569
Views
0
CrossRef citations to date
0
Altmetric
Original Article

2-swappable graphsFootnote

&
Pages 242-250 | Received 25 Jan 2017, Accepted 06 Nov 2017, Published online: 10 Jun 2020

Abstract

Let G=(V,E) be a graph, by V2, we mean the set of all 2-combinations of V. We say that a graph G is k-swappable if for every edge eE, there are two sets AE, BV2E such that eA, |A|k, and G(V,(EA)B). The swapping number of G is the minimum k>0 such that G is k-swappable.

Let d2 be an integer. Every tree with exactly one vertex of degree d and remaining vertices of degree d+1 or 1 we will call a d-nary tree.

In this paper we give sufficient conditions for general graphs as well as sufficient and necessary conditions for a d-nary tree for d>2 to be 2-swappable.

1 Introduction

In this paper, we deal with finite, simple and undirected graphs. Let us consider a graph G=(V,E), whose number of vertices we call the order and number of edges we call the size of G. If SV, then we will understand GS to mean the graph induced by G on the vertex set VS. By V2, we mean the set of all 2-combinations of V. A vertex xV is called primary if deg(x)3(degG(x)3). A leaf is a vertex of degree one.

The girth of a graph G is the length of a shortest cycle contained in the graph and it is denoted by girth(G). If the graph does not contain any cycles its girth is defined to be infinity.

A spider is a tree T in which exactly one vertex x has degree greater than 2. We call x the central vertex of T. An arm of a spider T is a connected component of T{x}. We denote a spider with central vertex of degree d by S(a1,a2,,ad), where a1,a2,,ad are the numbers of vertices in the arms; thus |V(S(a1,a2,,ad))|=1+a1+a2++ad.

Let d2 be an integer. Every tree with exactly one vertex of degree d and remaining vertices of degree d+1 or 1 we will call a d-nary tree. The vertex of degree d we will call the root of a d-nary tree and denote by r.

In this paper we will consider k-swappable graphs. The definition of k-swappable was given in [Citation1].

Definition 1

Let G=(V,E) be a graph and let k be a positive integer. We say that G is k-swappable if for every edge eE, there are two sets AE, BV2E such that eA, |A|k, and G(V,(EA)B). The swapping number of G is the minimum k such that G is k-swappable.

There is an interesting motivation for investigation of swappable graphs [Citation1]. Consider a network connecting different computing resources; such a network is modelled by a graph. Suppose that the network works if and only if it has an exact structure G (logical topology of the network), so if any of the connections breaks then the network does not work. Suppose that connections between nodes are expensive to add or remove, so we are looking for the minimum number of new connections in a broken network in order to obtain the structure G.

In [Citation1] Froncek et al. classified all 1-swappable trees and unicyclic graphs, and proved that the expected value of the swapping number grows linearly with the size of the graph. They proved the following theorem:

Theorem 1

Let T be a tree on two or more vertices. Then T is 1-swappable if and only if T is a path on at least three vertices or TS(2,1,1) or TS(3,2,1).

Ross found four infinite families of r3 regular graphs that are 2-swappable, [Citation2]. The aim of this article is a characterization of 2-swappable d-nary trees for d>2.

2 Some 2-swappable graphs

In this section we present some family of 2-swappable graphs.

Theorem 2

Let G=(V,E) be a graph satisfying the following conditions:

1.

girth(G)>4,

2.

any primary vertices x,yV are not neighbours,

3.

for any leaf there exists a leaf with different neighbour,

then G is 2-swappable graph.

Proof

Let e=xyE.

Suppose first that x is a leaf. By assumption 3 there exists a leaf xV such that yN(x)={y}. Let A={xy,xy} and B={xy,xy}. Notice that BV2E and G(V,(EA)B).

From now by assumption 2 without loss of generality we can assume that x has degree 2 and N(x)={y,z}. Let N(z)={x,zi,i=1,,deg(z)1} and N(y)={x,yi,i=1,,deg(y)1}. Let us consider two cases:

Case 1. deg(z)=1

Let A={xy} and B={yz}. Since BV2E, G(V,(EA)B).

Case 2. deg(z)=2

Let A={xy,zz1} and B={xz1,yz}. Since girth(G)>4, observe that z1y. Notice that BV2E and G(V,(EA)B).

Case 3. deg(z)>2

If deg(z1)=1 then A={xy} and B={yz1} and hence G(V,(EA)B).

If deg(z1)1 then by assumption 2 deg(z1)=2 and N(z1)={z,t}. Note that ty, because girth(G)>4. Let now A={xy,z1t} and B={xt,yz1}. It follows that BV2E and G(V,(EA)B).

Recall that there are r3 regular graphs that are 2-swappable [Citation2]. Therefore the condition 2 given in Theorem 2 is not necessary. Moreover, in we show that none of the conditions from Theorem 2 is necessary.

Fig. 1 Examples of 2-swappable graphs.

3 The family of d-nary trees

Let Td=(V,E) be a d-nary tree with the root r and let vV. Then there is exactly one path joining vertices r and v in Td. Let d(r,v) denote the distance of vertices r and v. We will usually refer to a path by one of the sequences of consecutive vertices. That is, for vertices v, w let (v,w)-path denote the path from v to w.

Let vV such that vvE and v is a vertex of the (r,v)-path. A branch of Td pointed by the vertex v, denoted by B[v], is a connected component of Td{v} with v as a vertex. Observe that B[v] is a d-nary tree or a vertex. Let uV such that v is a vertex of (r,u)-path. A twig pointed by the vertex v and determined by the vertex u, denoted by Tu[v], is the graph B[v]B[v], where vN(v)B[v] and v is a vertex of (r,u)-path.

Let (x1,,xi,,xk) be (x1,xk)-path and (y1,,yi,,yk) be (y1,yk)-path. If there are vertices ux, uyV such that twigs Tux[xi] and Tuy[yi] are isomorphic for i=1,,k, then we say that (x1,xk)-path and (y1,yk)-path are twins (or more precisely the paths are twins according to the vertex ux for (x1,xk)-path and the vertex uy for (y1,yk)-path). We say that (x1,xk)-path is symmetric (symmetric according to the vertex u) if there is a vertex uV such that for every i{1,,k} twigs Tu[xi] and Tu[xk+1i] are isomorphic and k2. The (x1,xk)-path is sequential (according to the vertex u) if there are a vertex u and j{1,,k1} such that j|k and twigs Tu[xi] and Tu[xi+j] are isomorphic graphs for i{1,,kj} and k3.

Let vwE such that v is a vertex of (r,w)-path. We will write that the edge vw is adjacent to a symmetric path if there is at least one of vertices u, u such that (u,v)-path is symmetric according to w or (w,u)-path is symmetric, where u is a vertex of (r,v)-path, w is a vertex of (r,u)-path. We say that the edge vw is added to a sequential path if (r,v)-path is sequential according to w. We say that the edge vw joins twin paths if there are vertices u, uV which are not vertices of (r,w)-path and such that (r,v)-path according to w and (w,u)-path according to u are twins (see ).

Fig. 2 An example of 2-swappable 3-nary tree.

Theorem 3

Let Td=(V,E), d3 be a d-nary tree with the root r. The only 2-swappable TdK1,d (d3) is K1,3. The graph TdK1,d is 2-swappable if and only if for every edge vwE, deg(w)>1, where v is a vertex of (r,w)-path, one of the following conditions holds:

(i) vw is adjacent to a symmetric path,

(ii) vw is added to a sequential path,

(iii) vw joins twins paths,

(iv) there is a vertex wVN(v) such that branches B[w], B[w] are isomorphic graphs,

(v) there is a vertex uv such that (r,v)-path according to w and (r,u)-path are twins

(vi) (only for d=3 and e=rw) there is a leaf lB[w] such that d(r,l){1,2} or (u,u)-path is symmetric according to l, where ru, ulE.

Proof

Let Td=K1,3. Set N(r)={v1,w1,u1}. Let e=rv1. If A={rv1,rw1}, B={u1v1,u1w1} we obtain that Td and (V,(EA)B) are isomorphic. It is easy to see that it is the only 2-swappable star K1,d.

We can assume that TdK1,d.

Let us consider e=vlE such that deg(l)=1. Since TdK1,d there is a leaf lV such that vN(l)={v}. Then Td(V,(E{vl,vl}){vl,vl}). Hence such edges have no influence property of 2-swappability. From now we will consider edges not at leaves.

Sufficiency. Let e=vwE and let v be a vertex of (r,w)-path. In every case we will find sets A and B such that eAE, BV2E, |A|2 and graphs (V,(EA)B) and Td are isomorphic.

Let us first suppose that vw is adjacent to a symmetric path. Let us assume that there are vertices u, uV such that (w,u)-path is symmetric according to u. We set A={vw,uu} and B={vu,wu}. If (v,r)-path is symmetric according to w then A={vw} and B={rw}. If there is a vertex uV{r} such that u is a vertex of (r,v)-path and (u,v)-path is symmetric according to w then we set A={vw,uu} and B={uv,uv}, where u is the only neighbour of u on (r,u)-path.

Suppose that vw is added to a sequential path. If (r,v)-path is an edge then A={vw} and B={rw}. Otherwise, there are consecutive vertices xj, xj+1, xj+2 of (r,v)-path such that Tw[r]Tw[xj+1], Tw[v]Tw[xj]. We set A={vw,xjxj+1}, B={xjw,rv}.

Assume that vw joins twins paths. Let u, u be vertices such that (r,v)-path according to w and (w,u)-path according to u are twins. Then we set A={vw,uu} and B={ru,vu}.

Suppose that there is a vertex w such that N(w)N(w)= and branches B[w], B[w] are isomorphic. Let vN(w) be the vertex of (r,w)-path. Then we write A={vw,vw} and B={vw,vw}.

Suppose that there is a vertex uv such that (r,v)-path according to w and (r,u)-path according to u are twins, where uN(u) and u is not a vertex of (r,u)-path. Then we set A={vw,uu} and B={vu,uw}.

Finally, let d=3, e=rw and there is a leaf lB[w] such that d(r,l){1,2} or (u,u)-path is symmetric according to l, where ru, ulE. Then A={rw,rs}, where sw, su and B={lw,ls}.

Necessity. Let eE. No vertices incident with e are leaves. Since we assume that Td is 2-swappable, for that edge e we have sets Ae, Be such that eAeE, |Ae|2, BeV2E and (V,(EAe)Be)Td. In the proof for that edge e we will consider every set A such that eA and |A|2. For every chosen set A we will reflect on every possible set B such that BV2E and (V,(EA)B)Td. In every case, for every considered sets A and B we will obtain that one of conditions (i)–(vi) holds. Since the proper sets Ae and Be are among considered pairs of sets A and B we will obtain that for that edge e at least one of the conditions (i)–(vi) holds.

We will consider possibilities of choosing A and then of choosing B. In every case σ denotes isomorphic mapping of Td on vertices of a new d-nary tree (V,(EA)B). By σ(Tu[v]) we will denote the image of Tu[v] by isomorphic mapping σ, that is the twig pointed by σ(v) and determined by σ(u) in d-nary tree (V,(EA)B).

Case 1. One vertex incident with e is the root r.

Denote e=rv1, deg(v1)=d+14. We have to consider five different types of the set A: {rv1}, {rv1,rw1}, {rv1,v1v2}, {rv1,vivi+1} where i2, d(r,vl)=l, l{1,2,i,i+1}, v1, vi are vertices of (r,vi+1)-path, {rv1,wjwj+1} where j1, d(r,wl)=l, l{j,j+1}, v1 is not a vertex of (r,wj+1)-path. In every case we try to choose the set B. Observe that, except for d=3 and A={rv1,rw1}, in every case new edges have to join only vertices of deleted edges and only one of that vertices could became a new root. No leaf could became a new root. We can construct a new d-nary tree for the last two types of A and for the second one for d=3.

Subcase 1.1. A={rv1,vivi+1}, i2, d(r,vl)=l, l{i,i+1}, v1, vi are vertices of (r,vi+1)-path.

Let vk denote the vertex of (r,vi+1)-path such that d(v,vk)=k, k=1,,i+1. We have two possibilities of choosing B: {rvi,v1vi+1} or {rvi,rvi+1}. Let us first suppose that B={rvi,v1vi+1}. Then σ(r)=r. Set W={w1:w1v1,rw1E,B[v1]B[w1]}. It is clear that σ(v1)W{vi} and for every w1W, σ(w1)W{vi}. Hence without loss of generality we may assume that σ(v1)=vi.

We will ask which vertices of (r,vi+1)-path satisfy property: σ(vk)=vi+1k, 1ki. Observe that for k=1 the property is satisfied. Let us consider k̄{1,,i} such that σ(vk̄)=vi+1k̄. It is easy to see that then σ(vk)=vi+1k for every k{1,,k̄}. Observe that for kk̄1 we have that σ(Tvi+1[vi])Tv[vi+1k]. Therefore Tvi+1[vi]Tvi+1[vi+1k] for kk̄1. If k̄1i2 we obtain that (v1,vi)-path is symmetric (according to vi+1) and hence e=rv1 satisfies condition (i). Let k̄1<i2 and σ(vk̄+1)vik̄. Then σ(vk̄+1) is a vertex of Tvi+1[vi+1k̄]. Let us consider σ(Tvi+1[vk̄]). Observe that the set of vertices of Tvi+1[vk̄] and, moreover, vertices v1, …,vk̄ are vertices of σ(Tvi+1[vk̄]). For that Tvi+1[vk̄], σ(Tvi+1[vk̄]) are isomorphic and have different number of vertices, a contradiction.

Let B={rvi,rvi+1}. Then σ(r)=v1. Let us suppose that for every k{1,,i1} we have σ(vk)=vk+1. Then σ(Tvi+1[vk])Tvi+1[vk+1], 1ki1 and then Tvi+1[vk]Tvi+1[vk+1], 1ki1. Since σ(r)=v1 and hence σ(Tvi+1[r])Tvi+1[v1] we obtain that Tvi+1[r]Tvi+1[vk], 1ki. Especially, (r,vi)-path is symmetric according to vi+1 and e=rv1 satisfies (i). Let us suppose that there is k0{1,,i2} such that σ(vk0)=vk0+1, σ(vk0+1)vk0+2. It is easy to see that σ(vk)=vk+1, kk0 and, like previously, we obtain that Tvi+1[r]Tvi+1[vk], kk0+1. We have that σ(vk0+1) is the vertex of Tvi+1[vk0+1]. Observe that vi and vertices of Tv1[r] are vertices of σ(Tvi+1[vk0]), contrary to that Tvi+1[r]Tvi+1[vk0]. Similarly we obtain that it is impossible that σ(v1)v2.

Subcase 1.2. A={rv1,wiwi+1}, i1, d(r,wl)=l, l{i,i+1} v1, wi are vertices of (r,wi+1)-path.

In this subcase B={rwi+1,v1wi}. Then σ(r)=r and like in Subcase 1.1., using the set W, we can assume that σ(wi+1)=v1. Hence B[v1]B[wi+1] and e satisfies (iv).

Subcase 1.3. A={rv1,rw1}, d=3, d(r,w1)=1.

Then B={lv1,lw1}, where l is a leaf and lB[v1], lB[w1]. Then σ(r)=l. If rlE then e=rv1 satisfies (vi). Let u1N(r){v1,w1}, deg(u1)=4. Like in Subcase 1.1. we may assume that σ(v1)=v1, σ(w1)=w1 and σ(u1)=u1. Moreover, using the same arguments as in Subcase 1.1. we can prove that σ(l)=r and (u1,u)-path is symmetric according to l, where ulE, and hence e=rv1 satisfies (vi).

Case 2. The edge e is not incident with the root.

Denote e=vivi+1, i1, d(vi)=d(vi+1)=d+14, d(r,vl)=l, l{i,i+1}. Let vk denote the vertex of (r,vi+1)-path such that d(r,vk)=k, k=1,,i+1.

Subcase 2.1. A={vivi+1}.

To obtain isomorphic d-nary tree we have to choose B={rvi+1}. Then σ(r)=vi. If σ(vi)r then there is a vertex w of Td such that viwE, B[w]B[vi+1] and vivi+1 satisfies (iv). Hence we may assume that σ(vi)=r and then, like in Subcase 1.1., σ(vk)=vik, k=1,,i1. Hence (r,vi)-path is symmetric according to vi+1 and e satisfies (i).

Subcase 2.2. A={vivi+1,vi+1vi+2}, where d(r,vi+2)=i+2.

Then B={rvi+1,vivi+2} and σ(r)=vi+1. Like in Subcase 2.1. if σ(vi+2)vi then vivi+1 satisfies (iv). Let σ(vi+1)=vi. Then σ(v1)=r, σ(vk)=vk1, k=1,,i+1. Observe that σ(Tvi+1[vk])Tvi+1[vk1], ki, σ(Tvi+1[v1])Tvi+1[r]. Hence Tvi+1[r]Tvi+1[vk], k<i and especially (r,vi)-path is symmetric according to vi+1 and e=vivi+1 satisfies (i).

Subcase 2.3. A={vivi+1,vjvj+1}, ji+2, d(r,vl)=l, l{i,i+1,j,j+1}.

We have several possibilities of choosing B: {vivj,vi+1vj+1}, {rvj,vi+1vj+1}, {rvi+1,vivj+1}, {rvj+1,vivj}, {rvj,vivj+1}. Edges in B join only vertices r, vi, vi+1, vj, vj+1. For chosen sets A, B let us reflect on branches of Td and branches of the new d-nary tree (V,(EA)B). Since Td and (V,(EA)B) are isomorphic, we obtain that families of branches pointed by all vertices from V are the same in both trees. Observe that if w is not a vertex of (r,vj)-path then the branch pointed by w in Td is isomorphic to the branch pointed by w in (V,(EA)B). For this reason the families of branches pointed by vertices of (r,vj)-path in Td and (V,(EA)B) are the same. Moreover, in Td branches pointed by vertices of (r,vj)-path have a property: for k̄1, B[vk̄] is a subgraph of B[r] and of B[vk], k{1,,k̄1}. Hence branches pointed by vertices of (r,vj)-path in (V,(EA)B) should have the same property.

I. B={vivj,vi+1vj+1}.

Then σ(r)=r and, by remark about branches, we can deduce that (vj,vi+1)-path according to vj+1 and (vi+1,vj)-path according to vj+1 are twins. Hence (vi+1,vj)-path is symmetric according to vj+1 and vivi+1 satisfies (i).

II. B={rvj,vi+1vj+1}.

Then σ(r)=vi. By remark about branches we have that (r,vi)-path and (vi,r)-path, both according to vi+1, are twins and hence (r,vi)-path is symmetric according to vi+1.

III. B={rvi+1,vivj+1}.

Then σ(r)=vj. We reflect on i+12 vertices of (r,vi)-path and ji2 vertices of (vi+1,vj)-path. We will find which paths are twins. Every pair of twins path in Td is taken according to vj+1. Let us first suppose that i+1=ji. Then by remark about branches, (vj,vi+1)-path and (r,vi)-path are twins. We also have that (r,vi)-path and (vi+1,vj)-path are twins. From above we obtain that (vi+1,vj)-path is symmetric according to vj+1 and hence (r,vi)-path is symmetric according to vi+1 and e satisfies (i).

Assume that i+1>ji. Observe that (r,vji1)-path and (vj,vi+1)-path are twins. We also have that (vji,vj)-path and (r,vi)-path are twins. Especially, their proper parts: (vi+1,vj)-path and (v2ij+1,vi)-path are twins, and hence (r,vji1)-path and (vi,v2ij+1)-path are twins. We consider vertices of (r,vi)-path. From above we also obtain that Tvi+1[vk]Tvi+1[vk+(ji)], k=1,,2ij, that is (r,v2ij)-path and (v2ij+1,vi)-path are twins. Moreover, we can observe that (vji,v2(ji)1)-path and (r,vji1)-path are twins, too and so on. Hence we have that Tvj+1[vk]Tvj+1[vk+(ji)], where k=1,,i(ji). Let i+1=n(ji)+α, where n, α are natural numbers such that n1, 0α<ji. Observe that (v(n1)(ji),vi)-path is symmetric according to vi+1 and vivi+1 satisfies (i).

Suppose that i+1<ji. We notice that (r,vi)-path and (vj,vj(i+1))-path are twins. Then we consider (vi+1,vj)-path and observe that (vj(i+1),vj)-path and (r,vi)-path are twins. Hence (vi,r)-path is symmetric according to vi+1.

IV. B={rvj+1,vivj}.

Then σ(r)=vi+1. We consider i+12 vertices of (r,vi)-path and ji2 vertices of (vi+1,vj)-path. We will find twins paths. Every path in Td will be taken according to vj+1.

Let us first suppose that i+1<ji. Then (r,vji1)-path and (vi+1,vj)-path are twins. We also have that (vji,vj)-path and (vi,r)-path are twins. Especially, (vji,vj)-path and (v2i+1,vi+1)-path are twins. From above we obtain that (r,vi)-path and (vi+1,v2i+1)-path are twins and, moreover, Tvj+1[vk]Tvj+1[vk+(i+1)], k=1,,ji1. Because of the structure of twigs of the first and of the last i+1 vertices of (vi+1,vj)-path we obtain that (vi+1,vj)-path is symmetric according to vj+1.

Suppose that i+1ji. Then (r,vji1)-path and (vi+1,vj)-path are twins and (vi+1,vj)-path and (vji1,r)-path are twins. Then (vi+1,vj)-path is symmetric according to vj+1 and vivi+1 satisfies (i).

V. B={rvj,vivj+1}.

We consider i+12 vertices of (r,vi)-path and ji2 vertices of (vi+1,vj)-path. It is obvious that σ(r)=vi+1. We will find twins paths. Every pair of twins paths in Td is taken according to vj+1. By the remark about branches we have that (r,vji1)-path, (vi+1,vj)-path are twins and (vji,vj)-path, (r,vi)-path are twins.

Let us first suppose that i+1ji. Then for (r,vi)-path we obtain that (vji,vj)-path and (r,vi)-path are twins. For (vi+1,vj)-path we obtain that (r,vi)-path and (vi+1,v2i+2)-path are twins. Hence we have that (r,vi)-path, (vi+1,v2i+2)-path are twins and vivi+1 joins twins paths.

Suppose that i+1>ji. For (vi+1,vj)-path we have that (r,vji1)-path, (vi+1,vj)-path are twins. For (r,vi)-path we know that Tvi+1[vk]Tvi+1[vk+(ji)], k=1,,i(ji), Tvi+1[r]Tvi+1[vji] and (vi+1,vj)-path, (vi(ji)+1vi)-path are twins.

From now we will consider (r,vi)-path. From above we obtain that (r,vji1)-path, (vi(ji)+1,vi)-path are twins and Tvi+1[vk]Tvi+1[vk+(ji)], k=1,,i(ji), Tvi+1[r]Tvi+1[vji]. Let i+1=n(ji)+α1, where n, α1 are natural numbers such that n1, 0α1<ji. If α1=0 then it is easy to see that n2 and (r,vi)-path is sequential according to vi+1. Assume that α1(0,ji). Using properties of (r,vi)-path we will describe its twigs, first starting with r and going to vi and then starting with vi and going to r. In the first way we obtain that Tvi+1[vt(ji)]Tvi+1[r], Tvi+1[vk+t(ji)]Tvi+1[vk], t{0,,n}, k{1,,ji1}, Tvi+1[vs+n(ji)]Tvi+1[vs], s{1,,α11}. In the other way we obtain that Tvi+1[vα1+t(ji)]Tvi+1[r], Tvi+1[vk+α1+t(ji)]Tvi+1[vk], t{0,,n1}, k{1,,ji1}, Tvi+1[vjiα1+s]Tvi+1[vs], s{1,,α11}, Tvi+1[vjiα1]Tvi+1[r]. Comparing these results we observe that Tvi+1[r]Tvi+1[vt(ji)]Tvi+1[vn(ji)]Tvi+1[vα1+t(ji)], t{1,,n1}, Tvi+1[vk]Tvi+1[vk+s(ji)]Tvi+1[vk+α1+s(ji)], s{0,,n1}, k{1,,ji1} and hence Tvi+1[vk]Tvi+1[vk+α1], k{1,,iα1}. Moreover, (r,vα11)-path and (vn(ji),vn(ji)+α11)-path ((vn(ji),vi)-path) are twins. Let i+1=mα1+α2, where m, α2 are natural numbers such that m1, 0α2<α1. If α2=0, then m2 and (r,vi)-path is sequential according to vi+1. If α2>0 we will repeat reasoning above replacing ji by α1 and α1 by α2. In every step we have the remainder less than the one in previous step. Hence there are the natural numbers αs, t such that i+1=tαs, and finally (r,vi)-path is sequential according to vi+1 the edge vivi+1 satisfies (iii).

Subcase 2.4. A={vivi+1,vi1vi}, i1 and (r,,vi1,vi,vi+1) is (r,vi+1)-path.

Observe that if i=1 and vi1=r it is impossible to fix the set B and a new d-nary tree. Assume that i2. Then B={rvi,vi1vi+1} and σ(r)=vi.

Let us suppose that σ(vi)vi1. Then w=σ(vi+1) is not a vertex of (r,vi1)-path and wvi+1. Since σ(vi)vi, we have that wN(vi). We have that B[w]B[vi+1]. Hence vivi+1 satisfies (iv).

Therefore we may assume that σ(vi)=vi1. Then σ(v1)=r, σ(vk)=vk1, k=2,,i. Hence Tvi+1[r]Tvi+1[vk], k=1,,i and especially (r,vi)-path is symmetric according to vi+1 and vivi+1 satisfies (i).

Subcase 2.5. A={vivi+1,vjvj+1}, 1ji2 and (r,,vj,vj+1,,vi,vi+1) is (r,vi+1)-path.

We have some possibilities of choosing B.

I. B={vjvi,vj+1vi+1}.

Then σ(r)=r. Observe that it is impossible that σ(vi)=vi.

If σ(vi)vj+1 then w=σ(vi+1) is not a vertex of (r,vi)-path and wvi+1, wN(vi), B[w]B[vi+1]. Hence vivi+1 satisfies (iv).

Therefore we may assume that σ(vi)=vj+1. Then σ(vj+1)=vi, σ(vj+1+k)=vik, k=1,,ij1, (σ(vi)=vj+1). Hence (vj+1,vi)-path is symmetric according to vi+1.

II. B={rvi,vj+1vi+1}.

Then σ(r)=vj. Similarly as in I., σ(vi)vi and if σ(vi)vj+1 then vivi+1 satisfies (iv). We may assume that σ(vi)=vj+1 and then σ(vj+k)=vi(k1), k=1,,ij, (σ(vj+1)=vi, σ(vi)=vj+1). Hence (vj+1,vi)-path is symmetric according to vi+1.

III. B={rvj+1,vjvi+1}.

Then σ(r)=vi. If σ(vi)vj then w=σ(vi+1) is not a vertex of (r,vi)-path, wvi+1, wN(vi), B[w]B[vi+1] and hence vivi+1 satisfies (iv).

Let σ(vi)=vj. Then σ(vk)=vik, k=1,,ij1, σ(vij)=r, σ(vij+l)=vl, l=1,,j and hence (r,vij1)-path, (vi,vj+1)-path are twins and (vij,vi)-path, (r,vj)-path are twins, in every case according to vj+1. We consider j+12 vertices of (r,vj)-path and ij vertices of (vj+1,vi)-path. Twins paths which we will find in Td, we will take according to vi+1.

Suppose that j+1ij. Then we have that (r,vj)-path and (vi,vi(j+1))-path are twins and also (vi(j+1),vi)-path and (r,vj)-path are twins. Hence (vi(j+1),vi)-path is symmetric according to vi+1 and vivi+1 satisfies (i).

Suppose that ij>j+1. Then we have that (r,vij1)-path and (vi,vj+1)-path are twins and also (vij,vi)-path and (r,vj)-path are twins. From above we obtain that Tvi+1[vk]Tvi+1[vk+(ij)], k=1,,j, Tvi+1[r]Tvi+1[vij]. Let us consider i+1 vertices of (r,vi)-path. Let i+1=n(ij)+α, where n, α are natural numbers such that n1, 0α<ij. Observe that (v(n1)(ij),vi)-path is symmetric according to vi+1 and vivi+1 satisfies (i).

IV. B={rvi+1,vjvi}.

Then σ(r)=vj+1. Once again, if σ(vi)r, then, since it is obvious that σ(vi)vi, vivi+1 satisfies (iv). Let σ(vi)=r. Then (r,vij+1)-path, (vj+1,vi)-path are twins, both according to vi+1 and (vij,vi)-path, (r,vj)-path are twins, both according to vi+1. We repeat arguments of Subcase 2.3.IV. replacing vj by vi, vj+1 by vi+1, vi by vj, vi+1 by vj+1 and obtain that (vj+1,vi)-path is symmetric according to vi+1.

V. B={rvi,vjvi+1}.

Then σ(r)=vj+1 and it is impossible that σ(vi)=vi. If σ(vi)vj we obtain that vivi+1 satisfies (iv). Let us assume that σ(vi)=vj. Then σ(vk)=vj+1+k, k=1,,ij1, σ(vij)=r, σ(vij+t)=vt, t=1,,j. We consider (r,vj)-path of j+1 vertices and (vj+1,vi)-path of ij vertices. In Subcase 2.3.V. we consider (x1,xu+1)-path such that Txu+1[xk]Txu+1[xk+w], k=1,,uw and (x1,xw)-path, (xuw+1,xu)-path are twins, both according to xu+1 where wu. We proved that if u=u1w+α1, where u1, α1 are nonnegative integers, u11, α1(0,w) then (x1,xu)-path has properties: Txu+1[xk]Txu+1[xk+α1], k=1,,uα1, (x1,xα1)-path, (xuα1+1,xu)-path are twins, both according to xu+1.

If α1=0, then u12 and (x1,xu)-path is sequential according to xu+1. Let d be the greatest common divisor of u and w. Observe that d is a divisor of α1, too. We can repeat the arguments for u and α1 and obtain that either u=u2α1 and (x1,xu)-path is sequential according to xu+1 or u=u2α1+α2, α2(0,α1), Txu+1[xk]Txu+1[xk+α2], (x1,xα2)-path, (xuα2+1,xu)-path are twins, both according to xu+1 and d is a divisor of α2. Since in every step we have the remainder less than the one in a previous step, then after several steps we obtain that there is αs such that u=us1αs, us1, αs are positive integers, d is a divisor of αs and (x1,xu)-path is sequential according to xu+1.

In the following part of V. considered path (in Td) are taken according to vi+1.

Assume that j+1ij. The mapping σ courses that (vj+1,vi)-path, (r,vij1)-path are twins and Tvj+1[r]Tvj+1[vij], Tvj+1[vk]Tvj+1[vk(ij)], k=ij+1,,i, (vj(ij)+1,vj)-path, (vj+1,vi)-path are twins.

Let d be the greatest common divisor of j+1 and ij. First we look at (r,vj)-path and use described above procedure. We obtain that there is a positive integer α1 such that j+1=n1α1, d is a divisor of α1 and Tvj+1[r]Tvj+1[vα1], Tvj+1[vk]Tvj+1[vk+α1], k=1,,j+1α1. We turn back to (vj+1,vi)-path. The mapping σ courses that (vj+1,vi)-path, (r,vij1)-path are twins. Hence Tvi+1[vk]Tvi+1[vk+α1], k=j+1,,iα1 and (vj+1,vj+α1)-path, (r,vα11)-path are twins and (viα1+1,vi)-path, (vijα1,vij1)-path are twins. Vertices of (vijα1,vij1)-path are vertices of (r,vj)-path, too. Hence (vijα1,vij1)-path, (vij,vij1+α1)-path are twins. Moreover, for (r,vj)-path we have that (vij,vij1+α1)-path, (r,vα11)-path are twins. We conclude that (vj+1,vj+α1)-path, (viα1+1,vi)-path are twins and Tvi+1[vk]Tvi+1[vk+α1], k=j+1,,iα1.

Let ij=mα1+β, where m, β are nonnegative integers and 0β<α1. If β=0, then (r,vi)-path is sequential. If β>0, then d is a divisor of β and we repeat procedure of finding the positive number β1 such that ij=m1β1, d is a divisor of β1 and Tvi+1[vk]Tvi+1[vk+β1], k=j+1,,iβ1.

It is clear that β1<α1. We check (r,vj)-path. Observe that properties associated with the value ij follow the same for β1, that is, Tvj+1[vk]Tvj+1[vk+β1], k=1,,jβ1, Tvj+1[r]Tvj+1[vβ1], (r,vβ11)-path, (vjβ1+1,vj)-path are twins.

Let j+1=nβ1+α where n, α are nonnegative integers 0α<β1. If α=0, then (r,vi)-path is sequential. If α>0, then we once more use described procedure and obtain that there is a positive integer α2 such that j+1=n2α2, d is a divisor of α2 and Tvj+1[vk]Tvj+1[vk+α2], k=1,,jα2, Tvj+1[r]Tvj+1[vα2].

It is obvious that α2<β1. Then we consider (vj+1,vi)-path and so on. We obtain less and less positive integer with d as a divisor. Hence finally we obtain that Tvi+1[v]Tvi+1[vd], Tvi+1[vk]Tvi+1[vk+d], k=1,,id and (r,vi)-path is sequential, hence vivi+1 satisfies (ii).

Assume that j+1<ij. The mapping σ for (vj+1,vi)-path courses that Tvi+1[vk]Tvi+1[vk+j+1], k=j+1,,ij1, (r,vj)-path, (vj+1,v2j+1)-path are twins, for (r,vj)-path courses that (vij,vi)-path, (r,vj)-path are twins. Hence (vj+1,v2j+1)-path, (vij,vi)-path are twins.

Let d be the greatest common divisor of j+1 and ij. If d=j+1, then (r,vi)-path is sequential If d<j+1, then we obtain that there is a positive integer β1<j+1 such that ij=m1β1, d is a divisor of β1, Tvi+1[vk]Tvi+1[vk+β1], k=j+1,,iβ1.

Let us consider (r,vj)-path. Since (r,vj)-path, (vj+1,v2j+1)-path are twins we have that Tvj+1[r]Tvj+1[vβ1], Tvj+1[vk]Tvj+1[vk+β1], k=1,,jβ1 and (r,vβ1)-path, (vj+1,vj+β1)-path are twins. Since (vj+1,vj+β1)-path, (viβ11,vi)-path and (viβ11,vi)-path, (vjβ11,vj)-path are twins, we obtain that (r,vβ1)-path, (vjβ11,vj)-path are twins.

If β1 is a divisor of j+1, then we obtain that (r,vi)-path is sequential. If d<β1, then there is a positive integer α1<β1 such that j+1=n1α1, d is a divisor of α1, Tvj+1[r]Tvj+1[vα1], Tvj+1[vk]Tvj+1[vk+α1], k=1,,jα1. We repeat arguments again and again and finally obtain that (r,vi)-path is sequential.

Subcase 2.6. A={vivi+1,rv1}, i2, (r,v1,,vi,vi+1) is (r,vi+1)-path.

I. B={rvi,v1vi+1}.

Then σ(r)=r. If σ(vi)=v1 then σ(vk)=vi(k1), k=1,,i and hence (v1,vi)-path is symmetric according to vi+1. If σ(vi)v1 then vivi+1 satisfies (iv).

II. B={rvi,rvi+1}.

Then σ(r)=v1. If σ(vi)=r then σ(vk)=vk+1, k=1,,i1. Hence Tvi+1[r]Tvi+1[vk], k=1,,i. Especially, (r,vi)-path is sequential according to vi+1. If σ(vi)r then vivi+1 satisfies (iv).

Subcase 2.7. A={vivi+1,ujuj+1}, i1, j1, ujuj+1,vivi+1E(B[v1]), ujuj+1E(B[vi+1]), vivi+1E(B[vi+1]), (r,v1,,vs,vs+1,,vi,vi+1) is (r,vi+1)-path, (r,v1,,vs,us+1,,uj,uj+1) is (r,uj+1)-path, 1s, si, sj.

Observe that if s=i=j it is impossible to create a new d-nary tree. We may assume that s<i or s<j.

We have several possibilities of choosing B.

I. B={viuj+1,ujvi+1}.

Then σ(r)=r. If σ(vi)=vi then B[uj+1] has to be isomorphic to B[vi+1] and vivi+1 satisfies (iv). If σ(vi)=uj then i=j and (r,vi)-path according to vi+1, (r,uj)-path according to uj+1 are twins. Hence vivi+1 satisfies (v). If σ(vi) is not a vertex of (r,vi)-path and is not a vertex of (r,uj)-path then σ(vi+1)vi+1, σ(vi+1)N(vi) and B[σ(vi+1)]B[vi+1]. Hence vivi+1 satisfies (iv).

II. B={ruj+1,ujvi+1}.

Then σ(r)=vi. If σ(vi)uj then σ(vi+1)vi+1 and hence vi+1N(σ(vi)). Therefore there is a vertex w such that σ(w)=vi+1. Observe that wN(vi) and B[w]B[vi] and then vivi+1 satisfies (iv).

We may assume that σ(vi)=uj. Then j=2s and σ(vk)=vik, k=1,,is, σ(vis+l)=us+l, l=1,,j. Observe that every vertex outside (r,vi)-path and outside (r,uj)-path points a branch in the new d-nary tree (V,(EA)B) isomorphic to a branch pointed in Td. Since the family of branches in Td pointed by all vertices is the same as the family of branches in the new d-nary tree (V,(EA)B) pointed by all vertices, we have that the family of branches pointed by vertices r,v1,,vs,vs+1,,vi,us+1,,uj in Td is the same as the family of branches pointed by vertices r,v1,,vs,vs+1,,vi,us+1,,uj in the new d-nary tree (V,(EA)B). Moreover, we have that the family of branches pointed by vertices us+1,,uj in Td is the same as family of branches pointed by vertices r,,vs1 in the new d-nary tree (V,(EA)B). Since B[uk+1] is a subgraph of B[uk], k=s+1,,j1, in Td, we obtain that (us+1,uj)-path according to uj+1 and (vs1,r)-path according to vs are twins. Then (r,uj)-path according to uj+1 and (vi,r)-path according to vi+1 are twins. Hence i=2s and hence (r,vi)-path is symmetric according to vi+1 and vivi+1 satisfies (i).

III. B={rvi+1,viuj+1}.

Then σ(r)=uj. If σ(vi){r,vi} then σ(vi+1)vi+1 and hence vi+1N(σ(vi)). Therefore there is a vertex w such that σ(w)=vi+1 and wN(vi), B[w]B[vi+1]. Then vivi+1 satisfies (iv). If σ(vi)=vi, then σ(vi1)=vi1 and σ(vi+1){uj+1}N(vi){vi1,vi+1}. Hence B[uj+1]B[vi+1] and vivi+1 satisfies (iv). We may assume that σ(vi)=r. Then (r,vi)-path according to vi+1, (uj,r)-path according to uj+1 are twins and hence i=j. Moreover, vertices of (r,vi)-path are mapped, in proper order, to vertices of (uj,r)-path. Observe that every vertex outside (r,vi)-path and outside of (r,uj)-paths points a branch in the new d-nary tree (V,(EA)B) isomorphic to a branch pointed by it in Td. Since the family of branches in Td pointed by all vertices is the same as the family of branches in the new d-nary tree (V,(EA)B) pointed by all vertices, we have that the family of branches pointed by vertices r,v1,,vi,us+1,,uj in Td is the same as the family of branches pointed by vertices r,v1,,vi,us+1,,uj in the new d-nary tree (V,(EA)B). Since vertices of (r,vi)-path are mapped to vertices of (uj,r)-path, we obtain that the family of branches in Td pointed by vertices us+1,,uj and the family of branches in the new d-nary tree (V,(EA)B) pointed by vertices vs+1,,vi are the same. Moreover, B[uk+1] is a subgraph of B[uk] for k=s+1,,j1 in Td. Hence we obtain that (vs+1,vi)-path according to vi+1 and (us+1,uj)-path according to uj+1 are twins. Then vivi+1 satisfies (v).

Subcase 2.8. A={vivi+1,rw1}, i1, (r,v1,,vi,vi+1) is (r,vi+1)-path.

Then B={rvi+1,viw1} and σ(r)=r. It is easy to see that B[vi+1] is a nonisomorphic subgraph of B[v1]. Hence without loss of generality we may assume that σ(w1)=vi+1, σ(vi+1)=w1. Then B[w1]B[vi+1] and vivi+1 satisfies (iv).

Subcase 2.9. A={vivi+1,wjwj+1}, i1, j1, wjwj+1E(B[v1]), (r,v1,,vi,vi+1) is (r,vi+1)-path, (r,w1,,wj,wj+1) is (r,wj+1)-path.

We have several possibilities of choosing B.

I. B={viwj+1,wjvi+1}.

We repeat the arguments of Subcase 2.7.I., replacing uj by wj and uj+1 by wj+1.

II. B={rwj+1,vi+1wj}.

Then σ(r)=vi. Since σ(vi)vi, we have that σ(vi+1)(N(vi){vi+1}). Observe that σ(vi+1) is not a vertex of (vi,r)-path and is not a vertex of (r,wj)-path because otherwise the branch pointed by vi+1 is a nonisomorphic subgraph of the branch pointed by σ(vi+1) in the new d-nary tree (V,(EA)B), which is impossible. Hence σ(vi) is not a vertex of (vi,v1)-path and (w1,wj1)-path. Since j1, it is impossible that σ(vi)=wj. Hence σ(vi+1)vi+1 and vivi+1 satisfies (iv).

III. B={rvi+1,viwj+1}.

Then σ(r)=wj. If σ(vi)r then σ(vi+1)vi+1 and hence vi+1N(σ(vi)). Hence there is a vertex w such that σ(w)=vi+1, and hence wN(vi), B[w]B[vi+1]. Then we obtain that vivi+1 satisfies (iv). Let us suppose that σ(vi)=r. Then (r,vi)-path according to vi+1, (wj,r)-path according to wj+1 are twins and the branch pointed by r in the new d-nary tree (V,(EA)B) is isomorphic to B[vi]. Observe that it follows that Twj[r]B[vi]. But B[vi] is a nonisomorphic subgraph of Twj[r], a contradiction.

Notes

Peer review under responsibility of Kalasalingam University.

References

  • Froncek D. Hlavacek A. Rosenberg S.J. Edge reconstruction and the swapping number of a graph Australas. J. Combin. 58 2014 1 15
  • M.S. Ross, 2-Swappability and the Edge-Reconstruction Number of Regular Graphs, http://arxiv:1503.01048 [math.CO].