On Convex Graphs Having Plane Spanning Subgraph of Certain Type
Niran Abbas Alia, Gek L. Chiab, Hazim Michman Traoc and Adem Kilicmand
a,c,d\/*Department of Mathematics,
Universiti Putra Malaysia, 43400 Serdang, Malaysia,
b\/Department of Mathematical and Actuarial Sciences,
Universiti Tunku Abdul Rahman, Sungai Long Campus, Malaysia
a,c\/Department of Mathematics, College of Science,
Al Mustansiriyah University, 10052 Filastin Street, Iraq*
Abstract
Motivated by a result of [17], we determine necessary and sufficient conditions on F\/ with ∣E(F)∣≤n−1\/ for which Kn−F\/ admits a g-angulation. For ∣E(F)∣≥n\/, we investigate the possibility of placing F\/ in Kn\/ such that Kn−F\/ admits a g-angulation for certain families of graphs F\/.
1 Introduction and Preliminary
By a geometric graph we mean a graph whose edges are straight line segments. By a convex graph, we mean a geometric graph whose vertices are in convex position.
Let S\/ be a set of n\/ points in general position in the plane. A g*-angulation * of S\/ is a plane graph in which each face interior to the convex hull of S\/ is a g-cycle.
A *convex g-angulation * is a g-angulation on S\/ of n\/ points in convex position in the plane.
We say that Gn is a g-angulation of a graph G(V,E) if E(Gn)⊆E.
In particular, the g-angulation is a triangulation if g=3.
The triangulation existence problem is the following: On a given graph G, decide whether there exists a triangulation of G.
This problem is NP-complete (see [14]). This article extends the problem of an article [17] on triangulability of convex geometric graphs to the g-angulation existence problem of a convex geometric graph G\/ by considering a spanning subgraph Fn of Kn with G=Kn−Fn is the convex graph obtained from Kn\/ by deleting the set of edges of Fn\/.
To decide whether G\/ admits a g-angulation or not, we first characterize the forbidding configurations for any possible g-angulation of Kn.
A configuration of n vertices in convex position (vertices of Kn) having a common edge with each g-angulation of Kn, is a forbidding configuration for g-angulations of Kn.
Determining the smallest size of the forbidding configurations for g-angulations of Kn is a natural Turaˊn-type question, as it is equivalent to determining the largest size of a convex geometric graph does not contain a g-angulation. We answer for the “forbidding configurations for g-angulations” question not only by determining their smallest size which is n−g+1\/, but rather giving a complete characterization of all those forbidding configurations of size at least n−g+1\/ and at most n−1\/.
We present the characterizations Fn,g(∗)\/ and Jn,g(β∗)\/ (β∈{1,2,…,2g−3}\/) that forbidding g\/-angulations of Kn\/.
By Theorem 2, Propositions 1, 2, and 3, we show that Kn−Jn,g(∗)\/ and Kn−Jn,g(β∗)\/ admits no g\/-angulation respectively.
If ∣E(Fn)∣≤n−g+1\/, we show by Theorem 3 that Kn−Fn\/ admits a g-angulation if and only if Fn=Fn,g(∗)\/.
If n−g+2≤∣E(Fn)∣≤n−1\/, we show by Theorem 4 that Kn−Fn\/ admits a g\/-angulation if and only if Fn=Jn,g(β∗)\/.
For the case where Fn\/ has at least n\/ edges, it seems difficult to obtain a characterization on Fn\/ such that Kn−Fn\/ admits a triangulation.
For this we confine our attention to seek for the possibility of arranging certain families of graphs Fn\/ as a convex geometric graph
such that Kn−Fn\/ admits a g-angulation.
If such a configuration exists for Fn\/, then we say that Fn\/ is potentially g-angulable in Kn\/.
Potentially g-angulable graphs are considered in Section 5 where we
(i) determine precisely the value of n\/ for which the n\/-cycle is potentially g-angulable in Kn\/ (Theorem 5), and
(ii) characterize all 2\/-regular graphs which are potentially g-angulable in Kn\/ (Theorem 6).
The potentially g-angulable problem is extended to the regular case in Section 6 where we characterize all 3\/-regular graphs which are potentially 4-angulable in Kn\/ (Theorem 7).
Throughout, we shall adopt the following notations. Unless otherwise stated, the vertices of a convex complete graph Kn\/ will be denoted by v0,v1,…,vn−1\/ in cyclic clockwise ordered. Also, unless otherwise specified, any operation on the subscript of vi\/ is reduced modulo n\/.
Lemma 1
Suppose g≥3\/ and t≥2\/ are natural numbers and assume that n=g+t(g−2)\/. Let F\/ be a subgraph of a convex complete graph Kn\/. Assume that F\/ has at most n−1\/ edges and having no boundary edge of Kn\/. Then Kn−F\/ has an edge of the form vjvj+g−1\/.
Proof: If the lemma is not true, then it implies that vj−g+1vj,vjvj+g−1∈E(F)\/, and recursively, this implies that F\/ is a spanning subgraph of Kn\/ with minimum vertex-degree at least 2\/. But this implies that ∣E(F)∣≥n\/, a contradiction.
2 g-angulable graphs
Theorem 1
:* Suppose g≥3\/ is a natural number and let n=g+t(g−2)\/ where t≥0 is any natural number. Let F\/ be a subgraph of the complete convex graph Kn. Suppose F\/ contains no boundary edge of Kn\/ and ∣E(F)∣≤n−g\/. Then Kn−F\/ admits a g-angulation.*
Proof: We prove this by induction on t\/.
The result is trivially true if t=0\/. If t=1\/, then n=2(g−1)\/ and F\/ has at most g−2\/ edges. Hence Kn−F\/ has a vertex, say vi\/ which is adjacent to every other vertex in Kn−F\/ and this means that vivi+g−1\/ together with the boundary edges of Kn\/ form a g\/-angulation of Kn−F\/.
Assume that t≥2\/ and the result is true for all natural numbers t′\/ where t′<t\/. Since g≥3\/, by Lemma 1, Kn−F\/ contains an edge of the form vivi+g−1\/. By relabeling the vertices of Kn−F\/, if necessary we may assume that i=0\/.
If, for some j∈{1,2,…,g−2}\/, vj\/ is adjacent to every other vertex in Kn−F\/, then vjvj+kg−2k+1\/ where k=1,2,…,t\/ together with the boundary edges of Kn\/ form a g\/-angulation of Kn−F\/.
Hence we assume that, for each j∈{1,2,…,g−2}\/, vj\/ is incident with at least one edge of F\/. Then the subgraph obtained from Kn−F\/ by deleting all the vertices v1,v2,…,vg−2\/ is a convex graph of the form Km−F′\/ where m=g+(t−1)(g−2)\/ and F′=F−{v1,v2,…,vg−2}\/. Moreover, F′\/ has at most m−g\/ edges (since there are at least g−2\/ edges of F\/ incident to the vertices v1,v2,…,vg−2\/).
By induction Km−F′\/ admits a g\/-angulation G′\/. As such, G′∪{v0v1v2⋯vg−1}\/ is a g\/-angulation for Kn−F\/.
This completes the proof.
Remark 1
:*
The result in Theorem 1 is tight with respect to ∣E(F)∣\/ since deleting n−g+1\/ edges from Kn\/ does not guarantee that the resulting graph admits a g\/-angulation. This follows directly from the next result in the next section.*
3 Graphs with at most n−g+1\/ edges
This section presents a characterization of Fn,g(∗)\/, that shares any possible g-angulation of Kn with at least one edge.
We show that G\/ admits no g-angulation when G=Kn−Fn,g(∗)\/, G\/ admits a g-angulation if and only if Fn=Fn,g(∗)\/.
Throughout, we let dG(v)\/ denote the degree of v\/ in the graph G\/.
Definition 1
Suppose g≥3\/ is a natural number and let n=g+t(g−2)\/ where t≥1 is any natural number.
Let Fn,g(∗)\/ denote a convex geometric graph with no isolated vertices and having n−g+1\/ edges such that
(i) for any vivj\/ in Fn,g(∗)\/, ∣j−i∣≡1 (mod (g−2))\/,
(ii) whenever d(vi)≥2\/, then vjvj−g+1\/ is an edge of Fn,g(∗)\/ for j=i+1,…,i+g−2\/,
(iii) the neighbor vi\/ of any pendant vertex satisfies the condition that vjvj−g+1\/ is an edge of Fn,g(∗)\/ for j=i+1,…,i+g−2\/,
(iv) for any two pendant vertices vr,vs\/ such that whenever vrvi\/ and vsvj\/ crosse each other in Fn,g(∗)\/, then ∣i−j∣≤g−2\/.
It is easy to see that in the definition, conditions (ii) and (iii) imply that all the pendant vertices of Fn,g(∗)\/ are in consecutive order. As such we may label the vertices of Fn,g(∗)\/ so that d(vi)=1\/ if and only if i=k,k+1,…,n−1\/ where k\/ is a natural number with g−2≤k≤n−2(g−1)\/ if t≥2\/, and k=0\/ if t=1\/.
Examples of a geometric graph Fn,g(∗)\/ are depicted in Figure1 (a), (b), (c) and (d).
In Figure1(a), F6,4(∗) with t=1\/, in (b) F12,4(∗)\/ with minimum value of k=g−2=2\/, in (c) F12,4(∗) with maximum value of k=n−2(g−2)=6, and in (d) F12,4(∗) satisfies condition (iv) of Definition1.
Theorem 2
Suppose g≥3\/ is a natural number and let n=g+t(g−2)\/ where t≥1 is any natural number.
Then Kn−Fn,g(∗)\/ admits no g-angulation.
Proof:
We first observe that, when t=1\/, we have n=2g−2\/ and since Fn,g(∗)\/ consists of g−1\/ edges of the form vjvj+g−1\/, j=0,1,…,g−2\/, Kn−Fn,g(∗)\/ admits no g-angulation.
Hence assume that t≥2\/.
Suppose Kn−Fn,g(∗)\/ admit a g\/-angulation Gn\/.
Note that any g-angulation on a set of points in convex position has a diagonal vqvq+g−1\/.
Let Fm=Fn,g(∗)−{vq+1,…,vq+g−2}\/ where m=n−g+2\/.
By the definition of Fn,g(∗)\/, we see that dFn,g(∗)(vi)=1\/ for each i=q+1,q+2,…,q+g−2.
It is readily checked that Fm\/ is of the form Fm,g(∗)\/ where m=g+(t−1)(g−2)\/. By induction, Km−Fm\/ admits no g-angulation, which is a contradiction with Gn is a g-angulation for Kn−Fn,g(∗)\/.
This completes the proof.
Theorem 3
:*
Suppose g≥3\/ is a natural number and let n=g+t(g−2)\/ where t≥1 is any natural number. Suppose Fn\/ is a subgraph of the convex complete graph Kn\/ such that ∣E(Fn)∣≤n−g+1\/ and Fn\/ contains no boundary edges of Kn\/. Then Kn−Fn\/ admits a g-angulation unless
Fn=Fn,g(∗)\/.*
Proof: In view of Theorem 2, we assume that Fn=Fn,g(∗)\/. Also, we may assume that Fn\/ contains no isolated vertices (otherwise Kn−Fn\/ admits a g-angulation).
We prove the result by induction on t\/. The result is clearly true for t=1\/. Assume that n=g+t(g−2)\/ with t≥2 and the result is true for all convex graphs Km−Fm\/ where m=g+t′(g−2)\/ with t′≤t−1.
By Lemma 1, Kn−Fn\/ contains an edge of the form vqvq+g−1\/ which is not an edge of Fn\/.
Let H\/ be the convex graph obtained from Kn−Fn\/ by deleting the vertices vq+1,…,vq+g−2. Clearly, H\/ is of the form Km−Fm\/ where m=g+(t−1)(g−2)\/.
If Fm=Fm,g(∗)\/, then Km−Fm\/ admits a g-angulation (by induction) and this implies that Kn−Fn\/ admits a g-angulation.
Hence assume that Fm=Fm,g(∗)\/. As such, the vertices vq+1,…,vq+g−2 are pendant vertices in Fn\/ (otherwise ∣E(Fm)∣≤m−g\/, a contradiction with assumption Fm=Fm,g(∗)\/).
There are two cases to consider. Let S={q+1,…,q+g−2}\/.
Case (1): m=2g−2\/.
For each i∈S\/, vi\/ is adjacent either (i) to vi+g−1\/ or else (ii) to vi−g+1\/.
If there exist i,j∈S\/ such that
i<j\/ and vi\/ is adjacent to vi+g−1\/ and vj\/ is adjacent to vj−g+1\/ (∗)\/
then vivi−g+1\/ and vjvj+g−1\/ are the diagonals of a required g\/-angulation.
Hence assume that no i,j∈S\/ satisfy the condition (∗)\/.
Let i\/ be the largest integer in S\/ such that vi\/ is adjacent to vi−g+1\/. This implies that vj\/ is adjacent to vj−g+1\/ if j<i\/, and vj\/ is adjacent to vj+g−1\/ otherwise. But this means that Fn\/ is Fn,g(∗)\/, a contradiction.
On the other hand, if there exists no i∈S\/ such that vi\/ is adjacent to vi−g+1\/, then vj\/ is adjacent to vj+g−1\/ for all j∈S\/. But again Fn\/ is is Fn,g(∗)\/, a contradiction.
Case (2): m≥3g−4\/.
Suppose there is an i∈S\/ such that vi\/ is adjacent to a vertex vj\/ in Fm\/ with ∣j−i∣≡1 (mod (g−2))\/. Then the diagonals
vivi+r(g−2)+1\/, r=1,2,…,t\/ together with the boundary edges of Kn\/ yield a g\/-angulation of Kn−Fn\/.
Hence assume that for any i∈S\/, vi\/ is adjacent to a vertex vj\/ in Fm\/ such that ∣j−i∣≡1 (mod (g−2))\/.
(i) Suppose both vq and vq+g−1 are non-pendent vertices in Fm,g(∗)\/.
If vq+1\/ is not adjacent to vq−g+2\/ in Fn\/, then obtained by deleting all the vertices vq−g+3,…,vq\/ is of the form Km−Fm′\/ where m=n−g+2\/, ∣E(Fm′)∣≤m−g\/ (because dFn(vq)≥2\/) and Fm′\/ contains no boundary edges of Km\/. By Theorem 1, Km−Fm′\/ has a g\/-angulation which together vq−g+1vq−g+2⋯vq+1\/ form a g\/-angulation for Kn−Fn\/.
If vq+1\/ is adjacent to vq−g+2\/ in Fn\/, then vq+1\/ is not adjacent to vq+g−2\/ in Fn\/. In this case, we consider the convex graph obtained by deleting the vertices vq+2,vq+3,…,vq+g−1\/ and apply similar argument before to conclude that Kn−Fn\/ admits a g\/-angulation.
(ii) Suppose only one of vq or vq+g−1 is a non-pendent vertex in Fm,g(∗)\/.
We can assume without loss of generality that vq (since we can relabel the vertices of Kn−Fn\/). Then vq+g−1\/ is a pendant vertex in Fn\/. If for some i∈S\/, vi\/ is not adjacent to vi−g+1\/ in Fn\/, then by the method similar to case in (i), we see that Kn−Fn\/ admits a g\/-angulation. On the other hand, if vi\/ is adjacent to vi−g+1\/ in Fn\/ for all i∈S\/, then Fn\/ is
Fn,g(∗)\/, a contradiction.
(iii) Suppose both vq and vq+g−1 are pendent vertices in Fm,g(∗)\/.
If i∈S\/, we let vsi be the neighbor of vi in Fn\/.
Suppose there exist i,j∈S\/ such that
i<j\/ and vivsi\/ and vjvsj\/ crosse in Fn\/ with sj−si>g−2\/ (⋆)\/
Let H1\/ and H2\/ denote the convex subgraphs of Kn−Fn\/ induced by the vertices
vsi+a,…,vsj,…,vi,…,vj\/ and vj,…,vsi,…,vsi+a\/ respectively
where a∈{1,…,g−2}\/ with ∣si+a−j∣≡1(mod (g−2))\/.
Then H1\/ and H2\/ each admits a g\/-angulation (since vi and vj is adjacent to every other vertex of H1\/ and H2\/ respectively) which together yields a g\/-angulation for Kn−Fn\/.
Hence we assume that no i,j∈S\/ satisfy the condition (⋆)\/.
Suppose that for some i∈S\/, vsi−jvsi−j+g−1∈/E(Fn) for some j∈{1,…,g−2}\/.
Then the subgraph obtained by deleting g−2\/ vertices vsi−j+1,…,vsi,…,\linebreakvsi−j+g−2\/ (from Kn−Fn\/) is of the form Km−Fm′\/ where m=n−g+2 with ∣E(Fm′)∣≤m−g\/ (because dFn(vsi)≥2\/) and Fm′\/ contains no boundary edges of Km\/. By Theorem 1, Km−Fm′\/ has a g\/-angulation which together with vsi−j⋯vsi⋯vsi−j+g−1\/ form a g\/-angulation for Kn−Fn\/.
Hence we assume that for any i∈S\/, vsi−jvsi−j+g−1\/ is an edge in Fn\/ for any j∈{1,…,g−2}\/.
Suppose there is a pendant vertex vr\/ such that r∈S\/ and the edge vrvs\/ (incident to vr\/) crosses vivsi\/ (i∈S\/) with ∣s−si∣>g−2\/, then again a g\/-angulation of Kn−Fn\/ can be constructed as in the previous case (where the condition (⋆)\/ is satisfied).
Hence, for any pendant vertex vr\/ where r∈S\/, vrvs\/ does not crosse vivsi\/ with ∣s−si∣>g−2\/ for any i∈S\/. But this implies that Fn\/ is Fn,g(∗)\/, a contradiction.
This completes the proof.
4 n−g+1+μ\/ edges
We present in this section a characterization of Jn,g(β∗)\/ (where β∈{1,2,…,2g−3}\/) of size at most n−1\/ that shares any possible g-angulation of Kn\/ by at least one edge.
We show that G=Kn−Fn\/ admits a g-angulation if and only if Fn=Jn,g(β∗)\/.
Definition 2
Suppose g≥3\/ is a natural number and let n=g+t(g−2)\/ where t≥1 is any natural number.
Let Jn,g(1∗)\/ denote a convex geometric graph with n\/ vertices and n−g+1+μ\/ edges
such that Jn,g(1∗)−{e1,e2,…,eμ}=Fn,g(∗)\/ for some μ edges e1,e2,…,eμ. Here 1≤μ≤g−2.
Proposition 1
Kn−Jn,g(1∗)\/* admits no g-angulation for any natural number n=g+t(g−2)\/ with t≥1.*
Proof: Suppose on the contrary that Kn−Jn,g(1∗)\/ admits a g-angulation Gn\/. Since
Jn,g(1∗) contains μ edges {e1,e2,…,eμ} such that Jn,g(1∗)−{e1,e2,…,eμ}=Fn,g(∗)\/, it follows that Kn−Fn,g(∗)\/ admits Gn\/, which is a contradiction.
This completes the proof.
Suppose g≥3\/ is a natural number and let n=g+t(g−2)\/ where t≥0 is any natural number. Let F\/ denote a subgraph of Kn\/.
A pair of vertices {vr,vs}\/ is called an α\/*-pair * in F\/ if whenever pair of edges vivr,vjvs\/ (with ∣r−i∣≡1(mod (g−2))\/ and ∣s−j∣≡1(mod (g−2))\/), then vivr\/ crosses vjvs\/ with ∣i−j∣>g−2\/.
A vertex vi∈V(F) is called a g*-angulable vertex * in F\/ if whenever edge vivj\/ (if there exists) in F\/ satisfies that ∣j−i∣≡1 (mod (g−2))\/.
Definition 3
Suppose g≥3\/ is a natural number and let n=g+t(g−2)\/ where t≥2 is any natural number.
Let Jn,g(2∗)\/ denote a convex geometric graph with n\/ vertices and n−g+1+μ\/ edges (where 1≤μ≤g−2\/) with no g\/-angulable vertex in Kn−Jn,g(2∗)\/
such that
(i) if {vr,vs}\/ is an α\/-pair of vertices in Jn,g(2∗)\/ (which occurs at most once), then vr,vs\/ are adjacent in Jn,g(2∗)\/ with ∣r−s∣≡1(mod (g−2))\/, and
(ii) whenever vivi+g−1∈E(Jn,g(2∗)) and {vi+1,…,vi+g−2}\/ contains non-pendant vertex (which is a vertex of an α\/-pair or a vertex in the neighbor of a vertex of an α\/-pair in case Jn,g(2∗)\/ has an α\/-pair),
then Jn,g(2∗)−{vi+1,…,vi+g−2}\/ is either
Fm,g(∗)\/ or Jm,g(1∗)\/ (where m=n−g+2\/).
Remark 2
Suppose vivi+g−1∈E(Jn,g(2∗)) and
either all vertices in {vi+1,\linebreak…,vi+g−2}\/ are pendant (in case Jn,g(2∗)\/ has no α\/-pair)
or {vi+1,…,vi+g−2}\/ contains a non-pendant vertex but neither a vertex of an α\/-pair nor a vertex in the neighbor of a vertex of an α\/-pair (in case Jn,g(2∗)\/ has an α\/-pair),
then it is easy to see that Jn,g(2∗)−{vi+1,…,vi+g−2}\/ is Jm,g(2∗) where m=n−g+2\/.
See Figure 12.
An example of a geometric graph Jn,g(2∗)\/ with and without α\/-pair is depicted in Figure 2 (a) and (b) respectively. In Figure 2 (a), {v6,v10}\/ is the α\/-pair in J14,5(2∗)\/.
Proposition 2
Kn−Jn,g(2∗)\/* admits no g-angulation for any natural number n=g+t(g−2)\/ with t≥2.*
**Proof:**We prove this by induction on n\/.
Consider the case t=2\/. Here n=3g−4\/.
Lemma 1 asserts the existence of an edge vqvq+g−1\/ in Kn−Jn,g(2∗)\/. Choose q\/ such that {vq+1,…,vq+g−2}\/ contains non-pendant vertex. By definition Jn,g(2∗)−{vq+1,…,vq+g−2}=Fm\/ is either Fm,g(∗)\/ or Jm,g(1∗)\/, Km−Fm\/ admits no g\/-angulation by Theorem 2 and Proposition 1. Here m=n−g+2\/.
Assume on the contrary that Kn−Jn,g(2∗)\/ admits a g-angulation Gn\/. Then Gn\/ does not contain the edge vqvq+g−1\/ (otherwise this implies that Km−Fm\/ admits g\/-angulation, a contradiction).
Since t=2\/, Gn\/ has only two diagonal edges. Clearly at least one of these diagonal edges, say e1\/ is incident to a vertex in {vq+1,…,vq+g−2}\/. This is clearly not possible if all vertices in {vq+1,…,vq+g−2}\/ are non-pendant. Hence assume that e1\/ is a incident to a pendant vertex in {vq+1,…,vq+g−2}\/. That is e1=vjvr\/ where j∈{q+1,…,q+g−2}\/. This means that vjvs\/ is an edge of Jn,g(2∗)\/ (where ∣r−s∣=g−2\/) and vs\/ is a non-pendant vertex in Jn,g(2∗)\/ (otherwise vs\/ is an isolated vertex in Fm=Jn,g(2∗)−{vq+1,…,vq+g−2}\/, a contradiction).
Let vpvp+g−1\/ be the second diagonal edge of Gn\/ (where p>j\/). It is easy to see that vs\/ is one of the vertex in {vp+1,…,vp+g−2}\/ (see for example Figure 3). But this means that vj\/ is an isolated vertex in Jn,g(2∗)−{vp+1,…,vp+g−2}\/ (a contradiction).
Now suppose t≥3\/.
Assume on the contrary that Kn−Jn,g(2∗) admits a g-angulation Gn\/.
Note that any g-angulation on a set of points in convex position has a diagonal vqvq+g−1\/.
Let Fm=Jn,g(2∗)−{vq+1,…,vq+g−2}\/ where m=n−g+2\/.
Then Gn−{vq+1,…,vq+g−2}\/ is a g-angulation for Km−Fm\/.
*Case (1): * Jn,g(2∗)\/ contains no α\/-pair.
(i) Suppose {vq+1,…,vq+g−2}\/ contains non-pendent vertex.
Then by the definition of Jn,g(2∗)\/, Fm\/ is either Fm,g(∗)\/ or Jm,g(1∗)\/.
By Theorem 2 or Proposition 1 respectively, Km−Fm\/ admits no g-angulation, a contradiction.
(ii) Suppose all vertices in {vq+1,…,vq+g−2}\/ are pendant vertices.
Then Fm=Jm,g(2∗)\/. By induction, Km−Fm\/ admits no g-angulation, again a contradiction that Gn is a g-angulation for Kn−Jn,g(2∗)\/.
*Case (2): * Jn,g(2∗)\/ contains an α\/-pair.
If {vq+1,…,vq+g−2}\/ contains a vertex of an α\/-pair or a vertex in the neighbor of a vertex of an α\/-pair, then (by definition of Jn,g(2∗)\/,) Fm\/ is either Fm,g(∗)\/ or Jm,g(1∗)\/.
If {vq+1,…,vq+g−2}\/ contains neither a vertex of an α\/-pair nor a vertex in the neighbor of a vertex of an α\/-pair, then
Fm\/ is Jm,g(2∗)\/.
In any case, by Theorem 2, Proposition 1 or by induction, Km−Fm\/ admits no g-angulation, a contradiction that Gn is a g-angulation for Kn−Jn,g(2∗)\/.
This completes the proof.
Definition 4
Suppose g≥4\/ and γ≥3\/ are natural numbers and let n=g+t(g−2)\/ where t≥3\/ is any natural number.
Let Jn,g(γ∗)\/ denote a convex geometric graph with n\/ vertices and n−g+1+μ\/ edges (where 1≤μ≤g−2\/) with no g-angulable vertex and having only γ−1\/ α\/-pairs of vertices such that
(i) whenever {vr,vs}\/ is an α\/-pair of vertices, then vr,vs\/ are adjacent in Jn,g(γ∗)\/ with ∣r−s∣≡1(mod (g−2))\/, and
(ii) whenever vivi+g−1∈E(Jn,g(γ∗)) and {vi+1,…,vi+g−2}\/ contains a vertex of an α\/-pair or a vertex in the neighbor of a vertex of an α\/-pair, then Jn,g(γ∗)−{vi+1,…,vi+g−2}\/ is either Fm,g(∗)\/ or Jm,g(β∗)\/ where m=n−g+2\/ and β≤γ−1\/.
An example of a convex geometric graph Jn,g(3∗)\/ is depicted in Figure 4.
In Figure 4 (a), {v9,v12} and {v13,v16} are two α\/-pairs in J18,4(3∗)\/,
in (b) {v13,v16} is the α\/-pair in the convex geometric graph J16,4(2∗)\/ which obtained from J18,4(3∗)\/ by deleting two vertices v7\/ and v8\/ (with v7\/ is a neighbor of a vertex (v12\/) of the α\/-pair {v9,v12}\/).
For more instance see Figure 13.
Proposition 3
Kn−Jn,g(γ∗)\/* admits no g-angulation for any natural number n=g+t(g−2)\/ with t≥3.*
Proof: We prove this by induction on n\/.
Consider the case t=3\/. Here n=4g−6\/.
Assume on the contrary that Kn−Jn,g(γ∗) admits a g-angulation Gn\/.
Note that Gn\/ has a diagonal vqvq+g−1\/ for some q\/.
Let Fm=Jn,g(γ∗)−{vq+1,…,vq+g−2}\/ where m=n−g+2\/. Then Gn−{vq+1,…,vq+g−2}\/ is a g-angulation of Km−Fm\/.
Let {vr,vs}\/ be an α-pair in Jn,g(γ∗)\/ and let
vivr\/ and vjvs\/ the edges incident to vr\/ and vs\/ respectively. Here ∣i−r∣≡1(mod (g−2))\/ and
∣j−s∣≡1(mod (g−2))\/. Since ∣i−j∣>g−2\/ and ∣r−s∣=g−1\/.
Since m=n−g+2\/ then t=2\/. Hence, there is no any α-pair in Fm\/.
Thus, {vq+1,…,vq+g−2}\/ contains a vertex of vi,vj,vr,vs\/.
By definition of Jn,g(γ∗)\/, Fm\/ is either Fm,g(∗)\/ or Jm,g(β∗)\/ with β=1,2\/. By Theorem 2 or Proposition 1 or else Proposition 2, Km−Fm\/ admits no g-angulation, a contradiction.
Now suppose t≥4\/.
Assume on the contrary that Kn−Jn,g(γ∗) admits a g-angulation Gn\/.
Then Gn\/ has a diagonal vqvq+g−1\/.
Let Fm=Jn,g(γ∗)−{vq+1,…,vq+g−2}\/ where m=n−g+2\/.
Then Gn−{vq+1,…,vq+g−2}\/ is a g-angulation for Km−Fm\/.
If {vq+1,…,vq+g−2}\/ contains a vertex of an α\/-pair or a vertex in the neighbor of a vertex of an α\/-pair, then (by definition of Jn,g(γ∗)\/,) Fm\/ is either Fm,g(∗)\/ or Jm,g(β∗)\/ where m=n−g+2\/ and β≤γ−1\/.
If {vq+1,…,vq+g−2}\/ contains neither a vertex of an α\/-pair nor a vertex in the neighbor of a vertex of an α\/-pair, then
Fm\/ is Jm,g(γ∗)\/.
In any case, by Theorem 2, Proposition 1, Proposition 2 or by induction, Km−Fm\/ admits no g-angulation, a contradiction that Gn is a g-angulation for Kn−Jn,g(γ∗)\/ (see for example Figure 12).
This completes the proof.
An example of a convex geometric graph Jn,g(γ∗)\/ is depicted in Figure 5.
In Figure 5 (a), J18,6(8∗)\/ has {v10,v15}\/, {v10−ℓ,v15−ℓ}\/ and {v6−ℓ,v1−ℓ}\/ as α-pairs where ℓ=1,2,3\/,
in Figure 5 (b) J18,6(9∗)\/ has {v7+ℓ,v12+ℓ}\/ and {v16+ℓ,v3+ℓ}\/ as α-pairs where ℓ=0,1,2,3\/.
Theorem 4
Suppose g≥3\/ is a natural number and let n=g+t(g−2)\/ where t≥1 is any natural number. Suppose Fn\/ is a subgraph of the convex complete graph Kn\/ such that n−g+2≤∣E(Fn)∣≤n−1\/ and Fn\/ contains no boundary edges of Kn\/. Then Kn−Fn\/ admits a g\/-angulation unless Fn\/ is Jn,g(β∗)\/ for some β∈{1,2,…,2g−3}\/.
Proof: In view of Propositions 1, 2 and 3, we assume that Fn=Jn,g(β∗)\/ for any β≥1\/.
The case g=3\/ has been treated in [17]. Hence we assume that g≥4\/.
We prove the result by induction on t\/.
Suppose t=1\/. If Kn−Fn\/ contains an edge of the form vjvj+g−1\/, then vjvj+g−1\/ together with the boundary edges of Kn\/ is a g\/-angulation of Kn−Fn\/.
On the other hand, if vjvj+g−1\/ is not an edge of Kn−Fn\/ for any j=0,1,…,g−2\/, (that is, vjvj+g−1\/ is an edge of Fn\/), then Fn\/ contains Fn,g(∗)\/ as a subgraph. This contradicts the assumption that Fn=Jn,g(1∗)\/.
Now, assume that t≥2 and the result is true for all convex graphs Km−Fm\/ where m=g+t′(g−2)\/, t′≤t−1.
By Lemma 1, Kn−Fn\/ contains an edge of the form vjvj+g−1\/ which is not an edge of Fn\/.
By relabeling if necessary, we may take j=0.
Now delete the set of vertices v1,…,vg−2 from Kn−Fn\/. Let Km−Fm\/ denote the resulting convex graph. Here m=g+(t−1)(g−2)\/.
If ∣E(Fm)∣≤m−g, then Km−Fm\/ admits a g-angulation by Theorem 1. Clearly, this g\/-angulation gives rise to a g-angulation for Kn−Fn\/.
Hence we assume that ∣E(Fm)∣≥m−g+1.
Suppose Fm\/ is neither Fm,g(∗)\/ nor Jm,g(β∗)\/ for any β\/. By Theorem 3 or by induction
Km−Fm\/ admits a g-angulation. Again this g\/-angulation gives rise to a g-angulation for Kn−Fn\/.
Hence assume that Fm\/ is either Fm,g(∗)\/ or Jm,g(β∗)\/ for some β∈{1,2,…2g−3}\/.
Suppose that {v1,…,vg−2}\/ contains no vertex of an α\/-pair in Fn\/.
If Fm=Fm,g(∗)\/, then Fn\/ is either Fn,g(1∗)\/ or Jn,g(2∗)\/. Either case is a contradiction.
Hence assume that Fm=Jm,g(β∗)\/. If {v1,…,vg−2}\/ contains a neighbor of a vertex of an α\/-pair in Fn\/, then Fn=Fn,g(γ∗)\/ (γ=β+1\/). If {v1,…,vg−2}\/ contains no neighbor of a vertex of an α\/-pair in Fn\/, then Fn=Fn,g(β∗)\/. Either case is a contradiction.
Now assume that {vr,vs}\/ is an α\/-pair in Fn\/ and vr∈{v1,…,vg−2}\/.
If vr,vs\/ are adjacent and ∣r−s∣≡1(mod (g−2))\/ then Fn=Jn,g(β∗)\/ for some β∈{2,…,2g−3}\/, a contradiction. Hence either vr,vs\/ are non-adjacent or else ∣r−s∣≡1(mod (g−2))\/ .
Assume without loss of generality that r<s\/. Let i\/ be the largest integer such that vi\/ is adjacent to vr\/ in Fn\/, and j\/ the smallest integer such that vj\/ is adjacent to vs\/ in Fn\/.
Let k\/ be an integer such that i<k<j\/ and ∣k−s∣≡1(mod (g−2))\/. Then vr\/ (respectively vs\/) is a g\/-angulable vertex in the convex subgraph induced by vk,vk+1,…,vs\/ (respectively vs,vs+1,…,vk\/). This yields a g\/-angulation for Kn−Fn\/.
This completes the proof.
5 Potentially g-angulable graphs
We now look at the possibility of placing a graph F\/ with n\/ vertices and n\/ edges in the convex complete graph Kn\/ so that Kn−F\/ admits a g-angulation. We shall confine our attention to the case where Fn\/ is a 2\/-regular graph.
Definition 5
:*
Let Kn\/ be a convex complete graph with n\/ vertices. F\/ is said to be potentially g-angulable if there exists a configuration of F\/ in Kn\/ such that Kn−F\/ admits a g-angulation.*
Theorem 5
:*
Suppose Fn\/ is an n\/-cycle and g≥4 is a natural number such that n=g+t(g−2). Then Fn\/ is potentially g-angulable if and only if n≥5\/.*
Proof: It is easy to see that Kn−Fn\/ admits no g-angulation if n≤4\/.
For n=6\/ suppose F\/ is a 6\/-cycle. Let F\/ be of the form v0v2v4v1v5v3v0\/, then K6−F\/ admits a g-angulation for g∈{4,6}, when g=4 the diagonal is v2v5\/.
For the rest of the proof, we assume that n≥5\/, where n=6.
When n\/ is odd, let Fn\/ takes the form
[TABLE]
When n\/ is even, let Fn\/ takes the form
[TABLE]
In both cases, the edges v2v3+i(g−2)\/, i=1,…,t\/ together with the boundary edges v0v1v2…vn−1v0\/ is a g-angulation of Kn−Fn\/.
This completes the proof.
Theorem 6
:*
Let Fn\/ be a 2\/-regular graph with n\/ vertices and g≥4 is a natural number such that n=g+t(g−2). Then Fn\/ is potentially g-angulable if and only if n≥5\/.*
Proof: If Fn\/ is connected, the result is true by Theorem 5. Hence we assume that Fn\/ is a union of disjoint cycles.
Let C be a smallest cycle in Fn and let vxvy and vyvz are two edges in C.
Consider F∗\/ to be a union of disjoint cycles in Fn−C and let ∣V(C)∣=p.
(i) If F∗ is a 4-cycle, Take F∗={v1v3v2v4v1}\/.
Insert vx, vy and vz of C into edges v4v1,v1v2,v2v3 respectively. Then K7−F7 admits 7-angulation with t=0.
In case that p=4, insert the fourth vertex vw into the edge v3v4. Then K8−F8 admits 4-angulation with two diagonal edges vyv3 and vyv4 and admits 5-angulation with a diagonal edge vyvw.
(ii) If F∗ is not a 4-cycle, then place F∗ on Kn−p∗ so that F∗ contains no boundary edge of Kn−p∗.
Let v1,v2,…,vn−p denotes the vertices of Kn−p∗.
Insert vx, vy and vz of C into edges v1v2,v2v3,v3v4 respectively.
In case that p>3 insert the rest of vertices of C, which are p−3 vertices, into the edges of the path v4v5⋯vn−pv1 and place C on Kn so that C contains no boundary edge of Kn.
Relabel the vertices of Kn to be u0,u1,u2,…,un−1 with vy=u0.
Hence, we have u0u1+i(g−2), i=1,2,…,t together with the edges u0u1u2u3…un−1u0\/ is a g-angulation of Kn−Fn\/ (since u0 is adjacent only to un−2 and u2).
This completes the proof.
6 Regular graphs
In view of the results in the preceding section, it is natural to ask which regular graph is potentially g-angulable in Kn\/.
Problem: Let r≥3\/ and g≥3\/ be two natural numbers and let G\/ be an r\/-regular graph with n\/ vertices where n=g+(g−2)t\/. It is true that there is a natural number n0(r,g)\/ such that when n≥n0(r,g)\/, then G\/ is potentially g-angulable in the convex complete graph Kn\/?
We believe that the above problem is true. However we do not have a complete answer for this even when restricted to the case r=3\/. Nevertheless we offer the following special case of a 3\/-regular graph which is well-known in the literature.
Suppose n\/ and k\/ are two integers such that 1≤k≤n−1\/ and n≥5\/. The generalized Petersen graph P(n,k)\/ is defined to have vertex-set
{ai,bi:i=0,1,…,n−1}\/ and edge-set E1∪E2∪E3\/ where E1={aiai+1:i=0,1,…,n−1}\/,
E2={bibi+k:i=0,1,…,n−1}\/ and E3={aibi:i=0,1,…,n−1}\/ with subscripts reduced modulo n\/. Edges in E3\/ are called the spokes of P(n,k)\/.
Proposition 4
:*
Suppose 1≤k<n/2\/ and g≥4 is a natural number such that 2n=g+t(g−2). Then the generalized Petersen graph P(n,k)\/ is potentially g-angulable in the convex complete graph K2n\/ where n≥5\/.*
Proof: Let the vertices of K2n\/ be denoted v1,v2,v3,…,v2n\/. We shall pack P(n,k)\/ on K2n\/ so that K2n−P(n,k)\/ admits a g-angulation.
Case (1): k=1\/
P(n,1)\/ consists of two n\/-cycles C=a0a1a2⋯cn−1a0\/ and C′=b0b1b2⋯bn−1b0\/ together with the edges aibi\/, i=0,1,2,…,n−1\/.
Place C\/ on K2n\/ so that C\/ takes the form v2v4v6⋯v2n−4v2n−2v2nv2\/
and that C′\/ takes the form v1v3v5v4v7⋯v2n−3v2n−1v1\/.
Case (2): 1<k<n/2\/
P(n,k)\/ consists of two n\/-cycles C=a0a1a2⋯cn−1a0\/ and C′={bibi+k, i=0,1,2,…,n−1}\/ together with the edges aibi\/, i=0,1,2,…,n−1\/.
Place C\/ and C′\/ on K2n\/ so that C\/ takes the form v2v4v6…v2n−2v2nv2\/
and C′\/ takes the form {vivi+2k : i=1,3,…,2n−1}\/. The operations on the subscripts are reduced modulo 2n\/.
In both cases: Let the spokes take the form vivi+3\/, i=0,2,4,…,2n−2\/. Here also the operations on the subscripts are reduced modulo 2n\/.
In both cases: if t=1, then Kn−Fn\/ has a g-angulation whose diagonal is v1vg.
If t≥2, consider the subgraph H\/ induced by the sets of vertices {vg,vg+1,…,\linebreakv2n−(g−1)}∪{v0,v1}\/.
Since the vertex v0\/ is not adjacent to every vertex in H\/ and ∣V(H)∣=2n−2(g−2), K2n−2(g−2)−H\/ admits a g-angulation G\/. Then G∪{v1v2⋯vg}∪{v2n−(g−1)v2n−(g−1)+1…v2n−1v0}\/ is a g-angulation for K2n−P(n,k)\/.
This completes the proof.
In this part, we shall prove that 3\/-regular graph with n=4+2t\/ vertices where t≥2\/ is potentially 4-angulable.
Lemma 2
:*
Let G\/ be a 3\/-regular graph with n=4+2t\/ vertices where t≥2\/.
Suppose V(G) can be labeled as v1,v2,…,vn such that*
(i) v1va,v1vb,v1vc∈E(G) where a,b,c\/ are distinct odd integers (different from 1\/, and
(ii) ∣i−j∣∈/{1,n−1} whenever vivj∈E(G).
Then G\/ is potentially 4-angulable.
Proof: Place the vertices v1,v2,…,vn\/ of G\/ in convex position and put them in clock wise order. Then
v1v4+2k\/, k=0,1,…,t−1 together with all boundary edges of Kn\/ is a 4-angulation of Kn−G.
In the next lemma, we shall show that all 3\/-regular graphs with at least 8\/ vertices admit a labeling as described in Lemma 2 unless it is the cube Q3\/ (on 8\/ vertices) in which case it has a labeling as shown in Figure 10. If Q3\/, with the given labeling, is placed on convex position (with clockwise order), then v2v5,v1v6\/ together with the boundary edges of K8\/ is a 4-angulation of K8−Q3. This implies that all connected 3\/-regular graphs are potentially 4\/-angulable.
To facilitate the proof of Theorem 7, we shall need to consider 3\/-regular graphs where double edges are allowed. By a 2\/-cycle in a graph, written uvu\/, we mean two edges of the graph of the form uv\/ and vu\/.
Let G\/ be a 3\/-regular graph on n\/ vertices and let e=xy be an edge in G. Suppose NG(x)={x1,x2,y} and NG(y)={y1,y2,x}.
Let Ge\/ be the graph obtained from G−e\/ by replacing the paths x1xx2\/ and y1yy2\/ with the edges x1x2\/ and y1y2\/ respectively. Then Ge\/ is a 3\/-regular graph with n−2\/ vertices.
In the case that G\/ has no 2\/-cycle, it is easy to see that G\/ contains an edge e\/ such that Ge\/ has at most one 2\/-cycle if n≥6\/.
Lemma 3
:*
Let G\/ be a connected 3\/-regular graph with n≥8\/ vertices having at most one 2\/-cycle.*
(i) Suppose G\/ has no multiple edges. Then V(G)\/ admits a labeling as described in Lemma 2 unless G\/ is the cube which has labeling as shown in Figure 10.
(ii) Suppose G\/ has a 2\/-cycle uvu\/. Then V(G)\/ can be labeled as v1,v2,…,vn\/ such that u=v1,v=va\/, v1vb∈E(G)\/ where a,b\/ are distinct odd integers different from 1\/, and that ∣i−j∣∈/{1,n−1} whenever vivj∈E(G).
Proof:
We prove this lemma by induction on n\/.
For n=8\/, we have checked that each cubic graph on 8\/ vertices, except Q3\/ cube admits a labeling on its vertices that satisfies the conditions (i) and (ii) (see Figure 11).
Let G\/ be a 3\/-regular graph on n\/ vertices where n≥10\/.
**Case (1) ** G has no 2\/-cycle.
Let e=xy be chosen such that Ge\/ has at most one 2\/-cycle.
(1.1) Ge has no 2\/-cycle.
If Ge\/ is the cube Q3\/, then G\/ is any one of the four 3\/-regular graphs depicted in Figure 8. Each of these graphs has a labeling that satisfies condition (i) of the lemma.
(b) Suppose v1∈{x1,x2,y1,y2}\/. See Figure 15.
(b1) Suppose {x1,x2}∩{y1,y2}=∅\/.
Let v∈{x1,x2,y1,y2}\/ be such that v\/ is not adjacent to vn−2\/ in Ge\/. Without loss of generality, assume that v=x1\/.
In Ge\/, change the label for v\/ from vi\/ to vn−1\/. Extend this labeling on Ge\/ to a labeling of G\/ by assigning x\/ with vi\/ and y\/ with vn\/.
(b2) Suppose {x1,x2}∩{y1,y2}=∅\/.
Assume that x2=y2\/. Then y1=x1\/.
Suppose that vn−2∈{x1,x2,y1}.
When vn−2=x2, then in Ge\/, change the label of x1 from vi\/ to vn−1\/.
In G\/, assign vi,vn\/ to x,y\/ respectively.
When vn−2=x2, then assume without loss of generality that vn−2=x1\/.
In Ge\/, change the label for y1\/ from vi\/ to vn\/ and change the label for x2\/ from vj\/ to vn−1\/. Extend this labeling on Ge\/ to a labeling of G\/ by assigning y\/ with vi\/ and x\/ with vj\/.
Hence assume that vn−2∈{x1,x2,y1}.
Suppose x2=vn−3. Let v∈{x1,y1}\/, say v=y1\/ be such that v\/ is not labeled with v3\/. Then in Ge\/, change the label for v2\/ to vn\/.
In G\/, assign vn−1,v2\/ to x,y\/ respectively.
Suppose x2=vn−3. Let v∈{x1,y1}\/, say v=y1\/ be such that v\/ is not labeled with vn−3\/. Then in Ge\/, change the labels for vn−2\/ to vn−1\/. In G\/, assign vn,vn−2\/ to x,y\/ respectively.
(1.2) Ge has a 2\/-cycle. See Figure 16.
Here we may assume that x1x2x1\/ is the 2\/-cycle in Ge\/. By induction, Ge\/ has a labeling v1,v2,…,vn−2\/ that satisfies condition (ii) of the lemma. There are two cases to consider.
(a) {x1,x2}∩{y1,y2}=∅\/.
We may assume without loss of generality that x1=v1\/ and x2=va\/.
Let v∈{y1,y2}\/, say v=y1\/ be such that v\/ is not labeled with vn−2\/. In Ge\/, change the label for y1\/ from vi\/ to vn\/. Extend this labeling of Ge\/ to a labeling of G\/ by assigning the label vn−1\/ and vi\/ to x\/ and y\/ respectively.
(b) {x1,x2}∩{y1,y2}=∅\/.
(b1) {x1,x2}={y1,y2}\/.
We may assume that x1=v1,x2=va\/ and x3=vb\/ where x3\/ is the other neighbor of x1\/ in Ge\/. To obtain a required labeling for G\/, we first change the label of a vertex in Ge\/ from vc\/ to vn\/ where c\/ is an odd integer and c∈{1,a,b}\/, and then assign vn−1,vc\/ to x,y\/ respectively.
(b2) x2=y2\/ and x1=y1\/.
Then either (b2.1) x2=v1\/ and x1=va\/ or (b2.2) x1=v1\/ and x2=va\/.
(b2.1) In this case, y1=vb\/. In Ge\/, change the label of y1\/ to vn\/. Then extend this new labeling on Ge\/ to a required labeling of G\/ by assigning vn−1,vb\/ to x,y\/ respectively.
(b2.2) In this case, we consider the label on y1\/.
Suppose y1=vn−2\/. First change the label of a vertex in Ge\/ from vc\/ to vn\/ where c\/ is an odd integer and c∈{1,a,b}\/, and then assign vn−1,vc\/ to x,y\/ respectively.
Suppose y1=vn−2\/. Then in Ge\/, if a=3\/, then change v6\/ to vn\/; if a=3\/, then change v2\/ to vn\/. In any case, in G\/, label x\/ with vn−1\/, and label y\/ with v6\/ and v2\/ respectively.
This gives a required labeling for G\/.
**Case (2) ** G has a 2\/-cycle. See Figure 14.
Let xyx\/ be the 2\/-cycle of G\/. Also, let x1\/ (respectively y1\/) be the other neighbor of x\/ (respectively y\/).
Here let e=xy\/. Then Ge\/ is a connected 3\/-regular graph on n−2\/ vertices having at most one 2\/-cycle.
(2.1) Ge has no 2\/-cycle. In this case, x1\/ and y1\/ are not adjacent in G\/.
If Ge\/ is the cube Q3\/, then G\/ the 3\/-regular graph depicted in Figure 10. which has a labeling that satisfies condition (ii) of the lemma.
Hence we assume that Ge\/ is not the cube Q3\/.
By induction, V(Ge) can be labeled as v1,v2,…,vn−2 which satisfies condition (i) of Lemma 2. We shall use this labeling on V(Ge)\/ to obtain a labeling for G\/ that satisfies the conditions of Lemma 2.
(a) Suppose v1∈{x1,y1}\/.
Assume without loss of generality that x1=v1\/ and y1=va\/. In Ge\/, change the labels of x1,y1\/ to vn−1,vn\/ respectively. Extend this new labeling on V(Ge)\/ to a required labeling of G\/ by labeling x,y\/ with v1,va\/ respectively.
(b) Suppose v1∈{x1,y1}\/.
(b1) vn−2∈{x1,y1}\/.
Suppose x1=vn−2\/.
If no vertex in NGe(vn−2)−y1\/ is vn−4\/, then in Ge\/, change the labels for v1\/ to vn\/, and interchange the labels of vn−3\/ and vn−2\/. Now, extend this new labeling on Ge\/ to a required labeling in G\/ by assigning v1,vn−1\/ to x,y\/ respectively.
If some vertex in NGe(vn−2)−y1\/ is vn−4\/, then in Ge\/, change the labels for x1,v1,vn−3\/ to vn−1,vn,vn−2\/ respectively. Now, extend this new labeling on Ge\/ to a required labeling in G\/ by assigning v1,vn−3\/ to x,y\/ respectively.
(b2) vn−2∈{x1,y1}\/.
Suppose the label of x1\/ is vi\/ where i\/ is odd. Then in Ge\/, change the label of v1\/ to vn\/. In G\/, assign v1,vn−1\/ to x,y\/ respectively.
Hence assume that x1=vi,y1=vj\/ are such that i\/ and j\/ are both even.
Suppose 2∈{i,j}. Assume without loss of generality that i=2.
If v4∈/NGe(x1)−y1\/, then in Ge, interchange the labels of v2\/ and v3\/. Also, change the label of v1\/ to vn\/. Now, extend this new labeling on Ge\/ to a required labeling in G\/ by assigning v1,vn−1\/ to x,y\/ respectively.
If v4∈NGe(x1)−y1\/, then in Ge\/, change the label of x1\/ from v2\/ to vn−1\/. Also, change the label of v1,v3\/ to vn,v2\/ respectively. Now, extend this new labeling on Ge\/ to a required labeling in G\/ by assigning v1,v3\/ to x,y\/ respectively.
Hence assume that 2∈/{i,j} and that i<j.
Let u be the vertex in Ge having the label vi+1. Note that such a vertex exists because i<n−2.
To extend the labeling of Ge\/ to a required labeling for G\/, we first interchange the labels of vi\/ and vi+1\/ (in Ge\/). There are two cases to consider.
Now, if u is not adjacent to a vertex with the label vi−1 in Ge, then change the label for v1\/ to vn\/. In G\/, we assign v1,vn−1\/ to x,y\/ respectively.
If u is adjacent to a vertex with the label vi−1 in Ge, then change the label for v1,vi−1\/ to vn−1,vn\/ respectively.
In G\/, assign v1,vi−1\/ to x,y\/ respectively.
(2.2) Ge has a 2\/-cycle.
In this case, x1x2∈E(G)\/. By induction, Ge\/ has a labeling v1,v2,…,vn−2\/ that satisfies condition (ii) of the lemma. To obtain a required labeling for G\/, we first change the label of the vertex v\/ (which is adjacent to x1\/) in Ge\/ from vb\/ to vn\/, and then extend this new labeling to G\/ by assigning v1,vn−1,vb\/ to x,y,x1\/ respectively.
This completes the proof.
Theorem 7
Let G\/ be a 3\/-regular graph with n=4+2t\/ vertices where t≥2\/. Then G\/ is potentially 4-angulable.
Proof: Suppose G\/ is the cube Q3\/ and has labeling as shown in Figure 10. Then Kn−G\/ admits a 4-angulation that has the diagonal edges v1v6\/ and v2v5\/.
Now suppose that G\/ is a cubic not Q3\/. By Lemma 3, V(G)\/ admits a labeling as described in Lemma 2. Then G\/ is potentially 4-angulable.