449
Views
0
CrossRef citations to date
0
Altmetric
Articles

Rainbow connection number of generalized composition

&

Abstract

Let G be a connected graph with V(G)={v1,,vn}. The rainbow connection number rc(G) is the smallest k for which there is a map γ:E(G){1,,k} such that any two vertices can be connected by a path whose edge colors are all distinct. The generalized composition G[H1,,Hn] is obtained by replacing each vi with the graph Hi. We prove rc(G[H1,,Hn])=diam(G) if each Hi has at least diam(G)4 vertices, improving known upper bounds of Basavaraju et al. and Gologranc et al. (2014). We prove the same result when diam(G)=3 but with some additional conditions. When diam(G)=2, we show that the largest possible value of rc(G[H1,,Hn]) is related to whether every vertex of G is contained in a triangle or not.

1 Introduction

In 2008 Chartrand et al. [Citation1] introduced new concepts that use edge-coloring to strengthen the connectedness property of a graph. An edge-coloring on a graph G is a map E(G){1,,k} (also called “k-coloring”). A rainbow path is a path whose edge colors are all distinct. A rainbow coloring is an edge-coloring in which any two vertices can be connected by a rainbow path. The rainbow connection number rc(G) is the smallest k for which G has a rainbow k-coloring. A strong rainbow coloring is an edge-coloring in which any two vertices can be connected by a rainbow geodesic. The strong rainbow connection number src(G) is the smallest k for which G has a strong rainbow k-coloring.

We have [Citation1] (1) diam(G)rc(G)src(G)|E(G)|.(1)

The reader is referred to [Citation2] for a detailed survey. It is known that computing rc and src is NP-hard [Citation3]. Many studies have focused on special classes of graphs or graph operations, such as Cartesian product, strong product, lexicographic product (see [Citation4] and [Citation5]), and graph join (see [Citation6]).

We study generalized composition, which can be thought of as “blowing-up” vertices into individual graphs. Let V(G)={v1,,vn}, and H1,,Hn be any graphs. The generalized composition G[H1,,Hn] is obtained by replacing each vi with Hi and adding a new edge between every vertex of Hi and every vertex of Hj whenever vivjE(G). We call this operation as G-composition. See .

Fig. 1 More examples, P3[P2,C3,K1] (left) and K1,3[2K1,P3,K1,P2] (right).

Fig. 1 More examples, P3[P2,C3,K1] (left) and K1,3[2K1,P3,K1,P2] (right).

Examples.

P2[H1,H2]=H1+H2 is the usual graph join, and a Pn-composition is known as a sequential join. The special case GH=G[H,H,,H] is known as composition or lexicographic product.

We always assume that G is non-trivial and connected. If G[H1,,Hn] is not a complete graph, then its diameter is max{2,diam(G)}. So (2) rc(G[H1,,Hn])diam(G).(2) If each Hi has at least diam(G)4 vertices, we show that (2) becomes an equality. When H1==Hn, this improves the results of Basavaraju et al. [Citation4] (rc(GH)2rad(G)) and Gologranc et al. [Citation5] (rc(GH)2diam(G)+1).

If diam(G)3, the bound (2) can be strict. However, we show that equality occurs when diam(G)=3 and some conditions are met (either each Hi has at least one edge, or G has the property that there is a 3-walk between every pair x,yV(G) possibly with x=y). When diam(G)=2, we show that the largest possible value of rc(G[H1,,Hn]) determines whether every vertex of G is contained in a triangle or not.

2 Results

2.1 A preliminary bound

Let QV(G). Its common neighborhood CN(Q) is the intersection of N(v)={w:vwE(G)} over all vQ. A set of vertices is independent if any two are non-adjacent, or co-neighboring if N(v)=N(w) for all v,wQ.

Lemma 1

Let QV(G) be a co-neighboring set. Then

(1)

src(G)|Q|1|CN(Q)|.

(2)

If moreover CN(Q) is independent, then rc(G)min4,|Q|1|CN(Q)|.

Proof

This is based on an idea in [Citation1]. Let CN(Q)={t1,,tb}. Given a k-coloring γ on G, define the color code of vQ with respect to CN(Q) as (3) code(v)=(γ(vt1),,γ(vtb)).(3) Note that there are at most kb distinct codes.

Claim

There is a rainbow geodesic between v,wQ if and only if code(v)code(w).

In fact, any geodesic between v and w has the form vtw with tCN(Q), which is rainbow if and only if γ(vt)γ(wt). Now we prove the lower bounds.

(1) Let k=|Q|b1. Suppose src(G)k, so there is a strong rainbow k-coloring on G. Since kb<|Q|, there are v,wQ with the same code. By the claim, there are no rainbow geodesics between them, a contradiction.

(2) Let k=min{3,|Q|b1}. Suppose rc(G)k, so there is a rainbow k-coloring on G. Since kb<|Q|, there are v,wQ with the same code. Let L:vxyw be a rainbow geodesic between v and w. Then x,yCN(Q), since Q is co-neighboring. By the claim, L is not a rainbow geodesic. So xy. Since CN(Q) is independent, dG(x,y)2. The length of L is at least 2+dG(x,y)4, contradicting k3. □

Below is the reason we only study the rc of G-compositions, not the src.

Corollary 1

The src of G-compositions cannot be bounded above in terms of G alone.

Proof

Let k,cN. Replace some vV(G) with mK1 such that m>ckdeg(v), and replace any other vertex with kK1, to get a G-composition graph A. The set Q=V(mK1) is co-neighboring and |CN(Q)|=kdeg(v), so by Lemma 1 (1) we have src(A)|Q|1|CN(Q)|>c. □

2.2 Diameter at least four

Theorem 1

Let diam(G)4 and n=|V(G)|. If each Hi has at least diam(G) vertices, then rc(G[H1,,Hn])=diam(G).

We prove the upper bound separately for later use.

Lemma 2

If each Hi has at least max{4,diam(G)} vertices, then (4) rc(G[H1,,Hn])max{4,diam(G)}.(4)

Proof

Let A=G[H1,,Hn] and V(Hi)={(i,j)|1jni}. We will construct a rainbow u-coloring on A, where u=max{4,diam(G)}. Define a map γ:E(A){0,1,,u1} arbitrarily on each E(Hi), and put (5) γ(i,j)(i,j)=j+j(modu)(5) for all i,i adjacent in G. We show that γ is a rainbow coloring.

Let x=(i,j), y=(i,j) with i,i non-adjacent in G. We will find a rainbow path between x and y.

Case 1: dG(i,i)=0 or 2.

Choose a common neighbor i1 of i,i. If jj(modu), then the path (i,j)2j(i1,j)2j+1(i,j+1)2j+3(i1,j+2)2j+2(i,j) is rainbow. Otherwise, the path (i,j)2j(i1,j)j+j(i,j) is rainbow.

Case 2: dG(i,i)=2m+13.

Choose a path ii1i2mi in G. Define the numbers j1,j2,,j2m{0,1,,u1} as j2k1=j+j+mk+1(modu) and j2k=jk(modu), for each k{1,2,,m}. Thus (6) j1=j+j+m,j3=j+j+m1,,j2m1=j+j+1(6) and (7) j2=j1,j4=j2,,j2m=jm.(7) Consider the following path, (8) L:(i,j)(i1,j1)(i2,j2)(i2m,j2m)(i,j).(8) The color sequence of this path is (9) 2j+j+m,2j+j+m1,2j+j+m2,,2j+jm.(9) The numbers above are 2m+1 consecutive integers in decreasing order. Since 2m+1u, these numbers are all different modulo u. So L is a rainbow path.

Case 3: dG(i,i)=2m4.

Use the path in Case 2 after deleting the penultimate vertex. □

2.3 Diameter three

The conclusion of Theorem 1 can fail when diam(G)=3 even if we control the number of vertices in each Hi. Below is a class of such examples.

Theorem 2

If diam(G)4 and some vertex of G is not contained in any triangle, then for any fixed k4 we have (10) maxkrc(G[H1,,Hn])=4(10) where the maximum is taken over all possible graphs H1,,Hn such that each Hi has at least k vertices.

Proof

By Lemma 2, the maximum is at most 4. To build a tight example, choose some vV(G) not in any triangle. Replace v with mK1 where m>3kdeg(v), and every other vertex with kK1 to form a G-composition graph A. Then Q=V(mK1) is co-neighboring and CN(Q) is independent (otherwise v would be in a triangle) so rc(A)min|Q|1|CN(Q)|,4>3 by Lemma 1. □

If some additional conditions are met, we do have an exact result.

Theorem 3

Let G be a connected graph with diam(G)=3, and let H1,,Hn be arbitrary graphs with at least three vertices each. Suppose that one of the following holds,

(1)

each Hi has at least one edge, or

(2)

G contains a 3-walk between every pair x,yV(G) (possibly with x=y).

Then rc(G[H1,,Hn])=3.

Proof

Let V(G)={1,,n} and A=G[H1,,Hn].

First, assume that (1) holds. We will construct a rainbow 3-coloring on A. We start with the following procedure.

(1)

For each i{1,,n}, choose a spanning forest Fi for Hi.

(2)

Choose a bipartition V(Fi)=ViWi so that |Vi|1 and |Wi|2. This is possible since we assumed |V(Hi)|3.

(3)

Put all isolated vertices of Hi (if any) in Wi.

(4)

Choose a non-isolated vertex of Hi (which exists, because we assumed E(Hi)), and put it in Wi. Denote that vertex by si.

Now we define a map γ:E(A){0,1,2} as follows.

(1)

γ(e)=0 for each eE(Hi) or e=xy with xVi{si} and yVj.

(2)

γ(xy)=1 if xWi{si} and yVj.

(3)

γ(xy)=2 if xWi and yWj.

We show that γ is a rainbow coloring. Let x,yV(A) be non-adjacent. Then xV(Hi) and yV(Hj) for some non-adjacent i,j in G.

Case 1: dG(i,j)=0 or dG(i,j)=2.

Choose a walk ii1j in G.

Subcase 1.1: xVi and yVj.

Choose w1WiN(x) and w2Wi1{si1}. Then x0w12w21y is rainbow.

Subcase 1.2: xVi and yWj.

The path x0si12y is rainbow.

Subcase 1.3: xWi is not isolated in Hi and yWj.

Choose vViN(x) and wWi1{si1}. Then x0v1w2y is rainbow.

Subcase 1.4: xWi is isolated in Hi and yWj is isolated in Hj.

Choose adjacent vVi1,wWi1. The path x1v0w2y is rainbow.

Case 2: dG(i,j)3.

Choose a path ii1i2j in G.

Subcase 2.1: xVi and yVj.

Choose wWi2{si2}. Then the path x0si12w1y is rainbow.

Subcase 2.2: xVi and yWj.

Choose vVi1 and wWi2{si2}. Then x0v1w2y is rainbow.

Subcase 2.3: xWi{si} and yWj.

Choose any vVi1. Then x1v0si22y is rainbow.

Subcase 2.4: x=si and yWj.

Choose vVi1 and wWi2{si2}. Then x0v1w2y is rainbow.

This proves that if (1) holds, then rc(G[H1,,Hn])=3. If (2) holds, we are done by the next lemma which we prove separately for later use. □

Lemma 3

If G contains a 3-walk between any x,yV(G) (possibly with x=y), and each Hi has at least 3 vertices, then rc(G[H1,,Hn])3.

Proof

We will construct a rainbow 3-coloring on A=G[H1,,Hn]. Let V(Hi)={(i,j)|1jni}. Define a map γ:E(A){0,1,2} arbitrarily on each E(Hi), and put (11) γ(i,j)(i,j)=j+j(mod3)(11) for all i,i adjacent in G. We prove that γ is a rainbow coloring.

Let x,yV(A) be non-adjacent, say x=(i,j) and y=(i,j) with i,i non-adjacent in G. Choose any walk ii1i2i in G. We use this walk to find a rainbow path between x and y.

If jj(mod3), use (i,j)2j+1(i1,j+1)2j+3(i2,j+2)2j+2(i,j). If jj(mod3), use (i,j)2j(i1,j)j+j(i2,j)2j(i,j) as a rainbow path. □

2.4 Diameter two

The rc of a graph can sometimes determine the structure of that graph. For instance, rc(G)=1 if and only if G is a complete graph, while rc(G)=|E(G)| if and only if G is a tree (see [Citation1]). Below, we show that if diam(G)=2 then the largest possible value of rc(G[H1,,Hn]) determines whether every vertex of G is contained in a triangle or not.

Theorem 4

If diam(G)=2 and k4, then (12) maxkrc(G[H1,,Hn])=3, if every vertex of G lies on a triangle4, otherwise.(12) where the maximum is taken over all possible graphs H1,,Hn such that each Hi has at least k vertices.

Proof

Suppose that every vertex is contained in a triangle. This assumption together with diam(G)=2 implies that G has a 3-walk between any x,yV(G) (possibly x=y). So by Lemma 3 we have rc(A)3 for any G-composition graph A.

Now we build a tight example. Replace a vertex vV(G) with mK1 where m>2kdeg(v), and every other vertex with kK1, to get a G-composition graph A. Then Q=V(mK1) is co-neighboring and |CN(Q)|=kdeg(v) so src(A)|Q|1|CN(Q)|>2 by Lemma 1 (1), i.e. src(A)3. This implies rc(A)3, because any rainbow 2-coloring must also be a strong rainbow coloring.

If some vertex does not lie on any triangle, we are done by Theorem 2. □

3 Concluding remarks and open problems

We have proved rc(G[H1,,Hn])=diam(G) when each Hi has at least diam(G)4 vertices. We have also studied what happens when diam(G)3, but we did not consider the case that some Hi has less than diam(G) vertices.

Theorem 3 (2) is a partial converse to Theorem 2 because if G contains a 3-walk between any x,yV(G) (possibly with x=y), then every vertex is contained in a triangle. These two statements are not equivalent, but it might be interesting to consider whether the weaker statement is enough. If it is, then the conclusion of Theorem 4 will also be true in the case diam(G)=3.

References

  • ChartrandG.JohnsG.L.McKeonK.A.ZhangP., Rainbow connection in graphs Math. Bohem. 1332008 85–98
  • LiX.SunY., Rainbow Connections of Graphs2012Springer
  • S. Chakraborty, E. Fischer, A. Matsliah, R. Yuster, Hardness and algorithms for rainbow connectivity, in: 26th International Symposium on Theoretical Aspects of Computer Science STACS, 2011, pp. 243–254.
  • BasavarajuM.ChandranL.S.RajendraprasadD.RamaswamyA., Rainbow connection number of graph power and graph products Graphs Combin. 302014 1363–1382
  • GolograncT.MekišG.PeterinI., Rainbow connection and graph products Graphs Combin. 302014 591–607
  • SeptyantoF.SugengK.A., Rainbow connections on graph joins Australas. J. Combin. 692017 375–381