1 Introduction

In this paper, all graphs considered are finite, simple, and undirected. The terminology and notation used but undefined in this paper can be found in [1]. Let \(G=(V,E)\) be a graph. We use \(V(G), E(G),\Delta (G)\), and \(\delta (G)\) to denote the vertex set, edge set, maximum degree, and minimum degree of G, respectively. Let \(d_G(v)\) or simply d(v) denote the degree of a vertex v in G. A vertex v is called a k-vertex (resp. \(k^{-}\)-vertex, or \(k^{+}\)-vertex) if \(d(v)=k\) [resp. \(d(v)\le k\), or \(d(v)\ge k\)]. A vertex is called a leaf of G if \(d(v)=1\). A 2-vertex is called bad if it is adjacent to a 2-vertex, otherwise we call it good. A 5-vertex is called bad if it is adjacent to four bad 2-vertices, otherwise we call it good. The girth of a graph G is the length of a smallest cycle in G, and we denote it by g(G). The maximum average degree of G is \({ {mad}}(G)=\max \{2|E(H)|/|V(H)|\}\), where the maximum is taken over all the non-empty subgraphs H of G.

A proper edge-k-coloring of a graph G is an assignment of k colors \(1,2,\ldots ,k\) to the edges of G such that no two adjacent edges receive the same color. Let c be a proper edge-k-coloring of G. By w(v) [resp. S(v)], we denote the sum (resp. set) of colors taken on the edges incident with v, i.e., \(w(v) =\sum _{uv\in E(G)}c(uv)\) [resp. \(S(v)=\{c(uv){\mid } uv\in E(G)\}\)]. We call the coloring c such that \(w(u)\ne w(v)\) [resp. \(S(u)\ne S(v)\)] for each edge \(uv\in E(G)\) a neighbor sum distinguishing (resp. neighbor distinguishing) edge-k-coloring of G. For simplicity, we use \({ {nsd}}\)-k-coloring (resp. nd-k-coloring) to denote the neighbor sum distinguishing (resp. neighbor distinguishing) edge-k-coloring of G. By \({ {ndi}}_{\sum }(G)\) [resp. \({ {ndi}}(G)\)], we denote the smallest value k such that G has an \({ {nsd}}\)-k-coloring (resp. nd-k-coloring) of G.

Obviously, a graph G has a neighbor sum distinguishing (neighbor distinguishing) coloring if and only if G has no isolated edges (we call it normal). Apparently, for any normal graph G, \({ {ndi}}(G)\le { {ndi}}_{\sum }(G)\). In 2002, Zhang et al. [2] proposed the following conjecture.

Conjecture 1.1

[2] If G is a normal graph with at least six vertices, then \({ {ndi}}(G)\le \Delta (G)+2\).

Balister et al. [3] proved Conjecture 1.1 for bipartite graphs and for graphs G with \(\Delta (G)=3\). If G is bipartite planar with maximum degree \(\Delta (G)\ge 12\), Conjecture 1.1 was confirmed by Edwards et al. [4]. Hatami [5] showed that if G is a normal graph and \(\Delta (G)>10^{20}\), then \({ {ndi}}(G)\le \Delta (G)+300\). Akbari et al. [6] proved that \({ {ndi}}(G)\le 3 \Delta (G)\) for any normal graph. Wang et al. [7, 8] confirmed Conjecture 1.1 for sparse graphs and \(K_4\)-minor free graphs. More precisely, in [7] they showed that if G is a normal graph and \({ {mad}}(G) <5/2\), then \({ {ndi}}(G)\le \Delta (G)+1\). Furthermore, \({ {ndi}}(G) = \Delta (G)+1\) if and only if G has two adjacent maximum degree vertices. Recently, Hocquard et al. [9] proved that for every normal graph with \(\Delta (G)\ge 5\) and \({ {mad}}(G)<13/5\), we have \({ {ndi}}(G)\le \Delta (G)+1\). Later, in [10] they proved that if G is a normal graph with \(\Delta (G)\ge 5\) and \({ {mad}}(G) <3-2/\Delta (G)\), then \({ {ndi}}(G)\le \Delta (G)+1\).

Recently, Flandrin et al. [11] studied the neighbor sum distinguishing colorings of cycles, trees, complete graphs, and complete bipartite graphs. Based on these examples, they proposed the following conjecture.

Conjecture 1.2

[11] If G is a connected graph on at least three vertices and \(G\ne C_5\), then \({ {ndi}}_{\sum }(G)\le \Delta (G)+2\).

Flandrin et al. [11] also proved that for each connected graph G with maximum degree \(\Delta \ge 2\), we have \({ {ndi}}_{\sum }(G)\le \lceil (7\Delta -4)/2\rceil \). Dong et al. [12] considered the neighbor sum distinguishing colorings of planar graphs and showed that if G is a normal planar graph, then \({ {ndi}}_{\sum }(G)\le \max \{2\Delta (G)+1,25\}\). In [13], Dong et al. proved that if G is a normal graph and \({ {mad}}(G)\le 5/2\), then \({ {ndi}}_{\sum }(G)\le k\) where \(k=\max \{\Delta (G)+1,6\}\). Other results on graph coloring problems are referred to [1416].

In this paper, we will prove the following results.

Theorem 1.3

Let G be a normal graph. If \({ {mad}}(G)<\frac{8}{3}\), then \({ {ndi}}_{\sum }(G)\le k\) where \(k=\max \{\Delta (G)+1,6\}\).

Corollary 1.4

Let G be a normal graph. If \({ {mad}}(G)<\frac{8}{3},\Delta (G)\ge 5\), then \({ {ndi}}_{\sum }(G)\le \Delta (G)+1\).

In [17], the authors obtained that \({ {mad}}(G)<2g/(g-2)\) if G is a planar graph with girth g. The following corollary is obvious.

Corollary 1.5

Let G be a normal planar graph. If \(g(G)\ge 8\) and \(\Delta (G)\ge 5\), then \({ {ndi}}_{\sum }(G)\le \Delta (G)+1\).

We note that if G contains two adjacent vertices of maximum degree, then \({ {ndi}}_{\sum }(G)\ge \Delta (G)+1\). So the bound \(\Delta (G)+1\) in Corollary 1.4 is sharp. Furthermore, Corollary 1.4 implies a result of Hocquard et al. [9] about the neighbor distinguishing coloring of sparse graphs.

2 Proof of Theorem 1.3

Firstly, we give two lemmas obtained by Dong et al. in [13], all the elements in each set are integers.

Lemma 2.1

[13] Let \(S_1,S_2\) be two sets and \(S_3=\{\alpha +\beta {\mid }\alpha \in S_1,\beta \in S_2,\alpha \ne \beta \}\).

  1. (i)

    If \(|S_1|=2\) and \(|S_2|=3\), then \(|S_3|\ge 3\).

  2. (ii)

    If \(|S_1|=2\) and \(|S_2|=4\), then \(|S_3|\ge 4\).

  3. (iii)

    If \(|S_1|=|S_2|=2\) and \(S_1\ne S_2\), then \(|S_3|\ge 3\).

Lemma 2.2

[13] Let S be a set of size \(k+1\). If \(S_1=\bigg \{\sum \limits _{i=1}^{k}x_i{\mid }x_i\in S,x_i\ne x_j\ \text{ if }\ 1\le i<j\le k\bigg \}\), then \(|S_1|\ge k+1\).

Let \(k=\max \{\Delta (G)+1,6\}\) and \([k]=\{1,2,\ldots ,k\}\). Suppose to the contrary that G is a counterexample to Theorem 1.3, such that |E(G)| is minimum. By the choice of G, it is clear that G is connected and any normal subgraph \(G'\) has an \({ {nsd}}\)-k-coloring c. We use w(v) and S(v) to denote the sum and the set of colors taken on the edges incident with v in the coloring c of \(G'\), i.e., \(w(v) =\sum _{v\in e,e\in E(G')}c(e)\) and \(S(v) =\{c(e)\mid v\in e,e\in E(G')\}\). In the following, we will extend c to the whole graph G.

Let H be the graph obtained by removing all the leaves of G. Obviously, H is a connected graph and \({ {mad}}(H)<8/3\). In the following, we give some properties of H.

Claim 2.3

H has the following properties:

  1. (i)

    [13] \(\delta (H)\ge 2\), where \(\delta (H)\) is the minimum degree of H.

  2. (ii)

    [13] Let \(v\in V(H)\) such that \(d_H(v) = 2\), then \(d_G(v) = 2\).

  3. (iii)

    Let uvxy be a path in H such that \(d_H(v) = d_H(x) = 2\), then \(d_G(u) = d_H(u)\) and \(d_G(y)=d_H(y)\) (Fig. 1).

    Fig. 1
    figure 1

    Illustration of Claim 2.3 (iii)

Proof

(iii) Let uvxy be a path in H such that \(d_{H}(v)=d_{H}(x)=2\). By Claim 2.3 (ii), \(d_{G}(v)=d_{G}(x)=2\). By contradiction suppose \(d_{G}(u)\ne d_{H}(u)\) [it follows from Claim 2.3 (i) and construction of H that \(d_{G}(u)\ge 3\)]. Hence there exists at least one 1-vertex adjacent to u in G, say \(u_1\). Consider \(G'=G\backslash \{vx\}\). By the minimality of G, \(G'\) admits an \({ {nsd}}\)-k-coloring c. If \(c(uv)\ne c(xy)\), then we color vx with a color distinct from \(c(uv),c(xy),w(u)-w(v),w(y)-w(x)\), then we obtain an \({ {nsd}}\)-k-coloring of G. Otherwise, we permute the colors assigned to \(uu_1\) and uv. The obtained coloring is still an \({ {nsd}}\)-k-coloring of \(G'\). We then extend this coloring to G as previously. This is a contradiction. \(\square \)

Claim 2.4

Let \(u\in V(H),d_H(u)=l,uu_i\in E(H),i=1,2,\ldots ,l\).

  1. (i)

    [13] If \(l=2\), then u is adjacent to at most one 2-vertex.

  2. (ii)

    (a) [13] If \(l=3\) and \(d_H(u)<d_G(u)\), then u is adjacent to at most one 2-vertex.

  3. (b)

    If \(l=3\), then u is not adjacent to any bad 2-vertex. Furthermore, u is adjacent to at most one good 2-vertex.

  4. (iii)

    If \(l=4\), then u is adjacent to at most one bad 2-vertex. Furthermore, if u is adjacent to one bad 2-vertex, then u is adjacent to at most two good 2-vertices.

  5. (iv)

    If \(l\ge 5\) and u is adjacent to \((l-1)\) bad 2-vertices, then u is adjacent to at most \((l-1)\) 2-vertices (Fig. 2).

Fig. 2
figure 2

Illustration of Claim 2.4

Proof

(ii) (b) Firstly, we prove that u is not adjacent to any bad 2-vertex. Suppose to the contrary that \(u_1\) is a bad 2-vertex. Let x be the other neighbor of \(u_1\) with \(d_H(x)=2\), and \(xy\in E(H)\), \(y\ne u_1\). By Claim 2.3 (iii), \(d_G(u)=d_H(u)=3\). Consider the graph \(G'=G\backslash \{uu_1\}\), then \(G'\) admits an \({ {nsd}}\)-k-coloring c. Color \(uu_1\) with a color \(\alpha \) in \(S=[k]\backslash \left( \{c(xy)\}\bigcup \{c(uu_2),c(uu_3)\}\bigcup \{w(u_2)-w(u)\}\bigcup \{w(u_3)-w(u)\}\right) \). Recolor \(u_1x\) with a color distinct from \(\alpha ,c(xy),w(y)-c(xy),c(uu_2)+c(uu_3)\) and we obtain an \({ {nsd}}\)-k-coloring of G, a contradiction.

In the following, we prove that u is adjacent to at most one good 2-vertex. Suppose to the contrary that \(d_H(u_1)=d_H(u_2)=2\), \(v_iu_i\in E(H),v_i\ne u,i=1,2\).

Case 1 \(d_G(u)>d_H(u)=3\). By Claim 2.4 (ii) (a), this Claim holds.

Case 2 \(d_G(u)=d_H(u)=3\).

Subcase 2.1 \(k\ge 7\). Consider the graph \(G'=G\backslash \{uu_1,uu_2\}\), then \(G'\) has an \({ {nsd}}\)-k-coloring c. Let \(S_i=[k]\backslash \left( \{c(u_iv_i)\}\bigcup \{c(uu_3)\}\bigcup \{w(v_i)-w(u_i)\}\bigcup \{w(u_{3-i})\right. \left. -c(uu_3)\}\right) , i=1,2\), then \(|S_i|\ge 3,i=1,2\). By Lemma 2.1 (ii), we can choose \(\alpha _i\in S_i,i=1,2\) such that \(\alpha _1\ne \alpha _2\), \(\alpha _1+\alpha _2+c(uu_3)\ne w(u_3)\). We obtain an \({ {nsd}}\)-k-coloring of G, which is a contradiction.

Subcase 2.2 \(k=6\). From the above discussion, \(d_G(v_i)\ge 3,i=1,2\). If \(d_G(v_1)=5\), then \(u_1\) can be distinguished from \(v_1\) under an arbitrary proper edge coloring of G. Consider \(G'=G\backslash \{uu_1,uu_2\}\), then \(G'\) has an \({ {nsd}}\)-6-coloring c. The colors in \(\left\{ c(u_1v_1)\}\bigcup \{c(uu_3)\}\bigcup \{c(u_2v_2)-c(uu_3)\right\} \) are forbidden for \(uu_1\). Let \(S_1=[6]\backslash \left( \{c(u_1v_1)\}\bigcup \{c(uu_3)\}\bigcup \{c(u_2v_2)-c(uu_3)\}\right) \), \(S_2=[6]\backslash \left( \{c(u_2v_2)\}\bigcup \{c(uu_3)\}\bigcup \{w(v_2)-w(u_2)\}\bigcup \{c(u_1v_1)-c(uu_3)\}\right) \), then \(|S_1|\ge 3,|S_2|\ge 2\). By Lemma 2.1 (ii), we can choose \(\alpha _i\in S_i,i=1,2\) such that \(\alpha _1\ne \alpha _2\), \(\alpha _1+\alpha _2+c(uu_3)\ne w(u_3)\). We obtain an \({ {nsd}}\)-6-coloring of G, which is a contradiction. Therefore, \(d_G(v_1)\ne 5\). Similarly, \(d_G(v_2)\ne 5\).

If \(d_G(v_1)=3\) and \(x_1,y_1\) are the other two neighbors of \(v_1\). Consider \(G'=G\backslash \{uu_1,uu_2,u_1v_1\}\), then \(G'\) has an \({ {nsd}}\)-6-coloring c. Let \(S_1=[6]\backslash \left( \{c(v_1x_1)\right. \left. +c(v_1y_1)\}\bigcup \{c(uu_3)\}\right) \), \(S_2=[6]\backslash \left( \{c(uu_3)\}\bigcup \{c(u_2v_2)\}\bigcup \{w(v_2)-w(u_2)\}\right) \), \(S_3=[6]\backslash \left( \{c(v_1x_1),c(v_1y_1)\}\bigcup \{w(x_1)-w(v_1)\}\bigcup \{w(y_1)-w(v_1)\}\right) \), then \(|S_1|\ge 4\), \(|S_2|\ge 3\), \(|S_3|\ge 2\). We can choose \(\alpha _i\in S_i,i=1,2,3\) such that \(\alpha _1\ne \alpha _2, \alpha _1\ne \alpha _3\), u can be distinguished from \(u_1,u_2,u_3\), and \(v_1\) can be distinguished from \(x_1,y_1\). We obtain an \({ {nsd}}\)-6-coloring of G, a contradiction. Therefore, \(d_G(v_1)\ne 3\). Similarly, \(d_G(v_2)\ne 3\).

Now we assume that \(d_G(v_1)=d_G(v_2)=4\), \(x_i,y_i,z_i\) are the other three neighbors of \(v_i\), \(i=1,2\). Consider \(G'=G\backslash \{uu_1,uu_2\}\), then \(G'\) has an \({ {nsd}}\)-6-coloring c. If \(c(v_1x_1)+c(v_1y_1)+c(v_1z_1)>6\), then \(u_1\) and \(v_1\) can be distinguished. Let \(S_1=[6]\backslash \left( \{c(uu_3)\}\bigcup \{c(u_1v_1)\}\bigcup \{w(u_2)-c(uu_3)\}\right) \), \(S_2=[6]\backslash \left( \{c(uu_3)\}\bigcup \{c(u_2v_2)\}\bigcup \{w(v_2)-w(u_2)\}\bigcup \{w(u_1)-c(uu_3)\}\right) \), then \(|S_1|\ge 3,|S_2|\ge 2\). We can choose \(\alpha _i\in S_i,i=1,2\) such that \(\alpha _1\ne \alpha _2\), \(\alpha _1+\alpha _2+c(uu_3)\ne w(u_3)\). We obtain an \({ {nsd}}\)-6-coloring of G, which is a contradiction. Therefore, \(c(v_1x_1)+c(v_1y_1)+c(v_1z_1)=6\). Similarly, \(c(v_2x_2)+c(v_2y_2)+c(v_2z_2)=6\). Without loss of generality we assume that \(c(v_ix_i)=1,c(v_iy_i)=2,c(v_iz_i)=3,i=1,2\). Suppose that \(c(u_1v_1)\ne c(u_2v_2)\) or \(c(uu_3)=6\), then we can obtain an \({ {nsd}}\)-6-coloring of G as previously. Hence, \(c(u_1v_1)=c(u_2v_2)\), \(c(uu_3)\ne 6\). From the above discussion, \(d_G(u_3)\ge 3\). If \(d_G(u_3)=3\), let \(x_3,y_3\) be the neighbors of \(u_3\) distinct from u. Consider the graph \(G'=G\backslash \{uu_1,uu_2,uu_3\}\), then \(G'\) has an \({ {nsd}}\)-6-coloring c. Let \(S_1=[5]\backslash \{c(u_1v_1)\}\), \(S_2=[5]\backslash \{c(u_2v_2)\}\), \(S_3=[6]\backslash \left( \{c(v_3x_3),c(v_3y_3)\}\bigcup \{w(x_3)-w(u_3)\}\bigcup \{w(y_3)-w(u_3)\}\right) \), then \(|S_1|\ge 4\), \(|S_2|\ge 4\), \(|S_3|\ge 2\). From the above discussion, we know that \(S_1=S_2=\{1,2,3,4\}\) or \(\{1,2,3,5\}\), so we can choose \(\alpha _i\in S_i,i=1,2,3\) such that \(\alpha _1,\alpha _2,\alpha _3\) are pairwise distinct and u can be distinguished from \(u_1,u_2,u_3\). We obtain an \({ {nsd}}\)-6-coloring of G, which is a contradiction. Therefore, \(d_G(u_3)\ge 4\). If \(c(uu_3)\in \{1,2,3\}\), color \(uu_1,uu_2\) properly with \(\{1,2,3\}\backslash \{c(uu_3)\}\). Otherwise, properly color \(uu_1,uu_2\) with colors in \(\{1,2,3\}\). In both cases, we obtain an \({ {nsd}}\)-6-coloring of G, a contradiction.

(iii) Suppose to the contrary that \(d_H(u_1)=d_H(u_2)=2\), \(u_iv_i\in E(H),d_H(v_i)=2,i=1,2\), \(x_i\) is the other neighbor of \(v_i,i=1,2\). By Claim 2.3 (iii), \(d_G(u)=d_H(u)=4\). Consider the graph \(G'=G\backslash \{uu_1,uu_2\}\), then \(G'\) has an \({ {nsd}}\)-k-coloring c. Let \(S_i=[k]\backslash \left( \{c(uu_3),c(uu_4)\} \bigcup \{c(v_ix_i)\}\right) ,i=1,2\), then \(|S_i|\ge 3\). By Lemma 2.1 (i), we can choose \(\alpha _i\in S_i,i=1,2\) such that \(\alpha _1\ne \alpha _2\) and u can be distinguished from \(u_3,u_4\). Recolor \(u_iv_i\) with a color distinct from \(\alpha _i,c(v_ix_i),w(x_i)-c(v_ix_i),\alpha _1+\alpha _2+c(uu_3)+c(uu_4)-\alpha _i\), \(i=1,2\), then u can be distinguished from \(u_1,u_2\). We obtain an \({ {nsd}}\)-k-coloring of G, a contradiction.

Now assume that u is adjacent to a bad 2-vertex \(u_1\) with \(u_1v_1\in E(H),d_H(v_1)=2,v_1x_1\in E(H),x_1\ne v_1\). Suppose to the contrary that \(d_H(u_i)=2,u_iv_i\in E(H),v_i\ne u,i=2,3,4\). By Claim 2.3 (iii), \(d_G(u)=d_H(u)=4\). Consider \(G'=G\setminus \{uu_1\}\), then \(G'\) has an \({ {nsd}}\)-k-coloring c. Let \(S=[k]\setminus (\{c(uu_2),c(uu_3),c(uu_4)\}\bigcup \{c(v_1x_1)\})\). When \(k\le 7\), if \(1\in S\), then color \(uu_1\) with \(\alpha \in S\setminus \{1\}\), otherwise color \(uu_1\) with \(\alpha \in S\setminus \{2\}\). In both cases \(w(u)+\alpha >w(u_i),i=2,3,4\). Then recolor \(u_1v_1\) with a color distinct from \(\alpha ,c(v_1x_1),w(u),w(x_1)-c(v_1x_1)\). We obtain an \({ {nsd}}\)-k-coloring of G, a contradiction. When \(k\ge 8\), \(|S|\ge 4\), we can choose \(\alpha \in S\) such that \(\alpha +w(u)\ne w(u_i),i=2,3,4\). Then recolor \(u_1v_1\) with a color distinct from \(\alpha ,c(v_1x_1),w(u),w(x_1)-c(v_1x_1)\). We obtain an \({ {nsd}}\)-k-coloring of G, a contradiction.

(iv) Suppose to the contrary that \(d_H(u_i)=2,i=1,2,\ldots ,l\), \(u_iv_i\in E(H),v_i\ne u,i=1,2,\ldots ,l\) and \(d_H(v_j)=2,v_jx_j\in E(H),x_j\ne v_j,j=1,2,\ldots ,l-1\). By Claim 2.3 (iii), \(d_G(u)=d_H(u)=l\). Let \(G'=G\backslash \{uu_1\}\), then \(G'\) has an \({ {nsd}}\)-k-coloring c. If \(l<\Delta =k-1\), color \(uu_1\) with \(\alpha \in [k]\setminus \left( \{c(uu_2),\ldots ,c(uu_l)\}\bigcup \{c(v_1x_1)\}\bigcup \left\{ \alpha +\sum \limits _{i=2}^{l-1}c(uu_i)\right\} \right) \). Otherwise color \(uu_1\) with \(\alpha \in [k]\setminus \left( \{c(uu_2),\ldots ,c(uu_l)\}\bigcup \{c(v_1x_1)\}\right) \). In both cases, u can be distinguished from \(u_l\). Properly recolor \(u_iv_i\) such that u can be distinguished from \(u_i\) and \(v_i\) can be distinguished from \(x_i,i=1,2,\ldots ,l-1\). We obtain an \({ {nsd}}\)-k-coloring of G, a contradiction. \(\square \)

Claim 2.5

Let \(u\in V(H), d_H(u)=5,uu_i\in E(H),i=1,2,3,4,5\).

  1. (i)

    If \(\Delta (G)\ge 6\), then u is adjacent to at most two bad 2-vertices. If \(\Delta (G)=5\) and u is adjacent to three bad 2-vertices, then u is adjacent to at most one good 2-vertex.

    Furthermore, if \(\Delta (G)=5\) and u is a bad 5-vertex, then by Claim 2.3 (iii), \(d_G(u)=d_H(u)=5\). Let \(d_H(u_i)=2,u_iv_i\in E(H)\), \(d_H(v_i)=2\), \(x_i\) be the other neighbor of \(v_i,i=1,2,3,4\), we have

  2. (ii)

    \(d_H(u_5)\ge 4\).

  3. (iii)

    If \(d_H(u_5)=4\), then \(u_5\) is adjacent to no bad 2-vertex.

  4. (iv)

    If \(d_H(u_5)=5\), then \(u_5\) is adjacent to at most two bad 2-vertices (Fig. 3).

Fig. 3
figure 3

Illustration of Claim 2.5

Proof

(i) Assume \(\Delta (G)\ge 6\). Suppose to the contrary that \(d_H(u_1)=d_H(u_2)=d_H(u_3)=2\), \(v_i\) is the other neighbor of \(u_i\) with \(d_H(v_i)=2\), \(x_i\) is the other neighbor of \(v_i,i=1,2,3\). By Claim 2.3 (iii), \(d_G(u)=d_H(u)=5\). Consider the graph \(G'=G\backslash \{uu_1,uu_2,uu_3\}\), then \(G'\) has an \({ {nsd}}\)-k-coloring c. Let \(S_i=[k]\backslash \left( \{c(uu_4),c(uu_5)\}\bigcup \{c(v_ix_i)\}\right) ,i=1,2,3\), then \(|S_i|\ge 4,i=1,2,3\). We can choose \(\alpha _i\in S_i,i=1,2,3\) such that \(\alpha _1,\alpha _2,\alpha _3\) are pairwise distinct and u can be distinguished by \(u_4,u_5\). Recolor \(u_iv_i\) with a color distinct from \(\alpha _i,c(v_ix_i),w(x_i)-c(v_ix_i),\sum \limits _{i=1}^{3}\alpha _i+w(u)-\alpha _i,i=1,2,3\), then u can be distinguished from \(u_1,u_2,u_3\). We obtain an \({ {nsd}}\)-k-coloring of G, a contradiction.

Assume that \(\Delta (G)=5\). Suppose to the contrary that \(d_H(u_i)=2\), \(u_iv_i\in E(H),v_i\ne u,i=1,2,\ldots ,5\) and \(d_H(v_j)=2,v_jx_j\in E(H),j=1,2,3\). By Claim 2.3 (iii), \(d_G(u)=d_H(u)=5\). Consider the graph \(G'=G\backslash \{uu_1\}\), then \(G'\) has an \({ {nsd}}\)-6-coloring c. Color \(uu_1\) with \(\alpha \in [6]\backslash \left( \{c(uu_2),\ldots ,c(uu_5)\}\bigcup \{c(v_1x_1)\}\right) \). Then recolor \(u_1v_1\) with a color distinct from \(\alpha ,c(v_1x_1),w(x_i)-c(v_1x_1),\sum \limits _{i=2}^{5}c(uu_i)\). It can be seen that \(w(u)+\alpha >w(u_i),i=2,\ldots ,5\), so we obtain an \({ {nsd}}\)-6-coloring of G, a contradiction.

(ii) By Claim 2.4 (iv), \(d_H(u_5)\ge 3\). Suppose to the contrary that \(d_H(u_5)=3\), \(v_{51},v_{52}\) are the other two neighbors of \(u_5\). Consider the graph \(G'=G\backslash \{uu_1,uu_2,uu_3,uu_4,uu_5\}\), then \(G'\) has an \({ {nsd}}\)-6-coloring c. Let \(S_i=[6]\backslash \{c(v_ix_i)\},i=1,2,3,4\), \(S_5=[6]\backslash \left( \{c(u_5v_{51}),c(u_5v_{52})\} \bigcup \{w(v_{51})-w(u_5)\}\bigcup \{w(v_{52})-w(u_5)\}\right) \), then \(|S_i|\ge 5,i=1,2,3,4\), \(|S_5|\ge 2\). We can choose \(\alpha _i\in S_i,i=1,2,3,4,5\) such that \(\alpha _1,\alpha _2,\alpha _3,\alpha _4,\alpha _5\) are pairwise distinct and u can be distinguished by \(u_5\). Obviously, u can be distinguished by \(u_1,u_2,u_3,u_4\). Recolor \(u_iv_i\) with a color distinct from \(\alpha _i,c(v_ix_i),w(x_i)-c(v_ix_i),i=1,2,3,4\) and we obtain an \({ {nsd}}\)-6-coloring of G, a contradiction.

(iii) Let \(v_{51},v_{52},v_{53}\) be the other three neighbors of \(u_5\). Suppose to the contrary that \(d_H(v_{51})=d_H(x_{51})=2,v_{51}x_{51}\in E(H)\), \(x_{51}y_{51}\in E(H),y_{51}\ne v_{51}\). By Claim 2.3 (iii), \(d_G(u_5)=d_H(u_5)=4\). Consider the graph \(G'=G\backslash \{uu_1,uu_2,uu_3,uu_4\}\), then \(G'\) has an \({ {nsd}}\)-6-coloring c. Assume that \(c(uu_5)=\alpha _5\), \(c(u_5v_{5j})=\beta _j,j=1,2,3\), \(c(v_ix_i)=\gamma _i,i=1,2,3,4\), \(c(x_{51}y_{51})=\eta \). Let \(S_i=[6]\backslash (\{\gamma _i\}\bigcup \{\alpha _5\}),i=1,2,3,4\), then \(|S_i|\ge 4,i=1,2,3,4\).

If there exists some \(\gamma _i=\alpha _5\) (\(i\in \{1,2,3,4\}\)) or \(\gamma _i\ne \gamma _j\) (\(i\ne j,i,j\in \{1,2,3,4\}\)), then we can choose \(\alpha _i\in S_i,i=1,2,3,4\) such that u can be distinguished by \(u_5\). Recolor \(u_iv_i\) with a color distinct from \(\gamma _i,\alpha _i,w(x_i)-c(v_ix_i),i=1,2,3,4\), we can obtain an \({ {nsd}}\)-6-coloring of G, a contradiction. Therefore, we assume that \(\gamma _1=\gamma _2=\gamma _3=\gamma _4=\gamma ,\alpha _5\ne \gamma \). Then \(S_1=S_2=S_3=S_4=[6]\backslash \{\alpha _5,\gamma \}\). Assume that \(S_i=\{\alpha _1,\alpha _2,\alpha _3,\alpha _4\},i=1,2,3,4\). Color \(uu_i\) with \(\alpha _i,i=1,2,3,4\).

If \(|\{\alpha _1,\alpha _2,\alpha _3,\alpha _4\}\bigcap \{\beta _1,\beta _2,\beta _3\}|=3\), then u can be distinguished by \(u_5\), recolor \(u_iv_i,i=1,2,3,4\) as previously and we obtain an \({ {nsd}}\)-6-coloring of G, a contradiction. Furthermore, \(\left| \{\alpha _1,\alpha _2,\alpha _3,\alpha _4\}\bigcap \{\beta _1,\beta _2,\beta _3\}\right| \ge 2\) because there are six colors in total. Therefore, \(\left| \{\alpha _1,\alpha _2,\alpha _3,\alpha _4\} \bigcap \{\beta _1,\beta _2,\beta _3\}\right| =2\). Without loss of generality we assume that \(\{\alpha _1,\alpha _2\}\subseteq \{\beta _1,\beta _2,\beta _3\}\).

If \(\beta _1=\gamma \), then \(\{\alpha _1,\alpha _2\}=\{\beta _2,\beta _3\}\). If \(\alpha _5\ne \eta \), suppose u can not be distinguished by \(u_5\), then \(\gamma =\alpha _3+\alpha _4\). Recolor \(uu_5\) with \(\gamma \) and recolor \(u_5v_{51}\) with \(\alpha _5\), we obtain an \({ {nsd}}\)-6-coloring of G, a contradiction. Therefore, \(\alpha _5=\eta \). Recolor \(u_5v_{51}\) with one of \(\alpha _3,\alpha _4\), or exchange the colors of \(uu_3\) and \(uu_5\), or exchange the colors of \(uu_4\) and \(uu_5\) such that \(u_5\) can be distinguished by \(u,v_{52},v_{53}\). It is easy to see that u can be distinguished by \(u_1,u_2,u_3,u_4\). Recolor \(u_iv_i,i=1,2,3,4\), and \(v_{51}x_{51}\) as previously and we can obtain an \({ {nsd}}\)-6-coloring of G, a contradiction.

If \(\beta _1\ne \gamma \). Without loss of generality we assume that \(\beta _1=\alpha _1,\beta _2=\alpha _2\). Recolor \(u_5v_{51}\) with one of \(\alpha _3,\alpha _4\), or exchange the colors of \(uu_3\) and \(uu_5\), or exchange the colors of \(uu_4\) and \(uu_5\) such that \(u_5\) can be distinguished by \(u,v_{52},v_{53}\). It is easy to see that u can be distinguished by \(u_1,u_2,u_3,u_4\). Recolor \(u_iv_i,i=1,2,3,4\), and \(v_{51}x_{51}\) as previously and we can obtain an \({ {nsd}}\)-6-coloring of G, a contradiction.

(iv) Let \(v_{51},v_{52},v_{53},v_{54}\) be the other four neighbors of \(u_5\). Suppose to the contrary that \(d_H(v_{5i})=d_H(x_{5i})=2,v_{5i}x_{5i}\in E(H)\), \(x_{5i}y_{5i}\in E(H),y_{5i}\ne v_{5i},i=1,2,3\). By Claim 2.3 (iii), \(d_G(u_5)=d_H(u_5)=5\). Consider the graph \(G'=G\backslash \{uu_1,uu_2,uu_3,uu_4\}\), then \(G'\) has an \({ {nsd}}\)-6-coloring c. Assume that \(c(uu_5)=\alpha _5\), \(c(u_5v_{5j})=\beta _j,j=1,2,3,4\), \(c(v_ix_i)=\gamma _i,i=1,2,3,4\). Let \(S_i=[6]\backslash \left( \{\gamma _i\}\bigcup \{\alpha _5\}\right) ,i=1,2,3,4\), then \(|S_i|\ge 4,i=1,2,3,4\).

If there exists some \(\gamma _i=\alpha _5\) (\(i\in \{1,2,3,4\}\)) or \(\gamma _i\ne \gamma _j\) (\(i\ne j,i,j\in \{1,2,3,4\}\)), then we can choose \(\alpha _i\in S_i,i=1,2,3,4\) such that u can be distinguished by \(u_5\). Recolor \(u_iv_i\) with a color distinct from \(\alpha _i,\gamma _i,w(x_i)-c(v_ix_i),i=1,2,3,4\). We obtain an \({ {nsd}}\)-6-coloring of G, a contradiction. Therefore, we assume that \(\gamma _1=\gamma _2=\gamma _3=\gamma _4=\gamma ,\alpha _5\ne \gamma \). Then \(S_1=S_2=S_3=S_4=[6]\backslash \{\alpha _5,\gamma \}\). Assume that \(S_i=\{\alpha _1,\alpha _2,\alpha _3,\alpha _4\}\). Color \(uu_i\) with \(\alpha _i\) and recolor \(u_iv_i\) with a color distinct from \(\gamma ,\alpha _i,w(x_i)-c_(v_ix_i),i=1,2,3,4\).

If \(\{\alpha _1,\alpha _2,\alpha _3,\alpha _4\}\ne \{\beta _1,\beta _2,\beta _3,\beta _4\}\), then u can be distinguished by \(u_5\), it is easy to see that u can be distinguished from \(u_1,u_2,u_3,u_4\). We obtain an \({ {nsd}}\)-6-coloring of G, a contradiction. Therefore, \(\{\alpha _1,\alpha _2,\alpha _3,\alpha _4\}= \{\beta _1,\beta _2,\beta _3,\beta _4\}\). Without loss of generality we assume that \(\alpha _i=\beta _i,i=1,2,3,4\). If \(c(x_{51}y_{51})\ne \gamma \), then recolor \(u_5v_{51}\) with \(\gamma \). We can see that u can be distinguished from \(u_5\). If \(u_5\) can not be distinguished from \(v_{54}\), then exchange the colors of \(uu_1\) and \(uu_5\). Recolor \(v_{51}x_{51}\) with a color distinct from \(\gamma _1,c(x_{51}y_{51}),w(y_{51})-c(x_{51}y_{51})\). We obtain an \({ {nsd}}\)-6-coloring of G, a contradiction. Similarly, \(c(x_{52}y_{52})=c(x_{53}y_{53})=\gamma \). Recolor \(uu_5\) with \(\gamma \). Then recolor \(u_5v_{51}\) or recolor \(u_5u_{52}\) with \(\alpha _5\) such that \(u_5\) can be distinguished by \(v_{54}\). Recolor \(v_{5i}x_{5i}\) with a color distinct from \(\gamma ,\alpha _5,\alpha _i,w(y_{5i})-c(x_{5i}y_{5i}),i=1,2\). We obtain an \({ {nsd}}\)-6-coloring of G, a contradiction. \(\square \)

In order to complete the proof, we use a discharging procedure. For every \(v\in V(H)\), we define the original charge of v to be \(ch(v)=d_H(v)=l\). We then redistribute the charges according to the rules R1, R2, and R3 (below). To complete the proof, our aim is to prove that, for every vertex v, the new charge \({ {ch}}^{*}(v)\) is at least 8/3.

The discharging rules are defined as follows:

  1. (R1)

    Every \(4^{+}\)-vertex gives \(\frac{2}{3}\) to each adjacent bad 2-vertex.

  2. (R2)

    Every \(3^{+}\)-vertex gives \(\frac{1}{3}\) to each adjacent good 2-vertex.

  3. (R3)

    If u is a bad 5-vertex, \(u_i,i=1,2,3,4,5\), are the neighbors of u, and if \(u_1,u_2,u_3,u_4\) are bad 2-vertices, then \(u_5\) gives \(\frac{1}{3}\) to u.

Case \(l=2\). Observe that \(ch(v)=2\). Suppose v is a good 2-vertex. Hence, by (R2), \({ {ch}}^{*}(v)\ge 2+2\times \frac{1}{3}=\frac{8}{3}\). Suppose v is bad, By Claim 2.4 (i) and Claim 2.4 (ii), v is adjacent to at most one 2-vertex and is adjacent to a \(4^{+}\)-vertex. Hence, by (R1), \({ {ch}}^{*}(v)= 2+1\times \frac{2}{3}=\frac{8}{3}\).

Case \(l=3\). Observe that \(ch(v)=3\). By Claim 2.4 (ii), v is adjacent to no bad 2-vertex and is adjacent to at most one good 2-vertex. By Claim 2.5 (ii), v is adjacent to no bad 5-vertex. By (R2) and (R3), \({ {ch}}^{*}(v)\ge 3-1\times \frac{1}{3}=\frac{8}{3}\).

Case \(l=4\). Observe that \(ch(v)=4\). Suppose v is not adjacent to a bad 2-vertex. Then, by (R2) and (R3), \({ {ch}}^{*}(v)\ge 4-4\times \frac{1}{3}=\frac{8}{3}\). Assume now, v is adjacent to a bad 2-vertex. By Claim 2.4 (iii), v is adjacent to at most two good 2-vertices. By Claim 2.5 (iii), v is adjacent to no bad 5-vertex. Hence by (R1), (R2), and (R3), \({ {ch}}^{*}(v)\ge 4-1\times \frac{2}{3}-2\times \frac{1}{3}=\frac{8}{3}\).

Case \(l=5\). Observe that \(ch(v)=5\). Suppose v is adjacent to at most two bad 2-vertices, then by (R1), (R2), and (R3), \({ {ch}}^{*}(v)\ge 5-2\times \frac{2}{3}-3\times \frac{1}{3}=\frac{8}{3}\). If v is adjacent to three bad 2-vertices, then by Claim 2.5 (i) and (iv), v is adjacent to at most one good 2-vertex and v is adjacent to no bad 5-vertex. Hence by (R1), (R2), and (R3), \({ {ch}}^{*}(v)\ge 5-3\times \frac{2}{3}-1\times \frac{1}{3}=\frac{8}{3}\). Assume now, v is a bad 5-vertex. By (R1), (R2), and (R3), \({ {ch}}^{*}(v)\ge 5-4\times \frac{2}{3}+1\times \frac{1}{3}=\frac{8}{3}\).

Case \(l\ge 6\). Observe that \(ch(v)=l\). By Claim 2.5 (i), any 5-vertex in H is good. By Claim 2.4 (iv), v is adjacent to at most \((l-1)\) bad 2-vertices. Moreover if v is adjacent to \((l-1)\) bad 2-vertices, then its last neighbor has degree at least 3. It follows by (R1), \({ {ch}}^{*}(v)\ge l-(l-1)\times \frac{2}{3}\ge \frac{8}{3}\).

This completes the proof of Theorem 1.3.