be good, contrary to Lemma 3.1(b). Hence p≥4. By Lemma 3.2(e), none of the vertices of P are bad. If x2 is fair, then both x1 and x3 must be good because neither x1 nor x3 are bad. By Lemma 3.1(c) applied to x2, dG∗(x4)=3, a contradiction. Thus x2 is good. Similarly, x3 is good. By Lemma 3.1(g, c), x1 is neither good nor fair, and so x1 must be bad, a contradiction. Thus every component of G∗[B] is isomorphic to K1, K2 or P3.
We shall apply the discharging method to obtain a contradiction.
For each vertex v∈V(G∗), let ω(v):=dG∗(v)−1124 be the initial charge of v. Then ∑v∈V(G∗)ω(v)=2e(G∗)−1124∣G∗∣=(2n2+3n3−n1)−1124(n2+n3)=119n3−2n2−11n1<0, because 9n3<2n2+11n1. Notice that for each v∈V(G∗), ω(v)=2−24/11=−2/11 if dG∗(v)=2 and ω(v)=3−24/11=9/11 if dG∗(v)=3. We will redistribute the charges of vertices in G∗ according to the following discharging rule:
(R): For each x∈V(G∗) with dG∗(x)=3 and exactly t≥1 neighbors of degree two in G∗, x sends 11t9≥113 charges to each of its neighbors of degree two in G∗.
Let ω∗ be the new charge of G∗ after applying the above discharging rule. We see that for any v∈V(G∗) with dG∗(v)=3, ω∗(v)≥0. We next show that ω∗(B):=∑v∈Bω∗(v)≥0.
By the above claim, each component P of G∗[B] is isomorphic to K1, K2 or P3. Thus each endpoint of P (with the endpoint of P1 counted twice) receives at least 3/11 charge from its neighbor in V(G∗)\B and so ω∗(P):=∑v∈V(P)ω∗(v)≥∑v∈V(P)ω(v)+3/11+3/11≥−6/11+3/11+3/11=0. Hence ω∗(B)=∑P∈G∗[B]ω∗(P)≥0, where P∈G∗[B] denotes that P is a component of G∗[B]. We see that
∑v∈V(G∗)ω∗(v)=∑v∈V(G∗)\Bω∗(v)+ω∗(B)≥0, contrary to the fact that ∑v∈V(G∗)ω∗(v)=∑v∈V(G∗)ω(v)<0.
Remark. Kerdjoudj, Kostochka and Raspaud [8] considered the list version of star edge-colorings of simple graphs. They proved that every subcubic graph is star list-8-edge-colorable, and further proved the following.
(a) **If mad(G)<7/3, then G is star list-5-edge-colorable.
(b) **If mad(G)<5/2, then G is star list-6-edge-colorable.
Acknowledgments. The authors would like to thank one anonymous referee for many helpful comments, in particular, for pointing out that the Heawood graph is star 5-edge-colorable and K4 with one subdivided edge has star chromatic index 6, and bringing Reference [3] to our attention. Yongtang Shi would like to thank Bojan Mohar for his helpful discussion and for mentioning the complexity problem on star edge-coloring during his visit to Simon Fraser University. The authors would like to thank Rong Luo for his helpful comments.
Bibliography13
The reference list from the paper itself. Each links out to its DOI / PubMed record.
1[1] M. O. Albertson, G. G. Chappell, H. A. Kierstead, A. Kündgen and R. Ramamurthi, Coloring with no 2-colored P 4 subscript 𝑃 4 P_{4} ’s, Electron. J. Combin. 11 (2004), #R 26.
2[2] Ľ. Bezegová, B. Lužar, M. Mockovčiaková and R. Soták, R. Škrekovski, Star edge coloring of some classes of graphs, J Graph Theory 81 (2016) 73–82.
3[3] 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 Math. 206 (1999) 77–89.
4[4] Y. Bu, D. W. Cranston, M. Montassier, A. Raspaud and W. Wang, Star coloring of sparse graphs, J Graph Theory 62 (2009) 201–219.
5[5] M. Chen, A. Raspaud and W. Wang, 6 6 6 -star-coloring of subcubic graphs, J Graph Theory 72 (2013) 128–145.
6[6] Z. Dvořák, B. Mohar and R. Šámal, Star chromatic index, J Graph Theory 72 (2013) 313–326.
7[7] I. Holyer, The NP-completeness of edge-coloring, SIAM J Comput. 10 (1981) 718–720.
8[8] S. Kerdjoudj, A. V. Kostochka and A. Raspaud, List star edge coloring of subcubic graphs, to appear in Discuss. Math. Graph Theory.