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

Upper bounds for the MDMD-numbers and characterization of extremal graphs111Supported by NSFC No.11871034 and 11531011.

Ping Li, Xueliang Li
Center for Combinatorics and LPMC
Nankai University, Tianjin 300071, China
Email: qdli [email protected], [email protected]
Abstract

For an edge-colored graph GG, we call an edge-cut MM of GG monochromatic if the edges of MM are colored with the same color. The graph GG is called monochromatic disconnected if any two distinct vertices of GG are separated by a monochromatic edge-cut. For a connected graph GG, the monochromatic disconnection number (or MDMD-number for short) of GG, denoted by md(G)md(G), is the maximum number of colors that are allowed in order to make GG monochromatic disconnected. For graphs with diameter one, they are complete graphs and so their MDMD-numbers are 11. For graphs with diameter at least 3, we can construct 22-connected graphs such that their MDMD-numbers can be arbitrarily large; whereas for graphs GG with diameter two, we show that if GG is a 22-connected graph then md(G)2md(G)\leq 2, and if GG has a cut-vertex then md(G)md(G) is equal to the number of blocks of GG. So, we will focus on studying 22-connected graphs with diameter two, and give two upper bounds of their MDMD-numbers depending on their connectivity and independent numbers, respectively. We also characterize the n2\left\lfloor\frac{n}{2}\right\rfloor-connected graphs (with large connectivity) whose MDMD-numbers are 22 and the 22-connected graphs (with small connectivity) whose MDMD-numbers archive the upper bound n2.\left\lfloor\frac{n}{2}\right\rfloor. For graphs with connectivity less than n2\frac{n}{2}, we show that if the connectivity of a graph is in linear with its order nn, then its MDMD-number is upper bounded by a constant, and this suggests us to leave a conjecture that for a kk-connected graph GG, md(G)nkmd(G)\leq\left\lfloor\frac{n}{k}\right\rfloor.
Keywords: monochromatic disconnection number, connectivity, diameter, independent number, upper bound, extremal graph.
AMS subject classification (2020): 05C15, 05C40, 05C35.

1 Introduction

Let GG be a graph and let V(G)V(G), E(G)E(G) denote the vertex-set and the edge-set of GG, respectively. We use |G||G| and G||G|| to denote the number of vertices and the number of edges of GG, respectively, and call them the order and the size of GG. If there is no confusion, we also use nn and mm to denote |G||G| and G||G||, respectively, throughout this paper. Let SS and FF be a vertex subset and an edge subset of GG, respectively. Then GSG-S is the graph obtained from GG by deleting the vertices of SS together with the edges incident with vertices of SS, and GFG-F is the graph whose vertex-set is V(G)V(G) and edge-set is E(G)FE(G)-F. Let G[S]G[S] and G[F]G[F] be the subgraphs of GG induced, respectively, by SS and FF. We use [r][r] to denote the set {1,2,,r}\{1,2,\cdots,r\} of positive integers. If r=0r=0, then set [r]=[r]=\emptyset. For all other terminology and notation not defined here we follow Bondy and Murty [4].

For a graph GG, let Γ:E(G)[r]\Gamma:E(G)\rightarrow[r] be an edge-coloring of GG that allows a same color to be assigned to adjacent edges. For an edge ee of GG, we use Γ(e)\Gamma(e) to denote the color of ee. If HH is a subgraph of GG, we also use Γ(H)\Gamma(H) to denote the set of colors on the edges of HH and use |Γ(H)||\Gamma(H)| to denote the number of colors in Γ(H)\Gamma(H). For an edge-colored graph GG and a vertex vv of GG, the color-degree of vv, denoted by dc(v)d^{c}(v), is the number of colors appearing on the edges incident with vv.

The three main colored connection colorings: rainbow connection coloring [8], proper connection coloring [5] and proper-walk connection coloring [3], monochromatic connection coloring [6], have been well-studied in recent years. As a counterpart concept of the rainbow connection coloring, rainbow disconnection coloring was introduced in [7] by Chartrand et al. in 2018. Subsequently, the concepts of monochromatic disconnection coloring and proper disconnection coloring were also introduced in [12] and [1, 9]. We refer to [2] for the philosophy of studying these so-called global graph colorings. More details on the monochromatic disconnection coloring can be found in [13]. We will further study this coloring in this paper and get some deeper and stronger results.

For an edge-colored graph GG, we call an edge-cut MM a monochromatic edge-cut if the edges of MM are colored with the same color. If there is a monochromatic uvuv-cut with color ii, then we say that color ii separates uu and vv. We use CΓ(u,v)C_{\Gamma}(u,v) to denote the set of colors in Γ(G)\Gamma(G) that separate uu and vv, and let cΓ(u,v)=|CΓ(u,v)|c_{\Gamma}(u,v)=|C_{\Gamma}(u,v)|.

An edge-coloring of a graph is called a monochromatic disconnection coloring (or MDMD-coloring for short) if each pair of distinct vertices of the graph has a monochromatic edge-cut separating them, and the graph is called monochromatic disconnected. For a connected graph GG, the monochromatic disconnection number (or MDMD-number for short) of GG, denoted by md(G)md(G), is defined as the maximum number of colors that are allowed in order to make GG monochromatic disconnected. An extremal MDMD-coloring of GG is an MDMD-coloring that uses md(G)md(G) colors. If HH is a subgraph of GG and Γ\Gamma is an edge-coloring of GG, we call Γ\Gamma an edge-coloring restricted on HH.

The following terminology and notation are needed in the sequel. Let GG and HH be two graphs. The union of GG and HH is the graph GHG\cup H with vertex-set V(G)V(H)V(G)\cup V(H) and edge-set E(G)E(H)E(G)\cup E(H). The intersect of GG and HH is the graph GHG\cap H with vertex-set V(G)V(H)V(G)\cap V(H) and edge-set E(G)E(H)E(G)\cap E(H). The Cartesian product of GG and HH is the graph GHG\Box H with V(GH)={(u,v):uV(G),vV(H)}V(G\Box H)=\{(u,v):u\in V(G),v\in V(H)\}, (u,v)(u,v) and (x,y)(x,y) are adjacent in GHG\Box H if either uxux is an edge of GG and v=yv=y, or vyvy is an edge of HH and u=xu=x. If GG and HH are vertex-disjoint, then let GHG\vee H denote the join of GG and HH which is obtained from GG and HH by adding an edge between every vertex of GG and every vertex of HH.

For a graph GG, a pendent vertex of GG is a vertex with degree one. The ends of GG is the set of pendent vertices, and the internal vertex set of GG is the set of vertices with degree at least two. We use end(G)end(G) and I(G)I(G) to denote the ends of GG and the internal vertex set of GG, respectively. The independent number of GG, denoted by α(G)\alpha(G), is the order of a maximum independent set of GG. For two vertices u,vu,v of GG, we use N(u)N(u) to denote the neighborhood of uu in GG, and N(u,v)N(u,v) to denote the set of common neighbors of uu and vv in GG. The distance between uu and vv in GG is denoted by d(u,v)d(u,v), and the diameter of GG is denoted by diam(G)diam(G). We call a cycle CC (path PP) a tt-cycle (tt-path) if |C|=t|C|=t (P=t||P||=t). If tt is even (odd), then we call the path an even (odd) path and the cycle an even (odd) cycle. A 33-cycle is also called a triangle. A matching-cut of GG is an edge-cut of GG, which also forms a matching in GG.

In [12, 13] we got the following results, which are restated for our later use.

Lemma 1.1.

[12]

  1. 1.

    If a connected graph GG has rr blocks B1,,BrB_{1},\cdots,B_{r}, then md(G)=i[r]md(Bi)md(G)=\sum_{i\in[r]}md(B_{i}) and md(G)=n1md(G)=n-1 if and only if GG is a tree.

  2. 2.

    md(G)=|G|2md(G)=\lfloor\frac{|G|}{2}\rfloor if GG is a cycle, and md(G)=1md(G)=1 if GG is a complete graph with order at least two.

  3. 3.

    If HH is a connected spanning subgraph of GG, then md(H)md(G)md(H)\geq md(G). Thus, md(G)n1md(G)\leq n-1.

  4. 4.

    If GG is connected, then md(vG)=1md(v\vee G)=1.

  5. 5.

    If vv is neither a cut-vertex nor a pendent vertex of GG and Γ\Gamma is an extremal MDMD-coloring of GG, then Γ(G)Γ(Gv)\Gamma(G)\subseteq\Gamma(G-v), and thus, md(G)md(Gv)md(G)\leq md(G-v).

Theorem 1.2.

[12] If GG is a 22-connected graph, then md(G)n2md(G)\leq\left\lfloor\frac{n}{2}\right\rfloor.

Theorem 1.3.

[13] If G1G_{1} and G2G_{2} are connected graphs, then md(G1G2)=md(G1)+md(G2)md(G_{1}\Box G_{2})=md(G_{1})+md(G_{2}).

Lemma 1.4.

[13] If GG has a matching-cut, then md(G)2md(G)\geq 2.

We will list some easy observations in the following, which will be used many times throughout this paper. Suppose Γ\Gamma is an MDMD-coloring of GG. If HH is a subgraph of GG, then Γ\Gamma is an MDMD-coloring restricted on HH. Every triangle of GG is monochromatic. If GG is a 44-cycle, then its opposite edges have the same color. If GG is a 55-cycle, then there are two adjacent edges having the same color.

Let VV be a set of vertices and let 2V\mathcal{E}\subseteq 2^{V}. Then a hypergraph =(V,)\mathcal{H}=(V,\mathcal{E}) is a linear hypergraph if |Ei|2|E_{i}|\geq 2 and |EiEj|1|E_{i}\cap E_{j}|\leq 1 for any Ei,EjE_{i},E_{j}\in\mathcal{E}. The size of \mathcal{H} is the number of hyperedges in \mathcal{H}. A hyperedge-coloring of \mathcal{H} assigns each hyperedge a positive integer. A linear hypergraph \mathcal{H} (say the size of \mathcal{H} is kk) is a linear hypercycle if there is a sequence of hyperedges of \mathcal{H}, say E1,,EkE_{1},\cdots,E_{k}, and there exist kk distinct vertices v1,,vkv_{1},\cdots,v_{k} of \mathcal{H}, such that E1Ek={vk}E_{1}\cap E_{k}=\{v_{k}\} and EiEi+1={vi}E_{i}\cap E_{i+1}=\{v_{i}\} for i[k1]i\in[k-1]. If we delete a hyperedge from a linear hypercycle and then delete the vertices only in this hyperedge, then we call the resulting hypergraph a linear hyperpath. A linear hypercycle (linear hyperpath) is called a linear hyper kk-cycle (linear hyper kk-path) if the size of this linear hypercycle (linear hyperpath) is kk.

2 Preliminaries

We need some more preparations before proceeding to our main results.

Lemma 2.1.

For two connected graphs G1G_{1} and G2G_{2}, if md(G1G2)=1md(G_{1}\cap G_{2})=1 then md(G1G2)=md(G1)+md(G2)1md(G_{1}\cup G_{2})=md(G_{1})+md(G_{2})-1.

Proof.

Let G=G1G2G=G_{1}\cup G_{2} and Γ\Gamma be an extremal MDMD-coloring of GG. Then |Γ(G1G2)|=1|\Gamma(G_{1}\cap G_{2})|=1 and Γ\Gamma is an MDMD-coloring restricted on G1G_{1} (and also G2G_{2}). So, md(G1G2)=|Γ(G1)|+|Γ(G2)||Γ(G1G2)|md(G1)+md(G2)1md(G_{1}\cup G_{2})=|\Gamma(G_{1})|+|\Gamma(G_{2})|-|\Gamma(G_{1}\cap G_{2})|\leq md(G_{1})+md(G_{2})-1. On the other hand, since E(G1G2)E(G_{1}\cap G_{2}) is monochromatic under any MDMD-coloring of G1G2G_{1}\cup G_{2}, let Γi\Gamma_{i} be an MDMD-coloring of GiG_{i} for i[2]i\in[2] such that Γ1(G1G2)=Γ2(G1G2)=Γ(G1)Γ(G2)\Gamma_{1}(G_{1}\cap G_{2})=\Gamma_{2}(G_{1}\cap G_{2})=\Gamma(G_{1})\cap\Gamma(G_{2}). Let Γ\Gamma^{\prime} be an edge-coloring of G1G2G_{1}\cup G_{2} such that Γ(e)=Γi(e)\Gamma^{\prime}(e)=\Gamma_{i}(e) if eE(Gi)e\in E(G_{i}), and let ww be a vertex of G1G2G_{1}\cap G_{2}. Then for any two vertices u,vu,v of G1G2G_{1}\cup G_{2}, if u,vV(Gi)u,v\in V(G_{i}), then CΓi(u,v)CΓ(u,v)C_{\Gamma_{i}}(u,v)\subseteq C_{\Gamma^{\prime}}(u,v); if uV(G1)V(G2)u\in V(G_{1})-V(G_{2}) and vV(G2)V(G1)v\in V(G_{2})-V(G_{1}), then (CΓ1(u,w)CΓ2(v,w))CΓ(u,v)(C_{\Gamma_{1}}(u,w)\cup C_{\Gamma_{2}}(v,w))\subseteq C_{\Gamma^{\prime}}(u,v). So, Γ\Gamma^{\prime} is an MDMD-coloring of GG, i.e., md(G1G2)|Γ(G1G2)|=md(G1)+md(G2)1md(G_{1}\cup G_{2})\geq|\Gamma(G_{1}\cup G_{2})|=md(G_{1})+md(G_{2})-1. Therefore, md(G1G2)=md(G1)+md(G2)1md(G_{1}\cup G_{2})=md(G_{1})+md(G_{2})-1.  

Lemma 2.2.

Let GG be a connected graph and let GG^{\prime} be a graph obtained from GG by replacing an edge e=abe=ab with a path PP. Then md(G)md(G)+P12md(G^{\prime})\geq md(G)+\left\lfloor\frac{||P||-1}{2}\right\rfloor.

Proof.

Let Γ\Gamma be an extremal MDMD-coloring of GG. Let P=t||P||=t and let P=ae1c1etbP=ae_{1}c_{1}\cdots e_{t}b. Let Γ\Gamma^{\prime} be an edge-coloring of GG^{\prime} such that Γ(f)=Γ(f)\Gamma(f)=\Gamma^{\prime}(f) when fE(G)ef\in E(G)-e, Γ(ei)=Γ(et+1i)=|Γ(G)|+i\Gamma^{\prime}(e_{i})=\Gamma^{\prime}(e_{t+1-i})=|\Gamma(G)|+i for i[t12]i\in[\left\lfloor\frac{t-1}{2}\right\rfloor], Γ(e)=Γ(et+12)\Gamma(e)=\Gamma^{\prime}(e_{\frac{t+1}{2}}) when tt is odd, and Γ(e)=Γ(et2)=Γ(et2+1)\Gamma(e)=\Gamma^{\prime}(e_{\frac{t}{2}})=\Gamma^{\prime}(e_{\frac{t}{2}+1}) when tt is even. It is easy to verify that Γ\Gamma^{\prime} is an MDMD-coloring of GG^{\prime}. Thus, md(G)md(G)+P12md(G^{\prime})\geq md(G)+\left\lfloor\frac{||P||-1}{2}\right\rfloor.  

Lemma 2.3.

Suppose u,vu,v are nonadjacent vertices of GG and Γ\Gamma is an extremal MDMD-coloring of GG. Let CΓ(u,v)={t}C_{\Gamma}(u,v)=\{t\} and ee an extra edge, and let Γ\Gamma^{\prime} be an edge-coloring of GeG\cup e that is obtained from Γ\Gamma by coloring the added edge ee with color tt. Then Γ\Gamma^{\prime} is an MDMD-coloring of GeG\cup e and md(G)=md(Ge)md(G)=md(G\cup e).

Proof.

Let HiH_{i} be the graph obtained from GG by deleting all the edges with color ii. Let G=GeG^{\prime}=G\cup e. If Γ\Gamma^{\prime} is not an MDMD-coloring of GG^{\prime}, then there are two vertices x,yx,y of GG^{\prime} such that CΓ(x,y)=C_{\Gamma^{\prime}}(x,y)=\emptyset. If tCΓ(x,y)t\in C_{\Gamma}(x,y), since x,yx,y are in different components of HtH_{t}, we have tCΓ(x,y)t\in C_{\Gamma^{\prime}}(x,y), a contradiction. If tCΓ(x,y)t\notin C_{\Gamma}(x,y), then let jCΓ(x,y)j\in C_{\Gamma}(x,y). Then there are two components D1,D2D_{1},D_{2} of HjH_{j} such that xV(D1)x\in V(D_{1}) and yV(D2)y\in V(D_{2}). Since jj does not separate x,yx,y in GG^{\prime}, the edge ee connects D1D_{1} and D2D_{2}, say uV(D1)u\in V(D_{1}) and vV(D2)v\in V(D_{2}). Thus, the color jj separates u,vu,v in GG, which contradicts that CΓ(u,v)={t}C_{\Gamma}(u,v)=\{t\}. Therefore, Γ\Gamma^{\prime} is an MDMD-coloring of GG^{\prime}. Since |Γ(G)|=|Γ(G)||\Gamma^{\prime}(G^{\prime})|=|\Gamma(G)| and Γ\Gamma is an extremal MDMD-coloring of GG, we have md(G)md(G)md(G^{\prime})\geq md(G). Since GG is a connected spanning subgraph of GG^{\prime}, by Lemma 1.1 (3) we have md(G)md(G)md(G)\geq md(G^{\prime}). So, md(G)=md(G)md(G)=md(G^{\prime}).  

Suppose Γ\Gamma is an MDMD-coloring of GG and GiG_{i} is the subgraph of GG induced by the set of edges with color ii, which, in what follows, is called the color ii induced subgraph of GG. Then for any component D1D_{1} of GiG_{i} and any component D2D_{2} of GjG_{j}, we have |V(D1)V(D2)|1|V(D_{1})\cap V(D_{2})|\leq 1; otherwise, suppose u,vV(D1)V(D2)u,v\in V(D_{1})\cap V(D_{2}). Then CΓ(u,v)=C_{\Gamma}(u,v)=\emptyset, a contradiction. We use Γ\mathcal{H}_{\Gamma} to denote a hyperedge-colored hypergraph with vertex-set V(G)V(G) and hyperedge-set {V(D)|D is a component of some Gi}\{V(D)\ |\ D\mbox{ is a component of some }G_{i}\}, and the hyperedge FF has color ii if FF corresponds to a component of GiG_{i}. Let HΓH_{\Gamma} be a graph with V(HΓ)=V(G)V(H_{\Gamma})=V(G) and

E(HΓ)={uv|u,v are in the same component of some Gi}.E(H_{\Gamma})=\{uv\ |\ u,v\mbox{ are in the same component of some }G_{i}\}.

Then each hyperedge of Γ\mathcal{H}_{\Gamma} corresponds to a clique of HΓH_{\Gamma}, and any two hyperedges of Γ\mathcal{H}_{\Gamma} (any two cliques of HΓH_{\Gamma}) share at most one vertex. Thus, Γ\mathcal{H}_{\Gamma} is a linear hypergraph. If FF is a hyperedge of Γ\mathcal{H}_{\Gamma} and u,vFu,v\in F, then cΓ(u,v)=1c_{\Gamma}(u,v)=1. According to Lemma 2.3, we have the following result.

Lemma 2.4.

If Γ\Gamma is an extremal MDMD-coloring of GG, then md(G)=md(HΓ)md(G)=md(H_{\Gamma}).

Suppose Γ\Gamma is an MDMD-coloring of GG and 𝒞\mathcal{C} is a hyper kk-cycle of Γ\mathcal{H}_{\Gamma}. Then there is a kk-cycle CC of HΓH_{\Gamma} such that any adjacent edges of CC have different colors. Thus, t3,5t\neq 3,5. Moreover, if k=4k=4, then the opposite hyperedges of 𝒞\mathcal{C} have the same color.

3 Graphs with diameter two

In this section, we show that md(G)2md(G)\leq 2 for a 22-connected graph GG if diam(G)2diam(G)\leq 2. However, for any integer d3d\geq 3, we can construct a 22-connected graph GG such that diam(G)=ddiam(G)=d and md(G)md(G) can be arbitrarily large. Thus, it makes sense to focus on studying the graphs with diameter two, since graphs with diameter 1 are complete graphs and their MDMD-numbers are 1.

Theorem 3.1.

Suppose GG is a graph with diam(G)=2diam(G)=2. Then

  1. 1.

    if GG has a cut-vertex, then md(G)md(G) is equal to the number of blocks of GG;

  2. 2.

    if GG is a 22-connected graph, then md(G)2md(G)\leq 2;

  3. 3.

    if any two nonadjacent vertices of GG has at least two common neighbors, then md(G)2md(G)\leq 2, and the equality holds if and only if G=KsKtG=K_{s}\Box K_{t}, where s,t2s,t\geq 2.

Proof.

The proof of statement (1) goes as follows. If vv is a cut-vertex of GG and diam(G)=2diam(G)=2, then vv connects every vertex of V(Gv)V(G-v). Thus, for each block DD of GG, DvD-v is connected and D=(Dv)vD=(D-v)\vee v, i.e., md(D)=1md(D)=1. Therefore, md(G)md(G) is equal to the number of blocks of GG.

Next, for the proof of statement (2) suppose Γ\Gamma is an MDMD-coloring of GG with |Γ(G)|3|\Gamma(G)|\geq 3. Then each hypercycle (hyperpath) of the above mentioned hypergraph Γ\mathcal{H}_{\Gamma} is a linear hypercycle (linear hyperedge). We now prove that there is a rainbow hyper 33-path (the colors of the three hyperedges are pairwise differently) in Γ\mathcal{H}_{\Gamma}. Since Γ\mathcal{H}_{\Gamma} does not have hyper 33-cycle, the union of three consecutive hyperedges forms a hyper 33-path. If every vertex zz of GG has dc(z)2d^{c}(z)\leq 2, then there is a rainbow hyper 33-path in Γ\mathcal{H}_{\Gamma}. If there is a vertex xx of GG with dc(x)3d^{c}(x)\geq 3, then there are three hyperedges, say D1,D2D_{1},D_{2} and D3D_{3}, such that xx is the common vertex of them. Then the colors of D1,D2D_{1},D_{2} and D3D_{3} are pairwise differently. Since GG is a 22-connected graph, there is a vertex ww of V(D1){x}V(D_{1})-\{x\} with dc(w)2d^{c}(w)\geq 2 (otherwise, xx is a cut-vertex of GG, a contradiction). Then there is a hyperedge FF of GG, such that ww is a common vertex of FF and D1D_{1}. Thus, either FD1D2F\cup D_{1}\cup D_{2} or FD1D3F\cup D_{1}\cup D_{3} is a rainbow hyper 33-path.

Let 𝒫\mathcal{P} be a rainbow hyper 33-path of \mathcal{H} and let V(Di)V(Di+1)={ui}V(D_{i})\cap V(D_{i+1})=\{u_{i}\} for i[2]i\in[2]. Let uV(D1){u1}u\in V(D_{1})-\{u_{1}\} and vV(D3){u2}v\in V(D_{3})-\{u_{2}\}. We use 𝒫u,v\mathcal{P}_{u,v} to denote a minimum hyperpath connecting uu and vv. Since diam(G)=2diam(G)=2, the size of 𝒫u,v\mathcal{P}_{u,v} is either one or two. Let 𝒞=𝒫u,v𝒫\mathcal{C}=\mathcal{P}_{u,v}\cup\mathcal{P}. If 𝒫u,v\mathcal{P}_{u,v} is a hyperedge, then 𝒞\mathcal{C} is a hyper 44-cycle. Since D1D_{1} and D3D_{3} are opposite hyperedges of 𝒞\mathcal{C} and they have different colors, a contradiction. If 𝒫u,v\mathcal{P}_{u,v} is a hyper 22-path, then let F1,F2F_{1},F_{2} be hyperedges of 𝒫u,v\mathcal{P}_{u,v}, and let V(F1)V(F2)={u3}V(F_{1})\cap V(F_{2})=\{u_{3}\}. If u3{u1,u2}u_{3}\notin\{u_{1},u_{2}\}, then 𝒞\mathcal{C} is a hyper 55-cycle, a contradiction. If u3{u1,u2}u_{3}\in\{u_{1},u_{2}\}, then 𝒞\mathcal{C} contains a hyper 33-cycle, a contradiction.

Finally, we show statement (3). It is obvious that diam(G)2diam(G)\leq 2, and GG is a 22-connected graph when n3n\geq 3. So, md(G)2md(G)\leq 2. Suppose G=KsKtG=K_{s}\Box K_{t} and s,t2s,t\geq 2. Then |N(u,v)|=2|N(u,v)|=2 for any nonadjacent vertices uu and vv of GG. By Lemma 1.1 (2) and Theorem 1.3, we have md(G)=md(Ks)+md(Kt)=2md(G)=md(K_{s})+md(K_{t})=2.

Suppose md(G)=2md(G)=2. Then n3n\geq 3 and GG is a 22-connected graph. Let Γ\Gamma be an extremal MDMD-coloring of GG and let G1,G2G_{1},G_{2} be the colors 1,21,2 induced subgraphs of GG, respectively. Since md(G)=2md(G)=2, we have dc(v)2d^{c}(v)\leq 2 for each vV(G)v\in V(G). If dc(v)=1d^{c}(v)=1, by symmetry, suppose vv is in a component DD of G1G_{1}. Since md(G)=2md(G)=2, we have DGD\neq G, i.e., there exists a vertex uu in V(G)V(D)V(G)-V(D). Then u,vu,v are nonadjacent and N(u,v)DN(u,v)\subseteq D. Let {a,b}N(u,v)\{a,b\}\subseteq N(u,v). Since Γ(va)=Γ(vb)=1\Gamma(va)=\Gamma(vb)=1, we have vavbuaubva\cup vb\cup ua\cup ub is a monochromatic 44-cycle, i.e., uV(D)u\in V(D), a contradiction. Thus, dc(v)=2d^{c}(v)=2 for each vV(G)v\in V(G). We use Du1D_{u}^{1} and Du2D_{u}^{2} to denote the components of G1G_{1} and G2G_{2}, respectively, such that V(Du1)V(Du2)=uV(D_{u}^{1})\cup V(D_{u}^{2})=u.

Suppose there are tt components of G1G_{1} and ss components of G2G_{2}. Since GG is a 22-connected graph, we have s,t2s,t\geq 2. Otherwise, if s=1s=1, then for each vertex vv of G1G_{1}, vv is a cut-vertex, a contradiction. We label the tt components of G1G_{1} by the numbers in [t][t] and label the ss components of G2G_{2} by the numbers in [s][s], respectively. We use l1(D)l_{1}(D) to denote the label of a component DD of G1G_{1}, and use l2(F)l_{2}(F) to denote the label of a component FF of G2G_{2}. For a vertex uu of GG, since dc(u)=2d^{c}(u)=2, we use (l1(Du1),l2(Du2))(l_{1}(D_{u}^{1}),l_{2}(D_{u}^{2})) to denote uu. For two vertices u,vu,v of GG, let u=(i,j)u=(i,j) and let v=(s,t)v=(s,t). In order to show G=KsKtG=K_{s}\Box K_{t}, we need to show that uvuv is an edge of GG when i=si=s and jtj\neq t, or isi\neq s and j=tj=t, and u,vu,v are nonadjacent vertices when isi\neq s and jtj\neq t. If isi\neq s and jtj\neq t, then vV(Du1Du2)v\notin V(D_{u}^{1}\cup D_{u}^{2}). Since N(u)V(Du1Du2)N(u)\subseteq V(D_{u}^{1}\cup D_{u}^{2}), u,vu,v are nonadjacent vertices of GG. If, by symmetry, i=si=s and jtj\neq t, then Du1=Dv1D_{u}^{1}=D_{v}^{1}. Let uV(Du2){u}u^{\prime}\in V(D_{u}^{2})-\{u\}. Then u,vu^{\prime},v are nonadjacent. Since N(v)V(Dv1Dv2)N(v)\subseteq V(D_{v}^{1}\cup D_{v}^{2}) and N(u)V(Du1Du2)N(u^{\prime})\subseteq V(D_{u^{\prime}}^{1}\cup D_{u^{\prime}}^{2}), we have

2|N(v,u)||V(Dv1Dv2)V(Du1Du2)|=|Dv1Du2|+|Du1Dv2|2.2\leq|N(v,u^{\prime})|\leq|V(D_{v}^{1}\cup D_{v}^{2})\cap V(D_{u^{\prime}}^{1}\cup D_{u^{\prime}}^{2})|=|D_{v}^{1}\cap D_{u^{\prime}}^{2}|+|D_{u^{\prime}}^{1}\cap D_{v}^{2}|\leq 2.

Thus, Dv1Du2N(v,u)D_{v}^{1}\cap D_{u^{\prime}}^{2}\subseteq N(v,u^{\prime}). Since Dv1Du2={u}D_{v}^{1}\cap D_{u^{\prime}}^{2}=\{u\}, we have uvuv is an edge of GG.  

Remark 1.

Suppose L1,,LrL_{1},\cdots,L_{r} are r(2)r\ (\geq 2) internal disjoint odd paths with an order 2ki+22k_{i}+2 for each i[r]i\in[r], and they have the same ends {u,v}\{u,v\}. Let Li=ue1ix1ie2ix2ix2kiiL_{i}=ue_{1}^{i}x_{1}^{i}e_{2}^{i}x_{2}^{i}\cdots x_{2k_{i}}^{i} e2ki+1ve_{2k_{i}+1}v. Let c0=1c_{0}=1 and ci=Σj=0ikjc_{i}=\Sigma_{j=0}^{i}k_{j}. If ki1k_{i}\geq 1 for i[r]i\in[r], then let Γ\Gamma be an edge-coloring of GG such that Γ(eji)=Γ(e2ki+2ji)=ci1+j\Gamma(e_{j}^{i})=\Gamma(e_{2k_{i}+2-j}^{i})=c_{i-1}+j and Γ(eki+1i)=1\Gamma(e^{i}_{k_{i}+1})=1 for each i[r]i\in[r] and j[ki]j\in[k_{i}]. Then Γ\Gamma is an MDMD-coloring of GG with |Γ(G)|=|G|2|\Gamma(G)|=\frac{|G|}{2}. Since GG is a 22-connected graph, we have md(G)=|G|2md(G)=\frac{|G|}{2}. If ki=1k_{i}=1 for each i[r]i\in[r], then GG is a 22-connected graph with diam(G)=3diam(G)=3 and md(G)=rmd(G)=r. Therefore, there exist 22-connected graphs with diameter three, but their MDMD-numbers can be arbitrarily large.

Let AnA_{n} be a graph with V(An)={v1,,vn2}{u1,,un2}V(A_{n})=\{v_{1},\cdots,v_{\left\lceil\frac{n}{2}\right\rceil}\}\cup\{u_{1},\cdots,u_{\left\lfloor\frac{n}{2}\right\rfloor}\} and E(An)={vivj:i,j[n2]}{uiuj:i,j[n2]}{viui:i[n2]}E(A_{n})=\{v_{i}v_{j}:i,j\in[\left\lceil\frac{n}{2}\right\rceil]\}\cup\{u_{i}u_{j}:i,j\in[\left\lfloor\frac{n}{2}\right\rfloor]\}\cup\{v_{i}u_{i}:i\in[\left\lfloor\frac{n}{2}\right\rfloor]\}. Then {viui:i[n2]}\{v_{i}u_{i}:i\in[\left\lfloor\frac{n}{2}\right\rfloor]\} is a matching-cut of GG. If nn is an odd integer, then let

𝒜n={AnE|E is either an emptyset or a matching of An[{v1,,vn12}]}.\mathcal{A}_{n}=\{A_{n}-E\ |\ E\mbox{ is either an emptyset or a matching of }A_{n}[\{v_{1},\cdots,v_{\frac{n-1}{2}}\}]\}.
Theorem 3.2.

Suppose GG is a n2\left\lfloor\frac{n}{2}\right\rfloor-connected graph and n4n\geq 4. Then md(G)2md(G)\leq 2 and

  1. 1.

    if nn is even, then md(G)=2md(G)=2 if and only if G=AnG=A_{n};

  2. 2.

    if nn is odd, then md(G)=2md(G)=2 if and only if G𝒜nG\in\mathcal{A}_{n}.

Proof.

Since N(x)+N(y)n1N(x)+N(y)\geq n-1 for any two nonadjacent vertices xx and yy, we have diam(G)2diam(G)\leq 2. So, md(G)2md(G)\leq 2.

It is obvious that GG is a n2\left\lfloor\frac{n}{2}\right\rfloor-connected graph if G=AnG=A_{n} or G𝒜nG\in\mathcal{A}_{n}. Moreover, by Lemma 1.4 and Theorem 3.1, we have md(G)=2md(G)=2.

Now suppose GG is a n2\left\lfloor\frac{n}{2}\right\rfloor-connected graph and md(G)=2md(G)=2. Since n4n\geq 4, GG is a 22-connected graph. We distinguish the following cases for our proof.

Case 1. nn is even.

For any two nonadjacent vertices u,vu,v of GG, |N(u)N(v)|2|N(u)\cap N(v)|\geq 2. By Theorem 3.1 (3), G=KsKtG=K_{s}\Box K_{t}, where s,t2s,t\geq 2. We need to prove that at least one of s,ts,t equals two. Suppose H1,H2H_{1},H_{2} are two cliques of order s,ts,t, respectively, and V(H1)V(H2)={u}V(H_{1})\cap V(H_{2})=\{u\}. Then N(u)V(H1H2)N(u)\subseteq V(H_{1}\cup H_{2}), i.e., s+t2n2s+t-2\geq\frac{n}{2}. Since n=stn=st, we have t(s2)2(s2)t(s-2)\leq 2(s-2). Thus, either s=2s=2 or t=2t=2.

Case 2. nn is odd.

Say n=2k+1n=2k+1 for some integer kk. Suppose Γ\Gamma is an extremal MDMD-coloring of GG and G1,G2G_{1},G_{2} are the colors 1,21,2 induced subgraphs, respectively.

Subcase 2.1 Every vertex vv of GG has dc(v)=2d^{c}(v)=2.

Suppose there are components D,FD,F of G1,G2G_{1},G_{2}, respectively, such that V(G)V(F)=V(G)\cap V(F)=\emptyset. Then let uV(D)u\in V(D) and vV(F)v\in V(F). Since dc(u)=dc(v)=2d^{c}(u)=d^{c}(v)=2, there are components DD^{\prime} of G1G_{1} and FF^{\prime} of G2G_{2}, such that V(D)V(F)={u}V(D)\cap V(F^{\prime})=\{u\} and V(F)V(D)={v}V(F)\cap V(D^{\prime})=\{v\}. Since V(D)V(F){u}V(D)\cup V(F^{\prime})-\{u\} and V(D)V(F){v}V(D^{\prime})\cup V(F)-\{v\} are vertex-cuts of GG, we have |V(D)V(F)|k+1|V(D)\cup V(F^{\prime})|\geq k+1 and |V(D)V(F)|k+1|V(D^{\prime})\cup V(F)|\geq k+1. Since |V(D)V(F)|1|V(D^{\prime})\cap V(F^{\prime})|\leq 1, we have n|V(D)V(F)|+|V(D)V(F)||V(D)V(F)|2k+1=nn\geq|V(D)\cup V(F^{\prime})|+|V(D^{\prime})\cup V(F)|-|V(D^{\prime})\cap V(F^{\prime})|\geq 2k+1=n, i.e., DDFF=GD\cup D^{\prime}\cup F\cup F^{\prime}=G. Then uu is a cut-vertex of GG, a contradiction. Therefore, for each component DD of G1G_{1} and each component FF of G2G_{2}, we have |V(G)V(F)|=1|V(G)\cap V(F)|=1. Then since dc(v)=2d^{c}(v)=2 for each vV(G)v\in V(G), any two components of G1G_{1} (and also G2G_{2}) have the same order, say ss (the order is tt). Then s,t>2s,t>2; otherwise, suppose s=2s=2, i.e., G1G_{1} is a matching. Since nn is odd, we have V(G)V(G1)V(G)-V(G_{1})\neq\emptyset. Thus, each vertex vv of V(G)V(G1)V(G)-V(G_{1}) has dc(v)=1d^{c}(v)=1, a contradiction. For a vertex xx of GG, let D1,D2D_{1},D_{2} be the components of G1,G2G_{1},G_{2}, respectively, containing xx. Then D1D2{x}D_{1}\cup D_{2}-\{x\} is a vertex-cut of GG, i.e., s+t2ks+t-2\geq k. However, 2k+1=n=st2k+1=n=st and s,t>3s,t>3, a contradiction.

Subcase 2.2 There is a vertex vv of GG with dc(v)=1d^{c}(v)=1.

Suppose DD is the component of G1G_{1} containing vv. Then since D{v}D-\{v\} is a vertex cut of GG, we have |D|k+1|D|\geq k+1. Since the set of vertices of DD with color-degree two is a vertex-cut of GG, there are at least kk vertices of DD, say v1,,vkv_{1},\cdots,v_{k}, such that dc(vi)=2d^{c}(v_{i})=2 for i[k]i\in[k]. Let FiF_{i} be the component of G2G_{2} containing viv_{i} and let U=i[k](V(Fi){vi})U=\bigcup_{i\in[k]}(V(F_{i})-\{v_{i}\}). Then |U|k|U|\geq k. Since n|D|+|U|2k+1=nn\geq|D|+|U|\geq 2k+1=n, we have |D|=k+1|D|=k+1, |U|=k|U|=k, and |Fi|=2|F_{i}|=2 for i[k]i\in[k]. Moreover, N(v)={v1,,vk}N(v)=\{v_{1},\cdots,v_{k}\}. Let V(Fi){vi}={ui}V(F_{i})-\{v_{i}\}=\{u_{i}\}. For i,j[k]i,j\in[k], if uiuju_{i}u_{j} is not an edge of GG, then U{ui,uj}+vjU-\{u_{i},u_{j}\}+v_{j} is a vertex-cut of GG with order k1k-1, which contradicts that GG is kk-connected. For each viv_{i}, if there are two vertices vj,vlv_{j},v_{l} such that vivjv_{i}v_{j} and vivlv_{i}v_{l} are not edges of GG, then V(D){vi,vj,vl}+uiV(D)-\{v_{i},v_{j},v_{l}\}+u_{i} is a vertex-cut of GG with order k1k-1, which contradicts that GG is kk-connected. Therefore, viv_{i} connects all but at most one vertex of DvD-v. So, G𝒜nG\in\mathcal{A}_{n}.  

4 Upper bounds

In this section, we give two upper bounds of the monochromatic disconnection number of a graph GG, one of which depends on the connectivity of GG, and the other depends on the independent number of GG. Note that for a kk-connected graph GG, when k=2k=2 (small) and kn2k\geq\left\lfloor\frac{n}{2}\right\rfloor (large), from Theorems 1.2 and 3.2 we know that md(G)nkmd(G)\leq\left\lfloor\frac{n}{k}\right\rfloor. This suggests us to make the following conjecture.

Conjecture 4.1.

Suppose GG is a kk-connected graph. Then md(G)nkmd(G)\leq\left\lfloor\frac{n}{k}\right\rfloor.

Suppose PP is a kk-path. Then md(KrP)=md(Kr)+md(P)=k+1md(K_{r}\Box P)=md(K_{r})+md(P)=k+1. Since n=|KrP|=r(k+1)n=|K_{r}\Box P|=r(k+1) and KrPK_{r}\Box P is an rr-connected graph, the bound is sharp for k2k\geq 2 if the conjecture is true.

The mean distance of a connected graph GG is defined as μ(G)=(n2)1Σu,vV(G)d(u,v)\mu(G)={n\choose 2}^{-1}\Sigma_{u,v\in V(G)}d(u,v). Plesnĺk in [14] posed the problem of finding sharp upper bounds on μ(G)\mu(G) for kk-connected graphs. Favaron et al. in [11] proved that if GG is a kk-connected graph of order nn, then

μ(G)n+k1kn1k2n1kn1,\displaystyle\mu(G)\leq\left\lfloor\frac{n+k-1}{k}\right\rfloor\cdot\frac{n-1-\frac{k}{2}\left\lfloor\frac{n-1}{k}\right\rfloor}{n-1}, (1)

and the bound is sharp when nn is even. If nn is odd and k3k\geq 3, then Dankelmann et al. in [10] proved that μ(G)n2k+1+30\mu(G)\leq\frac{n}{2k+1}+30 and this bound is, apart from an additive constant, best possible.

The following result gives a relationship between the monochromatic disconnection number and the connectivity of a graph, which means that if the connectivity of a graph is in linear of the order of the graph, then the monochromatic disconnection number of the graph is upper bounded by a constant.

Theorem 4.2.

For any 0<ε<120<\varepsilon<\frac{1}{2}, there is a constant C=C(ε)<(1+ε)24ε2(1ε)C=C(\varepsilon)<\frac{(1+\varepsilon)^{2}}{4\varepsilon^{2}(1-\varepsilon)}, such that for any εn\varepsilon n-connected graph GG, md(G)Cmd(G)\leq C.

Proof.

Suppose Γ\Gamma is an extremal MDMD-coloring of GG and V(G)={v1,,vn}V(G)=\{v_{1},\cdots,v_{n}\}. We use (i,j)(i,j) to denote an unordered integer pair in this proof. For each color ii of Γ(G)\Gamma(G), let

Si={(j,l): the color i sepatates vj and vl}.S_{i}=\{(j,l):\mbox{ the color }i\mbox{ sepatates }v_{j}\mbox{ and }v_{l}\}.

Then ΣiΓ|Si|=ΣjlcΓ(vj,vl)\Sigma_{i\in\Gamma}|S_{i}|=\Sigma_{j\neq l}c_{\Gamma}(v_{j},v_{l}).

Claim 4.3.

|Si|k(nk)|S_{i}|\geq k(n-k) for each iΓ(G)i\in\Gamma(G).

Proof.

Let εn=k\varepsilon n=k. The result holds obviously for k=1k=1. Thus, let k2k\geq 2. For each iΓ(G)i\in\Gamma(G), let GiG_{i} be the color ii induced subgraph of GG, and let HiH_{i} be the graph obtained from GG by deleting all the edges with color ii. Then HiH_{i} is a disconnected graph. Suppose there is a component DD of HiH_{i} with |D|>nk|D|>n-k. Let U={vj|vjV(D)V(Gi)}U=\{v_{j}\ |\ v_{j}\in V(D)\cap V(G_{i})\}. For a component BB of GiG_{i}, if V(B)V(D)V(B)\cap V(D)\neq\emptyset, then |V(B)V(D)|=1|V(B)\cap V(D)|=1. Since BB contains at least one vertex of V(GD)V(G-D), we have |U||V(GD)|<k|U|\leq|V(G-D)|<k. Since |D|>nk=n(1ε)>εn=k|D|>n-k=n(1-\varepsilon)>\varepsilon n=k, UU is a proper subset of V(D)V(D). So, UU is a vertex-cut of GG. Since |U|<k|U|<k and GG is kk-connected, this yields a contradiction. Thus, for each iΓ(G)i\in\Gamma(G), there is no component of HiH_{i} with order greater than nkn-k.

We partition the components of HiH_{i} into rr parts such that rr is minimum and the number of vertices in each part is at most nkn-k. Suppose the rr parts have n1,,nrn_{1},\cdots,n_{r} vertices, respectively. Then j[r]nj=n\sum_{j\in[r]}n_{j}=n. If r4r\geq 4, then since rr is minimum, nl+nj>nkn_{l}+n_{j}>n-k for each l,j[r]l,j\in[r]. Thus,

n(r1)=(r1)t[r]nt=l,j[r](nl+nj)>(r2)(nk),n(r-1)=(r-1)\sum_{t\in[r]}n_{t}=\sum_{l,j\in[r]}(n_{l}+n_{j})>{r\choose 2}(n-k),

and then r(nk)<2nr(n-k)<2n. Since k<n2k<\frac{n}{2}, this yields a contradiction. Therefore, rr is equal to 2 or 3. If r=2r=2, then |Si|n1n2k(nk)|S_{i}|\geq n_{1}\cdot n_{2}\geq k(n-k). If r=3r=3, then there is an nln_{l} such that knlnkk\leq n_{l}\leq n-k, say l=1l=1. Otherwise, nj<kn_{j}<k for each j[3]j\in[3], then n=j[3]nj<nn=\sum_{j\in[3]}n_{j}<n, a contradiction. Thus, |Si|>n1(n2+n3)n(nk)|S_{i}|>n_{1}\cdot(n_{2}+n_{3})\geq n(n-k).  

By the inequality (1) above, we have

μ(G)\displaystyle\mu(G) n+k1kn1k2n1kn1=n+k1k(1k2(n1)n1k)\displaystyle\leq\left\lfloor\frac{n+k-1}{k}\right\rfloor\cdot\frac{n-1-\frac{k}{2}\left\lfloor\frac{n-1}{k}\right\rfloor}{n-1}=\left\lfloor\frac{n+k-1}{k}\right\rfloor\cdot\left(1-\frac{k}{2(n-1)}\left\lfloor\frac{n-1}{k}\right\rfloor\right)
(n+k1k)[1k2(n1)(n1k1)]\displaystyle\leq\left(\frac{n+k-1}{k}\right)\cdot\left[1-\frac{k}{2(n-1)}\left(\frac{n-1}{k}-1\right)\right]
=n+k1kn+k12(n1)<(n+k)22k(n1).\displaystyle=\frac{n+k-1}{k}\cdot\frac{n+k-1}{2(n-1)}<\frac{(n+k)^{2}}{2k(n-1)}.

Since i,jd(vi,vj)=μ(G)(n2)\sum_{i,j}d(v_{i},v_{j})=\mu(G)\cdot{n\choose 2}, we have i,jd(vi,vj)<(n+k)2n4k\sum_{i,j}d(v_{i},v_{j})<\frac{(n+k)^{2}n}{4k}. It is obvious that d(vi,vj)cΓ(vi,vj)d(v_{i},v_{j})\geq c_{\Gamma}(v_{i},v_{j}) for any two vertices vi,vjv_{i},v_{j} of GG. Thus,

md(G)ΣiΓ|Si|k(nk)=i,jcΓ(vi,vj)k(nk)i,jd(u,v)k(nk)<(n+k)2n4k2(nk)=(1+ε)24ε2(1ε).md(G)\leq\frac{\Sigma_{i\in\Gamma}|S_{i}|}{k(n-k)}=\frac{\sum_{i,j}c_{\Gamma}(v_{i},v_{j})}{k(n-k)}\leq\frac{\sum_{i,j}d(u,v)}{k(n-k)}<\frac{(n+k)^{2}n}{4k^{2}(n-k)}=\frac{(1+\varepsilon)^{2}}{4\varepsilon^{2}(1-\varepsilon)}.

The proof is thus complete.  

Remark 2.

Since ε<12\varepsilon<\frac{1}{2}, we have (1+ε)24ε2(1ε)<(32)2/2ε2=98ε2.\frac{(1+\varepsilon)^{2}}{4\varepsilon^{2}(1-\varepsilon)}<(\frac{3}{2})^{2}/2\varepsilon^{2}=\frac{9}{8\varepsilon^{2}}. This means that when the connectivity of a graph increases, its MDMD-number could decrease, and the upper bound is 44 when ε\varepsilon is getting to 12\frac{1}{2}.

The following result gives a relationship between the monochromatic disconnection number and the independent number of a graph.

Theorem 4.4.

If GG is a 22-connected graph, then md(G)α(G)md(G)\leq\alpha(G). The bound is sharp.

Proof.

Let PP be a path and let t2t\geq 2 be an integer. Since α(KtP)=|P|=md(KtP)\alpha(K_{t}\Box P)=|P|=md(K_{t}\Box P), the bound is sharp if the result holds.

The proof proceeds by induction on the order nn of a graph GG. If n2α(G)n\leq 2\alpha(G), then since GG is a 22-connected graph, md(G)α(G)md(G)\leq\alpha(G). If GG has a vertex vv such that GvG-v is still 22-connected, then by Lemma 1.1 (5), we know md(Gv)md(G)md(G-v)\geq md(G). Since α(Gv)α(G)\alpha(G-v)\leq\alpha(G), by induction, we have md(G)md(Gv)α(Gv)α(G)md(G)\leq md(G-v)\leq\alpha(G-v)\leq\alpha(G). Thus, we only need to consider the graph GG with the property that GvG-v is not a 22-connected graph for any vertex vv of GG.

Let uu be a vertex of GG such that GuG-u has a maximum component. Let ={D1,,Ds}\mathcal{B}=\{D_{1},\cdots,D_{s}\} be the set of components of GuG-u and let DrD_{r} be a maximum component. Let SS be the set of cut-vertices of GuG-u. The block-tree of GuG-u, denoted by TT, is a bipartite graph with bipartition \mathcal{B} and SS, and a block DiD_{i} has an edge with a cut-vertex vv in TT if and only if DiD_{i} contains vv. Then the leaves of TT are blocks, say Dk1,,DklD_{k_{1}},\cdots,D_{k_{l}}. Since GG is 22-connected, there is a vertex viv_{i} of DkiSD_{k_{i}}-S such that uu connects viv_{i} in GG for i[l]i\in[l]. We use Pi,jP_{i,j} to denote the subpath of TT from DkiD_{k_{i}} to DkjD_{k_{j}}. We now prove that TT is a path and DiD_{i} is an edge for iri\neq r. If TT is not a path, then l3l\geq 3. There are two leaves of TT, say Dk1D_{k_{1}} and Dk2D_{k_{2}}, such that DrV(P1,2)D_{r}\in V(P_{1,2}). Then Gv3G-v_{3} has a component containing V(Dr){u}V(D_{r})\cup\{u\}, which contradicts that DrD_{r} is maximum. Thus, TT is a tree. Suppose rjr\neq j and DjD_{j} is not an edge, i.e., DjD_{j} is a 22-connected graph. Since TT is a path, we have W=V(Dj)S{v1,,vl}W=V(D_{j})-S-\{v_{1},\cdots,v_{l}\}\neq\emptyset. Let uWu^{\prime}\in W. Then GuG-u^{\prime} has a component containing V(Dr){u}V(D_{r})\cup\{u\}, which contradicts that DrD_{r} is maximum. Thus, DiD_{i} is an edge for iri\neq r.

Without loss of generality, suppose V(Di)V(Di+1)={ui}V(D_{i})\cap V(D_{i+1})=\{u_{i}\} for i[s1]i\in[s-1]. Then, D1,DsD_{1},D_{s} are leaves of TT, DiD_{i} is an edge for iri\neq r and S={u1,,us1}S=\{u_{1},\cdots,u_{s-1}\}. Let u0V(D1S)u_{0}\in V(D_{1}-S) and usV(DsS)u_{s}\in V(D_{s}-S) be two vertices adjacent to uu.

Let P1=i<rDiP_{1}=\bigcup_{i<r}D_{i} and let P2=i=r+1sDiP_{2}=\bigcup_{i=r+1}^{s}D_{i}. Then P1P_{1} and P2P_{2} are paths. There is an independent set UiU_{i} of PiP_{i} such that UiV(Dr)=U_{i}\cap V(D_{r})=\emptyset and |Ui|=|Pi|12|U_{i}|=\left\lceil\frac{|P_{i}|-1}{2}\right\rceil for i[2]i\in[2]. Let UU be a maximum independent set of DrD_{r}. Then UU1U2U\cup U_{1}\cup U_{2} is an independent set of GuG-u, i.e.,

α(G)\displaystyle\alpha(G) α(Gv)|UU1U2|=α(Dr)+|P1|12+|P2|12\displaystyle\geq\alpha(G-v)\geq|U\cup U_{1}\cup U_{2}|=\alpha(D_{r})+\left\lceil\frac{|P_{1}|-1}{2}\right\rceil+\left\lceil\frac{|P_{2}|-1}{2}\right\rceil
α(Dr)+|P1|+|P2|22=α(Dr)+s12.\displaystyle\geq\alpha(D_{r})+\left\lceil\frac{|P_{1}|+|P_{2}|-2}{2}\right\rceil=\alpha(D_{r})+\left\lceil\frac{s-1}{2}\right\rceil.

Let P={uu0,uus}(irDi)P=\{uu_{0},uu_{s}\}\cup(\bigcup_{i\neq r}D_{i}) and let G=DrPG^{\prime}=D_{r}\cup P. Then PP is an (s+1)(s+1)-path and GG^{\prime} is a 22-connected spanning subgraph of GG. By Lemma 1.1 (3), we have md(G)md(G)md(G)\leq md(G^{\prime}). Let Γ\Gamma be an extremal MDMD-coloring of GG^{\prime}. Then Γ\Gamma is an MDMD-coloring restricted on DrD_{r} and PP. We call DrD_{r} and each edge of PP the joints of GG^{\prime}. Let CC be the set of colors cΓ(G)c\in\Gamma(G^{\prime}) such that cc is in at least two joints of GG^{\prime}. For cCc\in C, we use ncn_{c} to denote the number of joints of GG having edges colored with cc. Then md(G)=|Γ(G)|=|Γ(Dr)|+PΣcC(nc1)md(G^{\prime})=|\Gamma(G^{\prime})|=|\Gamma(D_{r})|+||P||-\Sigma_{c\in C}(n_{c}-1). Since there is a color cc of CΓ(ur1,ur)C_{\Gamma}(u_{r-1},u_{r}) that separates ur1u_{r-1} and uru_{r}, we have cΓ(Dr)Γ(P)c\in\Gamma(D_{r})\cap\Gamma(P). By the same reason, for each eE(P)e\in E(P), either Γ(e)=Γ(f)\Gamma(e)=\Gamma(f) for an edge ff of PeP-e, or Γ(e)Γ(Dr)\Gamma(e)\subseteq\Gamma(D_{r}). Thus, ΣcC(nc1)s+22\Sigma_{c\in C}(n_{c}-1)\geq\left\lceil\frac{s+2}{2}\right\rceil. Therefore,

md(G)\displaystyle md(G) md(G)=|Γ(Dr)|+PΣcC(nc1)\displaystyle\leq md(G^{\prime})=|\Gamma(D_{r})|+||P||-\Sigma_{c\in C}(n_{c}-1)
α(Dr)+s+1s+22=α(Dr)+s2\displaystyle\leq\alpha(D_{r})+s+1-\left\lceil\frac{s+2}{2}\right\rceil=\alpha(D_{r})+\left\lfloor\frac{s}{2}\right\rfloor
=α(Dr)+s12α(G).\displaystyle=\alpha(D_{r})+\left\lceil\frac{s-1}{2}\right\rceil\leq\alpha(G).

The proof is thus complete.  

5 Characterization of extremal graphs

We knew that md(G)=n2md(G)=\left\lfloor\frac{n}{2}\right\rfloor if GG is a 22-connected graph. In this section, we characterize all the 22-connected graphs with MDMD-number n2\left\lfloor\frac{n}{2}\right\rfloor. We use =(L0;L1,,Lt)\mathcal{E}=(L_{0};L_{1},\cdots,L_{t}) to denote an ear-decomposition of GG, where L0L_{0} is a 22-connected subgraph of GG and LiL_{i} is a path for i[t]i\in[t]. Let Z={Li|i>0 and end(Li)V(L0)}Z_{\mathcal{E}}=\{L_{i}\ |\ i>0\mbox{ and }end(L_{i})\subseteq V(L_{0})\}.

If CC is a cycle of GG and vV(G)V(C)v\in V(G)-V(C), then we use κ(v,C)\kappa(v,C) to denote the maximum number of vvivv_{i}-path PiP_{i} of GG, such that V(Pi)V(Pj)={v}V(P_{i})\cap V(P_{j})=\{v\} and V(Pi)V(C)={vi}V(P_{i})\cap V(C)=\{v_{i}\}. We call H=C(i=1κ(v,C)Pi)H=C\cup(\bigcup_{i=1}^{\kappa(v,C)}P_{i}) a (v,C)(v,C)-umbrella of GG (or an umbrella for short) if κ(v,C)3\kappa(v,C)\geq 3. The vertices v1,,vκ(v,C)v_{1},\cdots,v_{\kappa(v,C)} divide CC into κ(v,C)\kappa(v,C) paths, say P1,,Pκ(v,C)P^{\prime}_{1},\cdots,P^{\prime}_{\kappa(v,C)}. We call PiP_{i} a spoke of HH and call PiP^{\prime}_{i} a rim of HH. If the size of each spoke is odd and the size of each rim is even, then we call the (v,C)(v,C)-umbrella a uniform (v,C)(v,C)-umbrella (or uniform umbrella for short).

A graph GG is called a θ\theta-graph if GG is the union of three internal disjoint paths T1,T2T_{1},T_{2} and T3T_{3} with end(T1)=end(T2)=end(T3)end(T_{1})=end(T_{2})=end(T_{3}). If each TiT_{i} is an even path, then we call GG an even θ\theta-graph and call each TiT_{i} a route.

Suppose =(L0;L1,Lt)\mathcal{E}=(L_{0};L_{1},\cdots L_{t}) is an ear-decomposition of GG. Then the concept normal ear-decomposition of GG is defined as follows.

\bullet If |G||G| is even, then \mathcal{E} is a normal ear-decomposition of GG if L0L_{0} is a cycle.

\bullet If |G||G| is odd and GG is not a bipartite graph, then \mathcal{E} is a normal ear-decomposition of GG if L0L_{0} is an odd cycle.

\bullet If |G||G| is odd and GG is a bipartite graph, then \mathcal{E} is a normal ear-decomposition of GG if L0L_{0} is either an umbrella or an even θ\theta-graph. Moreover, if L0L_{0} is an even θ\theta-graph, then for each LiZL_{i}\in Z_{\mathcal{E}}, end(Li)end(L_{i}) is contained in one route.

Lemma 5.1.

If GG is a 22-connected graph, then GG has a normal ear-decomposition.

Proof.

If nn is even or GG is a nonbipartite graph with nn odd, then GG has a normal ear-decomposition. If GG is a bipartite graph and nn is odd, then let ={L0;L1,,Lt}\mathcal{E}=\{L_{0};L_{1},\cdots,L_{t}\} be an ear-decomposition of GG with L0L_{0} an even cycle. Since n=|L0|+Σi[t](|Li|2)n=|L_{0}|+\Sigma_{i\in[t]}(|L_{i}|-2) and nn is odd, there is an even path among the ears, say LiL_{i}. Since H=l=0i1LiH=\bigcup_{l=0}^{i-1}L_{i} is a 22-connected bipartite graph, there is an even cycle CC of HH containing end(Li)end(L_{i}). Moreover, end(Li)end(L_{i}) divides CC into two even paths. So, L0=CLiL^{\prime}_{0}=C\cup L_{i} is an even θ\theta-graph, say the three routes are T1,T2T_{1},T_{2} and T3T_{3}. Let ={L0;L1,,Ls}\mathcal{E}^{\prime}=\{L^{\prime}_{0};L^{\prime}_{1},\cdots,L^{\prime}_{s}\} be an ear-decomposition of GG and let end(Lj)={ui,vi}end(L^{\prime}_{j})=\{u_{i},v_{i}\} for j[s]j\in[s]. If the ends of each LjL^{\prime}_{j} in ZZ_{\mathcal{E}^{\prime}} are contained in one route, then \mathcal{E}^{\prime} is a normal ear-decomposition of GG. Otherwise, suppose LjZL^{\prime}_{j}\in Z_{\mathcal{E}^{\prime}}, ujI(T1)u_{j}\in I(T_{1}) and vjI(T2)v_{j}\in I(T_{2}). Then κ(uj,T2T3)3\kappa(u_{j},T_{2}\cup T_{3})\geq 3, i.e., there is a (uj,T2T3)(u_{j},T_{2}\cup T_{3})-umbrella, say MM. Then there is a normal ear-decomposition of GG containing MM.  

Lemma 5.2.

Suppose GG is a 22-connected graph with md(G)=n2md(G)=\left\lfloor\frac{n}{2}\right\rfloor. Let =(L0;L1,,Lt)\mathcal{E}=(L_{0};L_{1},\cdots,L_{t}) be an ear-decomposition of GG with L0L_{0} a 22-connected subgraph of GG and end(Li)={ai,bi}end(L_{i})=\{a_{i},b_{i}\} for i[t]i\in[t]. Then we have the following results.

  1. 1.

    If HH is a 22-connected subgraph of GG, then each extremal MDMD-coloring of GG is an extremal MDMD-coloring restricted on HH, and md(H)=|H|2.md(H)=\left\lfloor\frac{|H|}{2}\right\rfloor.

  2. 2.

    If nn is even, then GG is a bipartite graph and LiL_{i} is an odd path for i[t]i\in[t].

  3. 3.

    If nn is odd, then when |L0||L_{0}| is even, exact one of {L1,,Lt}\{||L_{1}||,\cdots,||L_{t}||\} is even; when |L0||L_{0}| is odd, LiL_{i} is an odd path for i[t]i\in[t].

Proof.

Let Γ\Gamma be an extremal MDMD-coloring of GG. Then for each i[t]i\in[t], Γ(Li)Γ(l=0i1Ll)\Gamma(L_{i})\cap\Gamma(\bigcup_{l=0}^{i-1}L_{l})\neq\emptyset; otherwise, CΓ(ai,bi)=C_{\Gamma}(a_{i},b_{i})=\emptyset, a contradiction. Moreover, each color of Γ(Li)Γ(l=0i1Ll)\Gamma(L_{i})-\Gamma(\bigcup_{l=0}^{i-1}L_{l}) is used on at least two edges of LiL_{i}. Otherwise, suppose pΓ(Li)Γ(l=0i1Ll)p\in\Gamma(L_{i})-\Gamma(\bigcup_{l=0}^{i-1}L_{l}) and color pp is only used on one edge e=xye=xy of LiL_{i}. Then since Γ(l=0iLl)e\Gamma(\bigcup_{l=0}^{i}L_{l})-e is connected, CΓ(x,y)=C_{\Gamma}(x,y)=\emptyset, a contradiction. Therefore,

n2\displaystyle\left\lfloor\frac{n}{2}\right\rfloor =md(G)=|Γ(L0)|+Σi=1t|Γ(Li)Γ(l=0i1Ll)|\displaystyle=md(G)=|\Gamma(L_{0})|+\Sigma_{i=1}^{t}|\Gamma(L_{i})-\Gamma(\bigcup_{l=0}^{i-1}L_{l})|
md(L0)+Σi=1tLi12\displaystyle\leq md(L_{0})+\Sigma_{i=1}^{t}\left\lfloor\frac{||L_{i}||-1}{2}\right\rfloor
|L0|2+Σi=1tLi12\displaystyle\leq\left\lfloor\frac{|L_{0}|}{2}\right\rfloor+\Sigma_{i=1}^{t}\left\lfloor\frac{||L_{i}||-1}{2}\right\rfloor
|L0|2+Σi[t]Li12=n2.\displaystyle\leq\left\lfloor\frac{|L_{0}|}{2}+\Sigma_{i\in[t]}\frac{||L_{i}||-1}{2}\right\rfloor=\left\lfloor\frac{n}{2}\right\rfloor.

Then |Γ(L0)|=md(L0)=|L0|2|\Gamma(L_{0})|=md(L_{0})=\left\lfloor\frac{|L_{0}|}{2}\right\rfloor and |Γ(Li)|=Li12|\Gamma(L_{i})|=\left\lfloor\frac{||L_{i}||-1}{2}\right\rfloor for each i[t]i\in[t]. So, Γ\Gamma is an extremal MDMD-coloring restricted on L0L_{0}, and md(L0)=|L0|2md(L_{0})=\left\lfloor\frac{|L_{0}|}{2}\right\rfloor. Moreover, |Γ(Li)Γ(l=0i1Ll)|=1|\Gamma(L_{i})\cap\Gamma(\bigcup_{l=0}^{i-1}L_{l})|=1 when LiL_{i} is an odd path.

If GG is not a bipartite graph, nn is even and L0L_{0} an odd cycle, then the above inequality does not hold. Thus, GG is a bipartite graph when nn is even. Moreover, LiL_{i} is an odd path for each i[t]i\in[t]. If nn and |L0||L_{0}| are odd, then LiL_{i} is an odd path for i[t]i\in[t]. If nn is odd and |L0||L_{0}| is even, then exact one of {L1,,Lt}\{||L_{1}||,\cdots,||L_{t}||\} is even.  

For a normal ear-decomposition ={L0;L1,,Lt}\mathcal{E}=\{L_{0};L_{1},\cdots,L_{t}\} of a 22-connected graph GG, if L0L_{0} is an odd cycle and LiZL_{i}\in Z_{\mathcal{E}}, then end(Li)end(L_{i}) divides L0L_{0} into an odd path and an even path, which are denoted by fo(,i)f_{o}(\mathcal{E},i) and fe(,i)f_{e}(\mathcal{E},i), respectively. If L0L_{0} is an even cycle, LiZL_{i}\in Z_{\mathcal{E}} and eE(L0)e\in E(L_{0}), then we use g(,i,e)g(\mathcal{E},i,e) to denote the subpath of L0L_{0} with ends end(Li)end(L_{i}) and g(,i,e)g(\mathcal{E},i,e) contains ee. We define a function f(,i,j)f(\mathcal{E},i,j) for 0i<jt0\leq i<j\leq t as follows.

f(,i,j)={fo(,j)i=0,LjZ and L0 is an odd cycle;g(,i,e)i=0,LjZ and L0 is an even cycle with eE(L0);ajPbji=0,LjZ,L0 is an umbrella,P is either a spoke or a rim of L0 such that end(Lj)V(P);ajTbji=0,LjZ,L0 is an even θ-graph,T is one of the three routes such that end(Li)V(T);ajLibji>0 and end(Lj)V(Li);K4otherwise.f(\mathcal{E},i,j)=\begin{cases}f_{o}(\mathcal{E},j)&i=0,L_{j}\in Z_{\mathcal{E}}\mbox{ and }L_{0}\mbox{ is an odd cycle};\\ g(\mathcal{E},i,e)&i=0,L_{j}\in Z_{\mathcal{E}}\mbox{ and }L_{0}\mbox{ is an even cycle with }e\in E(L_{0});\\ a_{j}Pb_{j}&i=0,L_{j}\in Z_{\mathcal{E}},L_{0}\mbox{ is an umbrella},P\mbox{ is either a spoke}\mbox{ or a rim of }\\ &L_{0}\mbox{ such that }end(L_{j})\subseteq V(P);\\ a_{j}Tb_{j}&i=0,L_{j}\in Z_{\mathcal{E}},L_{0}\mbox{ is an even }\theta\mbox{-graph},\ T\mbox{ is one of the three }\\ &\mbox{routes such that }end(L_{i})\subseteq V(T);\\ a_{j}L_{i}b_{j}&i>0\mbox{ and }end(L_{j})\subseteq V(L_{i});\\ K_{4}&otherwise.\end{cases}

If L0L_{0} is not an even cycle, then the function depends only on ,i\mathcal{E},i and jj. If L0L_{0} is an even cycle and i=0i=0, then the function also depends on ee. Thus, we need to fix an edge ee of L0L_{0} in advance if L0L_{0} is an even cycle.

Lemma 5.3.

If GG is a uniform umbrella or an even θ\theta-graph other than K2,3K_{2,3}, then |G||G| is odd and md(G)=|G|2md(G)=\left\lfloor\frac{|G|}{2}\right\rfloor.

Proof.

It is obvious that |G||G| is odd. Fix an integer k3k\geq 3. Suppose GG^{\prime} is either a minimum even θ\theta-graph other than K2,3K_{2,3}, or a minimum uniform umbrella with kk spokes.

If GG^{\prime} is a minimum even θ\theta-graph other than K2,3K_{2,3}, then GG^{\prime} and one of its extremal MDMD-colorings are depicted in Figure 1 (1), which implies md(G)=3=|G|2md(G^{\prime})=3=\left\lfloor\frac{|G^{\prime}|}{2}\right\rfloor.

If GG^{\prime} is a minimum uniform umbrella with kk spokes, then each spoke is an edge and each rim is a 22-path. Suppose the kk spokes are e1=vv1,,ek=vvke_{1}=vv_{1},\cdots,e_{k}=vv_{k}, and the kk rims are P1=v1f1u1f2v2,,Pk=vkf2k1ukf2kv1P_{1}=v_{1}f_{1}u_{1}f_{2}v_{2},\cdots,P_{k}=v_{k}f_{2k-1}u_{k}f_{2k}v_{1}. We color each eie_{i} with ii. The colors of the edges of PiP_{i} obey the rule that opposite edges of any 44-cycle have the same color (see Figure 1).

Refer to caption
Figure 1: Extremal MDMD-colorings of the minimum even θ\theta-graph and the minimum uniform umbrella.

Since k3k\geq 3, we know that for v1v_{1}, {e1,f2,f2k1}\{e_{1},f_{2},f_{2k-1}\} is a monochromatic v1vv_{1}v-cut (it is also a monochromatic v1viv_{1}v_{i}-cut for i1i\neq 1, and a monochromatic v1uiv_{1}u_{i}-cut for i{1,2,k}i\neq\{1,2,k\}), {e2,f1,f4}\{e_{2},f_{1},f_{4}\} is a monochromatic v1u1v_{1}u_{1}-cut and {ek,f2k,f2k3}\{e_{k},f_{2k},f_{2k-3}\} is a monochromatic v1ukv_{1}u_{k}-cut. By symmetry, the edge-coloring is an MDMD-coloring of GG^{\prime} with kk colors. Since GG^{\prime} is 22-connected and |G|=2k+1|G^{\prime}|=2k+1, we have md(G)=k=|G|2md(G^{\prime})=k=\left\lfloor\frac{|G^{\prime}|}{2}\right\rfloor.

Suppose GG is a uniform umbrella with kk spokes (an even θ\theta-graph other than K2,3K_{2,3}). Then GG is obtained from GG^{\prime} by replacing some edges with odd paths, respectively. W.l.o.g., suppose GG is obtained from GG^{\prime} by replacing one edge with an odd path PP. Then by Lemma 2.2, we have md(G)md(G)+P12=|G|2md(G)\geq md(G^{\prime})+\left\lfloor\frac{||P||-1}{2}\right\rfloor=\left\lfloor\frac{|G|}{2}\right\rfloor, i.e., md(G)=|G|2md(G)=\left\lfloor\frac{|G|}{2}\right\rfloor. The proof is thus complete.  

Lemma 5.4.

If GG is a bipartite graph of odd order and md(G)=n2md(G)=\left\lfloor\frac{n}{2}\right\rfloor, then each umbrella of GG is a uniform umbrella.

Proof.

Suppose GG is a bipartite graph of odd order and md(G)=n2md(G)=\left\lfloor\frac{n}{2}\right\rfloor. Let HH be a (v,C)(v,C)-umbrella of GG. We show that HH is a uniform umbrella.

If κ(v,C)=3\kappa(v,C)=3, then let R1,R2R_{1},R_{2} and R3R_{3} be spokes of HH and RiR_{i} be a vvivv_{i}-path. Then CC is divided into three paths by vertices v1,v2v_{1},v_{2} and v3v_{3} (say, the three paths are W1,W2W_{1},W_{2} and W3W_{3}, such that end(W1)={v1,v2}end(W_{1})=\{v_{1},v_{2}\}, end(W2)={v2,v3}end(W_{2})=\{v_{2},v_{3}\} and end(W3)={v1,v3}end(W_{3})=\{v_{1},v_{3}\}). If each RiR_{i} is an odd path, then since GG is a bipartite graph, each WiW_{i} is an even path, HH be a uniform (v,C)(v,C)-umbrella of GG. If, by symmetry, R1R_{1} is an even path and R2,R3R_{2},R_{3} are odd paths, then W1,W3W_{1},W_{3} are odd paths and W2W_{2} is an even path. Then since (W1W3R2R3;R1,W2)(W_{1}\cup W_{3}\cup R_{2}\cup R_{3};R_{1},W_{2}) is an ear-decomposition of HH containing even paths R1R_{1} and W2W_{2}, by Lemma 5.2 (1) and (3) this yields a contradiction. If, by symmetry, R1R_{1} is an odd path and R2,R3R_{2},R_{3} are even paths, then HH is a uniform (v1,R2R3W2)(v_{1},R_{2}\cup R_{3}\cup W_{2})-umbrella. If each RiR_{i} is an even path, then (C;R1R2,R3)(C;R_{1}\cup R_{2},R_{3}) is an ear-decomposition of HH containing two even paths, a contradiction.

If κ(v,C)4\kappa(v,C)\geq 4, then let W1,W2,W3,W4W_{1},W_{2},W_{3},W_{4} be four spokes of HH (let WiW_{i} be a vvivv_{i} path for i[4]i\in[4]). Then CC is divided into two paths by v2v_{2} and v3v_{3} (say, the two paths are Y1Y_{1} and Y2Y_{2}). W.l.o.g., suppose W1W_{1} is an even path. Then (Y1W2W3;Y2,W4,W1)(Y_{1}\cup W_{2}\cup W_{3};Y_{2},W_{4},W_{1}) is an ear-decomposition of HH. Since md(H)=|H|2md(H)=\left\lfloor\frac{|H|}{2}\right\rfloor and W1W_{1} is an even path, by Lemma 5.2 (4), Y2Y_{2} is an odd path. Since HH is a bipartite graph, either W2W_{2} or W3W_{3} is an even path (say W2W_{2}). Then (CW3W4;W1,W2)(C\cup W_{3}\cup W_{4};W_{1},W_{2}) is an ear-decomposition of HH containing two even paths, a contradiction. So, each spoke of HH is an odd path. Since HH is a bipartite graph, each rim of HH is an even path.  

Suppose =(L0;L1,Lt)\mathcal{E}=(L_{0};L_{1},\cdots L_{t}) is an ear-decomposition of GG. Then \mathcal{E} can have the following possible properties.

Q: If end(Lj)I(Li)end(L_{j})\cap I(L_{i})\neq\emptyset, then end(Lj)V(Li)end(L_{j})\subseteq V(L_{i}).

R: If end(Lj)I(f(,k,i))end(L_{j})\cap I(f(\mathcal{E},k,i))\neq\emptyset, then f(,k,j)f(\mathcal{E},k,j) is a proper subpath of f(,k,i)f(\mathcal{E},k,i).

The concept standard ear-decomposition of GG is defined as follows.

\bullet If |G||G| is even, then \mathcal{E} is a standard ear-decomposition of GG if L0L_{0} is an even cycle.

\bullet If |G||G| is odd and GG is not a bipartite graph, then \mathcal{E} is a standard ear-decomposition of GG if L0L_{0} is an odd cycle and fe(,i)fe(,j)f_{e}(\mathcal{E},i)\cap f_{e}(\mathcal{E},j)\neq\emptyset for Li,LjZL_{i},L_{j}\in Z_{\mathcal{E}}.

\bullet If |G||G| is odd and GG is a bipartite graph, then \mathcal{E} is a standard ear-decomposition of GG if L0L_{0} is either a uniform umbrella or a even θ\theta-graph other than K2,3K_{2,3}. Moreover, for each LiZL_{i}\in Z_{\mathcal{E}}, if L0L_{0} is a uniform umbrella, then end(Li)end(L_{i}) is contained in either a rim or a spoke; if L0L_{0} is an even θ\theta-graph other than K2,3K_{2,3}, then end(Li)end(L_{i}) is contained in one route.

Therefore, a standard ear-decomposition of GG is also a normal ear-decomposition of GG.

Lemma 5.5.

If =(L0;L1,,Lt)\mathcal{E}=(L_{0};L_{1},\cdots,L_{t}) is a standard ear-decomposition of GG and \mathcal{E} has properties Q and R, then there exist integers 0k<rt0\leq k<r\leq t such that end(Lr)V(Lk)end(L_{r})\subseteq V(L_{k}), and d(u)=2d(u)=2 for each uI(f(,k,r))I(Lr)u\in I(f(\mathcal{E},k,r))\cup I(L_{r}).

Proof.

For i[t]i\in[t], let end(Li)={ai,bi}end(L_{i})=\{a_{i},b_{i}\}. We use mrm_{r} (nrn_{r}) to demote the minimum integer such that arV(Lmr)a_{r}\in V(L_{m_{r}}) (brV(Lnr)b_{r}\in V(L_{n_{r}})). Since I(L0)=V(L0)I(L_{0})=V(L_{0}), we have aiI(Lmr)a_{i}\in I(L_{m_{r}}) and brI(Lnr)b_{r}\in I(L_{n_{r}}). Since \mathcal{E} has property Q, we know for each i[t]i\in[t], either end(Li)V(Lmi)end(L_{i})\subseteq V(L_{m_{i}}), or end(Li)V(Lni)end(L_{i})\subseteq V(L_{n_{i}}). Let lil_{i} be the minimum integer such that end(Li)V(Lli)end(L_{i})\subseteq V(L_{l_{i}}).

Let DD be a digraph with vertex-set V(D)={s0,s1,,st}V(D)=\{s_{0},s_{1},\cdots,s_{t}\} and arc-set A(D)={(si,sj)|f(,i,j)K4}A(D)=\{(s_{i},s_{j})\ |\ f(\mathcal{E},i,j)\neq K_{4}\}. We use djd_{j} to denote the length of a minimum directed path from s0s_{0} to sjs_{j}. If end(Lj)I(Li)end(L_{j})\cap I(L_{i})\neq\emptyset, then dj=di+1d_{j}=d_{i}+1. Let U={j|dj is maximum}U=\{j\ |\ d_{j}\mbox{ is maximum}\}. If jUj\in U, then dG(u)=2d_{G}(u)=2 for each uI(Lj)u\in I(L_{j}).

Let ii be an integer in UU such that |f(,li,i)||f(\mathcal{E},l_{i},i)| is minimum. If there is a vertex vv of I(f(,li,i))I(f(\mathcal{E},l_{i},i)) such that dG(v)3d_{G}(v)\geq 3, then there is a path LkL_{k} such that vend(Lk)I(f(,li,i))v\in end(L_{k})\cap I(f(\mathcal{E},l_{i},i)). Since \mathcal{E} has property R, f(,li,k)f(\mathcal{E},l_{i},k) is a proper subpath of f(,li,i)f(\mathcal{E},l_{i},i), i.e., |f(,li,k)|<|f(,li,i)||f(\mathcal{E},l_{i},k)|<|f(\mathcal{E},l_{i},i)|. Since |f(,li,i)||f(\mathcal{E},l_{i},i)| is minimum, we have kUk\notin U. Then there is a path, say LpL_{p}, such that end(Lp)I(Lk)end(L_{p})\cap I(L_{k})\neq\emptyset. Thus, dp>dk=did_{p}>d_{k}=d_{i}, a contradiction. Hence, dG(u)=2d_{G}(u)=2 for each uI(f(,li,i))u\in I(f(\mathcal{E},l_{i},i)).  

Theorem 5.6.

Suppose GG is a 22-connected graph and =(L0;L1,Lt)\mathcal{E}=(L_{0};L_{1},\cdots L_{t}) is a normal ear-decomposition of GG. Then md(G)=n2md(G)=\left\lfloor\frac{n}{2}\right\rfloor if and only if \mathcal{E} is a standard ear-decomposition of GG that has properties Q and R, LiL_{i} is an odd path for each i[t]i\in[t], and f(,i,j)f(\mathcal{E},i,j) is an odd path if f(,i,j)K4f(\mathcal{E},i,j)\neq K_{4}.

Proof.

For i[t]i\in[t], let end(Li)={ai,bi}end(L_{i})=\{a_{i},b_{i}\}.

For the necessity, suppose md(G)=n2md(G)=\left\lfloor\frac{n}{2}\right\rfloor. If nn is even, then L0L_{0} is an even cycle. By Lemma 5.2 (2), GG is a bipartite graph and LiL_{i} is an odd path for i[t]i\in[t]. Since f(,i,j)Ljf(\mathcal{E},i,j)\cup L_{j} is an even cycle, f(,i,j)f(\mathcal{E},i,j) is an odd path. If nn is odd, then since \mathcal{E} is normal, |L0||L_{0}| is odd. By Lemma 5.2 (4), LiL_{i} is an odd path for i[t]i\in[t]. Suppose there are integers i,ji,j such that f(,i,j)f(\mathcal{E},i,j) is an even path. If i=0i=0 and L0L_{0} is an odd cycle, then f(,i,j)=fo(i,j)f(\mathcal{E},i,j)=f_{o}(i,j) is an odd path, a contradiction. If i>0i>0 and L0L_{0} is an odd cycle, then H=Lj(c=0iLc)H=L_{j}\cup(\bigcup_{c=0}^{i}L_{c}) is a 22-connected subgraph of GG and (L0;L1,Li1,LiLjI(f(,i,j)),f(,i,j))(L_{0};L_{1}\cdots,L_{i-1},L_{i}\cup L_{j}-I(f(\mathcal{E},i,j)),f(\mathcal{E},i,j)) is an ear-decomposition of HH with L0L_{0} an odd cycle and f(,i,j)f(\mathcal{E},i,j) an even path, and by Lemma 5.2 (1) and (3) this yields a contradiction. If L0L_{0} is an umbrella or an even θ\theta-graph other than K2,3K_{2,3}, then GG is a bipartite graph. Since f(,i,j)Ljf(\mathcal{E},i,j)\cup L_{j} is an even cycle and LjL_{j} is an odd path, f(,i,j)f(\mathcal{E},i,j) is an odd path, a contradiction. Thus, f(,i,j)f(\mathcal{E},i,j) is an odd path if nn is odd.

We need to prove that \mathcal{E} is standard and \mathcal{E} has properties Q and R below.

Claim 5.7.

\mathcal{E} is standard.

Proof.

If nn is even, then since GG is a bipartite graph, L0L_{0} is an even cycle. Thus, \mathcal{E} is standard.

If GG is not a bipartite graph and nn is odd, then L0L_{0} is an odd cycle. Suppose \mathcal{E} is not a standard ear-decomposition of GG. Then there are paths LiL_{i} and LjL_{j} of ZZ_{\mathcal{E}} such that E(fe(,i))E(fe(,j))=E(f_{e}(\mathcal{E},i))\cap E(f_{e}(\mathcal{E},j))=\emptyset. Let D=LiLj[L0I(fe(,i)fe(,j))]D=L_{i}\cup L_{j}\cup[L_{0}-I(f_{e}(\mathcal{E},i)\cup f_{e}(\mathcal{E},j))]. Then DD is 22-connected subgraph of L0LjLiL_{0}\cup L_{j}\cup L_{i}. Since (D;fe(,i),fe(,j))(D;f_{e}(\mathcal{E},i),f_{e}(\mathcal{E},j)) is an ear-decomposition of L0LiLjL_{0}\cup L_{i}\cup L_{j} and fe(,i),fe(,j)f_{e}(\mathcal{E},i),f_{e}(\mathcal{E},j) are even paths, by Lemma 5.2 (1) and (3) this yields a contradiction. Thus, \mathcal{E} is standard.

If GG is a bipartite graph, nn is odd and L0L_{0} is an even θ\theta-graph, then L0K2,3L_{0}\neq K_{2,3}. Otherwise L0L_{0} is a 22-connected subgraph of GG with md(L0)=1<|L0|2md(L_{0})=1<\left\lfloor\frac{|L_{0}|}{2}\right\rfloor, and by Lemma 5.2 (1) this yields a contradiction. Thus, \mathcal{E} is standard.

If GG is a bipartite graph, nn is odd and L0L_{0} is an umbrella, then suppose the rims of L0L_{0} are W1,,WkW_{1},\cdots,W_{k}, where k3k\geq 3 and WiW_{i} is a vivi+1v_{i}v_{i+1}-path for i[k1]i\in[k-1]. Suppose the spokes are R1,,RkR_{1},\cdots,R_{k}, where RiR_{i} is a vvivv_{i}-path. Let C=i[k]WiC=\bigcup_{i\in[k]}W_{i}. Since md(G)=n2md(G)=\left\lfloor\frac{n}{2}\right\rfloor, by Lemma 5.4, L0L_{0} is a uniform umbrella, i.e., each WiW_{i} is an even path and each RiR_{i} is an odd path. Suppose there is a path LiL_{i} of ZZ_{\mathcal{E}} such that end(Li)end(L_{i}) is neither contained in any spoke nor contained in any rim. If aiI(Rj)a_{i}\in I(R_{j}) and biV(L0)V(Rj)b_{i}\in V(L_{0})-V(R_{j}), then aia_{i} divides RjR_{j} into two subpaths Rj1=vLjaiR^{1}_{j}=vL_{j}a_{i} and Rj2=aiLjvjR^{2}_{j}=a_{i}L_{j}v_{j}. Since k3k\geq 3, w.l.o.g., let bjI(Wk)b_{j}\notin I(W_{k}). Then Hs=RjsLi(lkWl)(ljRl)H_{s}=R^{s}_{j}\cup L_{i}\cup(\bigcup_{l\neq k}W_{l})\cup(\bigcup_{l\neq j}R_{l}) is a 22-connected graph for s[2]s\in[2]. Since LjL_{j} is an odd path, one of Rj1R^{1}_{j} and Rj2R^{2}_{j} is an even path, say Rj1R^{1}_{j}. Since (H2;Wk,Rj1)(H_{2};W_{k},R^{1}_{j}) is an ear-decomposition of L0LiL_{0}\cup L_{i} and Wk,Rj1W_{k},R^{1}_{j} are even paths, by Lemma 5.2 (1) and (3) this yields a contradiction. If end(Li)V(C)end(L_{i})\subseteq V(C), then since GG is a bipartite graph, LiL_{i} is an odd path and each WjW_{j} is an even path, we have |end(Li){v1,,vk}|1|end(L_{i})\cap\{v_{1},\cdots,v_{k}\}|\leq 1. Therefore, there is a rim WjW_{j} such that aia_{i} divides WjW_{j} into two odd paths Wj1=vjWjaiW^{1}_{j}=v_{j}W_{j}a_{i} and Wj2=aiWjvj+1W^{2}_{j}=a_{i}W_{j}v_{j+1}. (w.l.o.g., suppose 1j<k1\leq j<k). Since there is no rim containing end(Li)end(L_{i}), we have biV(Wj)b_{i}\notin V(W_{j}). Note that end(Li)end(L_{i}) divides CC into two subpaths C1C^{1} and C2C^{2} such that vjV(C1)v_{j}\in V(C^{1}) and vj+1V(C2)v_{j+1}\in V(C^{2}). Since k3k\geq 3, by symmetry, suppose |C1{v1,,vk}|2|C^{1}\cap\{v_{1},\cdots,v_{k}\}|\geq 2. Then there is an integer l[k]{i+1}l\in[k]-\{i+1\} such that C1C^{1} contains viv_{i} and vlv_{l}. Then there is an ear-decomposition (C;P1,P2,)(C^{\prime};P^{\prime}_{1},P^{\prime}_{2},\cdots) of L0LiL_{0}\cup L_{i} such that C=C1Li,P1=RiRlC^{\prime}=C^{1}\cup L_{i},P^{\prime}_{1}=R_{i}\cup R_{l} and P2=Wi2Ri+1P^{\prime}_{2}=W^{2}_{i}\cup R_{i+1}. Since P1P^{\prime}_{1} and P2P^{\prime}_{2} are even paths, by Lemma 5.2 (3) this yields a contradiction. Thus \mathcal{E} is standard.  

Claim 5.8.

\mathcal{E} has property Q.

Proof.

Let mim_{i} (nin_{i}) be the minimum integer such that aiV(Lmi)a_{i}\in V(L_{m_{i}}) (biV(Lni)b_{i}\in V(L_{n_{i}})). Since I(L0)=V(L0)I(L_{0})=V(L_{0}), we have aiI(Lmi)a_{i}\in I(L_{m_{i}}) and biI(Lni)b_{i}\in I(L_{n_{i}}). Let lil_{i} be an integer such that end(Li)I(Lli)end(L_{i})\cap I(L_{l_{i}})\neq\emptyset.

Suppose \mathcal{E} does not have property Q. Then there are integers 0j<rt0\leq j<r\leq t such that arI(Lj)a_{r}\in I(L_{j}) and brV(Lj)b_{r}\notin V(L_{j}). Since brI(Lnr)b_{r}\in I(L_{n_{r}}), by symmetry, suppose j>lnrj>l_{n_{r}}. For convenience, let lnr=il_{n_{r}}=i. Since LjL_{j} is an odd path, let ajLjara_{j}L_{j}a_{r} be an even path. Let l=max{mj,nj,nr}l=\max\{m_{j},n_{j},n_{r}\} and H=LjLr(h=0lLh)H=L_{j}\cup L_{r}\cup(\bigcup_{h=0}^{l}L_{h}). Then HH is a 22-connected graph with an ear-decomposition (L0;L1,,Ll,arLjbjLr,ajLjar)(L_{0};L_{1},\cdots,L_{l},a_{r}L_{j}b_{j}\cup L_{r},a_{j}L_{j}a_{r}). If L0L_{0} is an odd cycle, or a uniform umbrella, or an even θ\theta-graph other than K2,3K_{2,3}, then since |L0||L_{0}| is odd and ajLjara_{j}L_{j}a_{r} is an even path, by Lemma 5.2 (1) and (3) this yields a contradiction. If L0L_{0} is an even cycle, then by Lemma 5.2 (1) and (2) this yields a contradiction.  

Claim 5.9.

\mathcal{E} has property R.

Proof.

If \mathcal{E} does not have property R, then there are integers r,i,jr,i,j such that end(Lj)I(f(,r,i))end(L_{j})\cap I(f(\mathcal{E},r,i))\neq\emptyset and f(,r,j)f(\mathcal{E},r,j) is not a subpath of f(,r,i)f(\mathcal{E},r,i). Since \mathcal{E} has property Q, f(,r,j)f(\mathcal{E},r,j) is a subpath of LrL_{r}. Then end(Li)end(L_{i}) and end(Lj)end(L_{j}) appear alternately on L=f(,r,i)f(,r,j)L=f(\mathcal{E},r,i)\cup f(\mathcal{E},r,j), say ai,aj,bi,bja_{i},a_{j},b_{i},b_{j} are consecutively on LL. Here, LL is a subpath of the path LrL_{r} if r>0r>0; LL is a subpath of either a rim or a spoke of LrL_{r} if r=0r=0 and L0L_{0} is a uniform umbrella; LL is a subpath of a route if r=0r=0 and L0L_{0} is an even θ\theta-graph other than K2,3K_{2,3}; LL is a subpath of a cycle LrL_{r} if r=0r=0 and L0L_{0} is a cycle. Let W1=aiLaj,W2=ajLbiW^{1}=a_{i}La_{j},W^{2}=a_{j}Lb_{i} and W3=biLbjW^{3}=b_{i}Lb_{j}. Since f(,r,i)f(\mathcal{E},r,i) and f(,r,j)f(\mathcal{E},r,j) are odd paths, either W1,W3W^{1},W^{3} are even paths and W2W^{2} is an odd path, or W2W^{2} is an even path and W1,W3W^{1},W^{3} are odd paths. Let H=(l=0rLl)LiLjH=(\bigcup_{l=0}^{r}L_{l})\cup L_{i}\cup L_{j}.

Suppose W1,W3W^{1},W^{3} are even paths and W2W^{2} is an odd path. Let HH^{\prime} be a graph obtained from HH by removing W1W^{1} and W3W^{3}. Then HH^{\prime} is a 22-connected graph. Since (H;W1,W3)(H^{\prime};W^{1},W^{3}) is an ear-decomposition of HH and W1,W3W^{1},W^{3} are even paths, by Lemma 5.2 this yields a contradiction.

Suppose W2W^{2} is an even path and W1,W3W^{1},W^{3} are odd paths. Let HiH_{i} be a graph obtain from HH by removing WiW^{i} for i[3]i\in[3]. It is obvious that each HiH_{i} is a 22-connected graph. If L0L_{0} is an even cycle, then (H2;W2)(H_{2};W^{2}) is an ear-decomposition of GG, and by Lemma 5.2 (1) and (2) this yields a contradiction. If r=0r=0 and L0L_{0} is an odd cycle, then P=L0I(L)P=L_{0}-I(L) is an even path and C=H2I(P)C=H_{2}-I(P) is an even cycle. Since (C;P,W2)(C;P,W^{2}) is an ear-decomposition of HH and P,W2P,W^{2} are even paths, by Lemma 5.2 (1) and (3) this yields a contradiction. If r=0r=0 and L0L_{0} is an even θ\theta-graph, then suppose T1,T2T_{1},T_{2} and T3T_{3} are routes of L0L_{0}, and suppose LL is a subpath of T1T_{1}. Then (H2I(T2);T2,W2)(H_{2}-I(T_{2});T_{2},W^{2}) is an ear-decomposition of HH and T2,W2T_{2},W^{2} are even paths, a contradiction. If r=0r=0 and L0L_{0} is a uniform umbrella, then there is a rim WW of L0L_{0} such that LL is not a subpath of WW. Then (H2I(W);W,W2)(H_{2}-I(W);W,W^{2}) is an ear-decomposition of HH and W,W2W,W^{2} are even paths, a contradiction. If r>0r>0 and nn is odd, then (L0;,W2)(L_{0};\cdots,W^{2}) is an ear-decomposition of HH. Since |L0||L_{0}| is odd and W2W^{2} is an even path, by Lemma 5.2 (1) and (3) this yields a contradiction.  

Now for the sufficiency, suppose =(L0;L1,,Lt)\mathcal{E}=(L_{0};L_{1},\cdots,L_{t}) satisfies all conditions of the theorem, i.e., \mathcal{E} is a standard ear-decomposition of GG that has properties Q and R, LiL_{i} is an odd path for i[t]i\in[t], and f(,j,i)f(\mathcal{E},j,i) is an odd path when f(,j,i)K4f(\mathcal{E},j,i)\neq K_{4}. Recall the definitions of digraph DD, set UU and integer lil_{i} in Lemma 5.5. We choose an integer rr from UU such that |f(,lr,r)||f(\mathcal{E},l_{r},r)| is minimum. For convenience, let l=lrl=l_{r}. Then for each vertex uu of I(f(,l,r))I(Lr)I(f(\mathcal{E},l,r))\cup I(L_{r}), we have dG(u)=2d_{G}(u)=2. The proof proceeds by induction on tt. By Lemmas 1.1 (2) and 5.3, the result holds for t=0t=0.

If LrL_{r} is not an edge, then let GG^{\prime} be a graph obtained from GG by replacing f(,l,r)f(\mathcal{E},l,r) with an edge f=arbrf=a_{r}b_{r}, let G1=GI(Lr)G^{\prime}_{1}=G^{\prime}-I(L_{r}) and G2=LrfG^{\prime}_{2}=L_{r}\cup f. Let L=[LlI(f(,l,r))E(f(,l,r))]fL=[L_{l}-I(f(\mathcal{E},l,r))-E(f(\mathcal{E},l,r))]\cup f. Let \mathcal{E}^{\prime} be an ear-decomposition of G1G^{\prime}_{1} obtained from \mathcal{E} by removing LrL_{r}, and then replacing LlL_{l} with LL. If l>0l>0, then since f(,l,r)f(\mathcal{E},l,r) is an odd path, LL is an odd path and \mathcal{E}^{\prime} satisfies all the conditions. If l=0l=0 and LlL_{l} is a uniform umbrella (an odd cycle or an even cycle), then LL is also a uniform umbrella (an odd cycle, an even cycle), i.e., \mathcal{E}^{\prime} satisfies all the conditions in this case. If l=0l=0 and LlL_{l} is an even θ\theta-graph, then \mathcal{E}^{\prime} satisfies all the conditions except for L=K2,3L=K_{2,3}. Thus, \mathcal{E}^{\prime} satisfies all the conditions unless L=K2,3L=K_{2,3}.

If LK2,3L\neq K_{2,3}, then \mathcal{E}^{\prime} satisfies all the conditions. Since the number of paths in \mathcal{E}^{\prime} is t1t-1, by the induction hypothesis we have md(G1)=|G1|2md(G^{\prime}_{1})=\left\lfloor\frac{|G^{\prime}_{1}|}{2}\right\rfloor. Since G2G^{\prime}_{2} is an even cycle, we have md(G2)=|G2|2md(G^{\prime}_{2})=\frac{|G^{\prime}_{2}|}{2}. Thus, by Lemma 2.1, md(G)=md(G1)+md(G2)1=|G|2md(G^{\prime})=md(G^{\prime}_{1})+md(G^{\prime}_{2})-1=\left\lfloor\frac{|G^{\prime}|}{2}\right\rfloor. Since GG is a graph obtained from GG^{\prime} by replacing ff with the odd path f(,l,r)f(\mathcal{E},l,r), by Lemma 2.2 we have md(G)md(G)+f(,l,r)12=n2md(G)\geq md(G^{\prime})+\left\lfloor\frac{||f(\mathcal{E},l,r)||-1}{2}\right\rfloor=\left\lfloor\frac{n}{2}\right\rfloor. Therefore, md(G)=n2md(G)=\left\lfloor\frac{n}{2}\right\rfloor.

If L=K2,3L=K_{2,3}, then l=0l=0 and r=1r=1. Since rUr\in U, drd_{r} is maximum and dr=1d_{r}=1 (the definition drd_{r} is in the proof of Lemma 5.5). Thus, LiZL_{i}\in Z_{\mathcal{E}} for each i[t]i\in[t]. Let T1,T2T_{1},T_{2} and T3T_{3} be routes of L0L_{0} with |T1||T2||T3||T_{1}|\leq|T_{2}|\leq|T_{3}|. Then T1T_{1} and T2T_{2} are 22-paths and f(,0,r)f(\mathcal{E},0,r) is a subpath of T3T_{3} with |f(,0,r)|=|T3|1|f(\mathcal{E},0,r)|=|T_{3}|-1. Since L0K2,3L_{0}\neq K_{2,3}, we have |f(,0,r)|=|T3|14|f(\mathcal{E},0,r)|=|T_{3}|-1\geq 4. For each LiL_{i}, if end(Li)I(Tj)end(L_{i})\cap I(T_{j})\neq\emptyset for j[2]j\in[2], then |f(,0,i)|=2<|f(,l,r)||f(\mathcal{E},0,i)|=2<|f(\mathcal{E},l,r)|, a contradiction; if end(Li)=end(T3)end(L_{i})=end(T_{3}), then f(,0,i)f(\mathcal{E},0,i) is an even path, a contradiction. Thus, f(,0,i)f(\mathcal{E},0,i) is a proper subpath of T3T_{3} and |f(,0,i)|=|f(,0,r)||f(\mathcal{E},0,i)|=|f(\mathcal{E},0,r)| for each i[t]i\in[t]. If end(Li)end(Lr)end(L_{i})\neq end(L_{r}) for i,j[t]i,j\in[t], then end(Li)I(f(,0,r))end(L_{i})\cap I(f(\mathcal{E},0,r))\neq\emptyset and f(,0,i)f(\mathcal{E},0,i) is not a proper subpath of f(,0,r),f(\mathcal{E},0,r), i.e., \mathcal{E} does not have property R, a contradiction. Therefore, end(Li)=end(Lj)end(L_{i})=end(L_{j}) for each i,j[t]i,j\in[t]. Let H=T2T3(i[t]Li)H=T_{2}\cup T_{3}\cup(\bigcup_{i\in[t]}L_{i}). Then HH is a graph constructed in Remark 1. Thus, md(H)=|H|2md(H)=\frac{|H|}{2}. Suppose Γ\Gamma is an extremal MDMD-coloring of HH (see Remark 1). Let T1=ue1ae2vT_{1}=ue_{1}ae_{2}v and T2=uf1bf2vT_{2}=uf_{1}bf_{2}v. Since G=HT1G=H\cup T_{1}, let Γ\Gamma^{\prime} be an edge-coloring of GG such that Γ(e)=Γ(e)\Gamma(e)=\Gamma^{\prime}(e) for each eE(H)e\in E(H), and Γ(e1)=Γ(f2)\Gamma(e_{1})=\Gamma^{\prime}(f_{2}) and Γ(e2)=Γ(f1)\Gamma(e_{2})=\Gamma^{\prime}(f_{1}). Then Γ\Gamma^{\prime} is an MDMD-coloring of GG with n2\left\lfloor\frac{n}{2}\right\rfloor colors, i.e., md(G)=n2md(G)=\left\lfloor\frac{n}{2}\right\rfloor.

If LrL_{r} is an edge, then replace LlL_{l} by LlLrI(f(,l,r))L_{l}\cup L_{r}-I(f(\mathcal{E},l,r)) and replace LrL_{r} by f(,l,r)f(\mathcal{E},l,r). Then the new ear-decomposition also satisfies all the conditions. Moreover, drd_{r} is maximum and |f(,lr,r)|=2|f(\mathcal{E},l_{r},r)|=2 is minimum in the new ear-decomposition. Since LrL_{r} is not an edge in the new ear-decomposition, this case has been discussed above.  

Remark 3.

Recalling the proof of Lemma 5.1, we can find a normal ear-decomposition for a given 22-connected graph in polynomial time. For a normal ear-decomposition \mathcal{E} of GG, deciding whether \mathcal{E} satisfies all the conditions of Theorem 5.6 can be done in polynomial time. Thus, given a 22-connected graph GG, deciding whether md(G)=n2md(G)=\left\lfloor\frac{n}{2}\right\rfloor is polynomially solvable.

Corollary 5.10.

If GG is a 22-connected graph with md(G)=|G|2md(G)=\left\lfloor\frac{|G|}{2}\right\rfloor, then GG is a planar graph.

Proof.

By Theorem 5.6, there is a standard ear-decomposition ={L0;L1,,Lt}\mathcal{E}=\{L_{0};L_{1},\cdots,L_{t}\} of GG that has properties Q and R. Since GG is a planar graph if GG is a cycle, an umbrella or a θ\theta-graph, the result holds for t=0t=0. Our proof proceeds by induction on tt. Suppose t>0t>0. By Lemma 5.5, there are integers k,ik,i such that f(,k,i)f(\mathcal{E},k,i) is a path of order at least two, and dG(u)=2d_{G}(u)=2 for each uI(f(,k,i))I(Li)u\in I(f(\mathcal{E},k,i))\cup I(L_{i}). Let GG^{\prime} be a graph obtained from GG by removing LiL_{i}. By Lemma 5.2 (1), md(G)=|G|2md(G^{\prime})=\left\lfloor\frac{|G^{\prime}|}{2}\right\rfloor. By the induction hypothesis, GG^{\prime} is a planar graph. Since dG(u)=2d_{G}(u)=2 for each uI(f(,k,i))u\in I(f(\mathcal{E},k,i)), there is a face FF of GG^{\prime} such that f(,k,i)f(\mathcal{E},k,i) is a subpath of FF. Therefore, LiL_{i} can be embedded in FF and GG is a planar graph.  

References

  • [1] X. Bai, Y. Chen, M. Ji, X. Li, Y. Weng, W. Wu, Proper disconnection of graphs, arXiv:1906.01832 [math.CO].
  • [2] X. Bai, X. Li, Graph colorings under global structural conditions, arXiv:2008.07163 [math.CO].
  • [3] J. Bang-Jensen, T. Ballitto, A. Yeo, Proper-walk connection number of graphs, J. Graph Theory, DOI: 10.1002/jgt.22609, in press (2020), 1–23.
  • [4] J.A. Bondy, U.S.R. Murty, Graph Theory, GTM 244, Springer, 2008.
  • [5] V. Borozan, S. Fujita, A. Gerek, C. Magnant, Y. Manoussakis, L. Montero, Zs. Tuza, Proper connection of graphs, Discrete Math. 312(17) (2012), 2550–2560.
  • [6] Y. Caro, R. Yuster, Colorful monochromatic connectivity, Discrete Math. 311 (2011), 1786–1792.
  • [7] G. Chartrand, S. Devereaux, T.W. Haynes, S.T. Hedetniemi, P. Zhang, Rainbow disconnection in graphs, Discuss. Math. Graph Theory 38(4) (2018), 1007–1021.
  • [8] G. Chartrand, G.L. Johns, K.A. McKeon, P. Zhang, Rainbow connection in graphs, Math. Bohem. 133 (2008), 85–98.
  • [9] Y. Chen, P. Li, X. Li, Y. Weng, Complexity results for the proper disconnection of graphs, Proceedings of 14th International Frontiers of Algorithmics Workshop (FAW 2020), LNCS No.12340.
  • [10] P. Dankelmann, S. Mukwembi, H.C. Swart, Average distance and vertex-connectivity, J. Graph Theory 62(2) (2010), 157–177.
  • [11] O. Favaron, M. Kouider, M. Mahéo, Edge-vulnerability and mean distance, Networks 19 (1989), 493–504.
  • [12] P. Li, X. Li, Monochromatic disconnection of graphs, accepted for publication in Discrete Appl. Math. arXiv:1901.01372 [math.CO].
  • [13] P. Li, X. Li, Monochromatic disconnection: Erdős-Gallai-type problems and product graphs, arXiv:1904.08583 [math.CO].
  • [14] J. Plesnĺk, On the sum of all distances in a graph or digraph, J. Graph Theory 8 (1984), 1–24.