On Central-Peripheral Appendage Numbers of Uniform Central Graphs
Sul-Young Choi and Jonathan Needleman
Abstract
In a uniform central graph (UCG) the eccentric verticies of a central vertex is the same for all central verticies. This collection of eccentric verticies is the centered periphery. For a pair of graphs (C,P) the central-peripheral appendage number, Aucg(C,P), is the minimum number verticies needed to be adjoined to the graphs C and P in order to construct a uniform central graph H with center C and centered-periphery P. We compute Aucg(C,P) in terms of the radius and diameter of P and whether or not C is a complete graph. In the process we show Aucg(C,P)≤6 if diam(P)>2. We also provide structure theorems for UCGs in terms of the centered periphery.
1 Introduction
Let G=(V(G),E(G)) be a simple graph. The eccentricity of a vertex v, denoted by e(v), is defined by max{d(u,v):u∈V(G)} where d(u,v) is the distance between two vertices u and v. A vertex x is called an eccentric vertex of v if d(v,x)=e(v); and the set of all eccentric vertices of v is denoted by EC(v). The radius of G, r(G) or r, is the minimum eccentricity of the vertices in G and the diameter of G, diam(G), is the maximum eccentricity of the vertices in G.
A (u,v)-path is called a distance measuring path, or a dmpath, when the length of the path is d(u,v). For a central vertex u, a (u,v)-dmpath is called a radial path if the length of the path is r(G). For any nonempty subset S of vertices in G, ⟨S⟩ represents the induced subgraph of G by S. For terminology not defined in this paper, the reader is referred to [5].
Let Z(G) be the center of the graph G and define the centered periphery of G as
[TABLE]
the set of vertices that are far from the center. The periphery, which is the set of maximally eccentric vertices, coincides with centered periphery for some graphs. However, in many cases they differ. For instance, the graph in figure 1 has periphery {p0,p1,p2,p5,p6,p7} while the centered periphery is {p1,…,p6}.
A graph G is called a uniform central graph, or UCG, if EC(c) is same for all central vertices c of G, i.e. EC(c)=CP(G) for every vertex c in the center. It is known that a radial path in a uniform central graph contains only one central vertex [2]. In the study of UCGs it is also useful to define the set of “intermediate” vertices of the graph as I(G)=V(G)−(Z(G)∪CP(G)).
An appendage number of a graph G is the minimum number of vertices to be added to G to obtain a supergraph H of G so that H satisfies some prescribed properties. Buckley, Miller and Slater [1] studied the appendage number of a graph G satisfying that G is the center of its supergraph. This result is extended by Gu [3], where the supergraph is a UCG with a given G as its center. More recently S. Klavžar, K. Narayankar, and S. Lokesh [4] looked at appendage numbers of graphs where the supergraph is a UCG with G as any subgraph.
In this paper, we study the appendage number for a pair of given graphs (C,P), where the supergraph H is a uniform central graph satisfying ⟨Z(H)⟩=C and ⟨CP(H)⟩=P. We denote the central-peripheral appendage number, Aucg(C,P), as the minimum number of “intermediate” vertices needed to construct such a uniform central graph H. By convention Aucg(C,P)=∞ if there is no graph satisfying the above conditions.
Gu’s results in [3] depend on whether or not the desired center is a complete graph. This distinction also appears in our work. In light of this we organize our paper in the following way. Section 2 focuses on results that do not depend on the center. These results are then applied to graphs with specific centers; complete graphs in section 3 and all other graphs in section 4 to classify central-peripheral appendage numbers. The results in sections 3 and 4 are given in terms of sizes of various coverings of P. In section 5 computes the sizes of these coverings. Section 6 summarizes the previous work and describes the central-peripheral appendage numbers in terms of the diameter and radius of P. Finally, in section 7, using different techniques than Gu, we obtain her results as a corollary of our results.
2 General Centers
This section develops results about the structure of a uniform central graph H in terms of P=⟨CP(H)⟩. We do this by studying coverings of the centered periphery.
A covering of a graph G=(V,E) is a set V={V1,…,Vk} where Vi⊂V with ∪Vi=V. We say k is the size of the covering. A subcovering of a covering V is a subset of V that is also a covering.
Throughout this paper we are interested in coverings satisfying various properties. The simplest and most important of these cinditions is condition A. Let P be a graph with covering P={P1,…,Pk}. We say the pair (P,P) satisfies
Condition A:
if for each 1≤i≤k, there is a vertex p∈/Pi satisfying d(Pi,p)≥2.
When there is no confusion to the graph P, we simply say P satisfies condition A.
Elements of a cover can overlap. It is useful to minimize this overlap in some sense.
Lemma 2.1**.**
Let {P1,…,Pk} be a covering of P satisfying condition A. Then there is a subcovering {P1′,…,Pη′} satisfying condition A and for each i (1≤i≤η)
[TABLE]
That is, for each i there exists p~i∈Pi such that p~i∈Pj for all j=i.
Proof.
Suppose there exists an i satisfying
[TABLE]
Then {Pj:j=i} is a subcovering satisfying condition A.
∎
The main idea of this paper is to take the question of determining Aucg(C,P) and to transfer it to questions about the size of coverings of P satisfying various conditions. This is possible because for any UCG there exists a natural covering on P. It turns out this covering satisfying condition A. These coverings are used to obtain lower and upper bounds on Aucg(C,P). To do this we first introduce some notation.
For a uniform central graph H let
[TABLE]
for 0≤m≤r=r(H). This gives a stratification of H with D0=Z(H) and Dr=CP(H). For an enumeration D1={x1,…,xk} let Li be the set of all radial paths from C to P containing xi
Lemma 2.2**.**
Let H be a uniform central graph with ⟨CP(H)⟩=P. Then the graph H induces a covering on P satisfying condition A.
Proof.
Let H be a uniform central graph with ⟨CP(H)⟩=P. Furthermore, let C=⟨Z(H)⟩ and D1={x1,…,xk} for some k. Let Pi=V(P)∩V(Li). Then {P1,…,Pk} is a covering of P since each radial path from C to P contains a vertex in D1.
Assume {P1,…,Pk} does not satisfy the condition A. Then there is an i satisfying that for every vertex p∈Pi there is a p′∈Pi with d(p,p′)≤1. This shows xi∈V(C), a contradiction.
Since xi∈D1, there is a central vertex c with d(xi,c)=1.
For each vertex v of H not in P, d(c,v)≤r(H)−1 and hence d(xi,v)≤r(H).
If p is a vertex in Pi, d(xi,p)=r(G)−1 from the construction of Pi.
If p is a vertex in P but not in Pi. Then there is a vertex p′∈Pi with d(p,p′)=1 and so
[TABLE]
and so xi∈V(C), a contradiction.
∎
Corollary 2.3**.**
If H is a UCG with P=⟨CP(H)⟩, then r(P)>1.
Proof.
If r(P)=1, then there is q∈V(P) with e(q)=1. Let P={P1,…,Pk} be the induced cover of P. Without loss of generality we may assume q∈P1. Then for all p∈V(P), d(p,P1)≤1 and condition A is not met. Hence by lemma 2.2 P is not a centered periphery for any UCG.
∎
Corollary 2.4**.**
If P is a graph with r(P)≤1, then Aucg(C,P)=∞ for all graphs C.
For H, a UCG with C=⟨Z(H)⟩ and P=⟨CP(H)⟩, we use the induced covering from lemma 2.2 to gain a better understanding of the structure of I(H).
Let D1={x1,…,xk} and {P1,…,Pk} be the induced covering of P defined as in lemma 2.2. Without loss of generality assume {P1,…,Pk′} with k′≤k, is subcovering with non-empty elements . Also let P={P1,…,Pk′′} and {p~1,…,p~k′′}, with k′′≤k′, be as in lemma 2.1. For 1≤m≤r(H)−1 we define the following subsets of Dm.
[TABLE]
Dm and Dm′ are well-defined with respect to H, however Dm′′(P) depends on a choice of subcover. When this choice is clear from context, we simply use Dm′′.
The following proposition describes the structure of I(H) that is the key to the rest of this paper.
Proposition 2.5**.**
Let H be a UCG with C=⟨Z(H)⟩, P=⟨CP(H)⟩, P={P1,…,Pk′′} and {p~1,…,p~k′′} as in lemma 2.1 with respect to the induced covering from lemma 2.2. Then any (xi,p~i)-dmpath and (xj,p~j)-dmpath are disjoint if i=j.
Proof.
For Li a (xi,p~i)-dmpath and Lj a (xj,p~j)-dmpath with i=j, assume there exists a y∈V(Li)∩V(Lj). Let Li′ be the subpath of Li from xi to y and Lj′ the subpath of Lj from y to p~j. The concatenation of Li′ and Lj′ yields a (xi,p~j)-dmpath and so p~j∈Pi, a contradiction.
∎
Corollary 2.6**.**
With the assumptions of proposition 2.5, ∣D1′′∣≤∣Dm′′∣ for 1≤m≤r(H)−1.
Proof.
For each 1≤i≤k′′ there exists Li, a (xi,p~i)-dmpath. Let L={L1,…,Lk′′} and Fm=V(L)∩Dm′′. By proposition 2.5 ∣F1∣=∣Fm∣. Then by the construction of Fm
[TABLE]
∎
We often work with UCGs with few intermediate verticies. For such UCGs we can say more about thier structure.
Lemma 2.7**.**
Let H be a UCG with P=⟨CP(H)⟩ and r=r(H). Furthermore, let P={P1,…,Pk′′} be a subcovering of the induced covering and {p~1,…,p~k′′} an associated set of vertices as in lemma 2.1. If ∣D1′′∣=∣Dr−1′′∣ then for each Pi∈P there is a unique yi∈Dr−1′′ such that yi is adjacent to every vertex in Pi. Furthermore, if i=j, then yi=yj.
Proof.
Let D1′′={x1,…,xk′′}. For each i, let Li be an (xi,p~i)-dmpath and let yi be the vertex on Li adjacent to p~i. Then yi∈Dr−1′′. By proposition 2.5 yi=yj when i=j. Furthermore Dr−1′′={y1,…,yk′′}. We claim yi is adjacent to every vertex in Pi.
For a vertex p∈Pi there is a c∈Z(H) and a (c,p)-radial path L that contains xi. Let y be the vertex on L adjacent to p. Since y∈Dr−1′′, y=yj for some 1≤j≤k′′. Since yj is adjacent to p~j, there is a (xi,p~j)-dmpath of length r−1. Therefore, p~j∈Pi which means j=i, y=yi and yi is adjacent to p.
∎
Lemma 2.8**.**
Let H be a UCG with C=⟨Z(H)⟩ and P=⟨CP(H)⟩ and assume the induced covering P={P1,…,Pk} satisfies the conditions of the subcovering in lemma 2.1. That is D1=D1′′. Then each vertex in C is adjacent to each vertex in D1.
Proof.
Let D1={x1,…,xk} and {p~1,…,p~k} be an associated set of vertices to P as in lemma 2.1. For each vertex c in the center and each p~i, there is a (c,p~i)-radial path. By construction of P and definition of p~i, each (c,p~i)-radial paths must contain xi. Therefore, c and xi are adjacent.
∎
For a graph P, let covA(P) be the smallest size of a covering of P satisfying condition A. Observe, from the definition of condition A it follows that covA(P)≥2 for any graph P with r(P)>1. The following result is crirtical in obtaining lower bounds on Aucg(C,P).
Proposition 2.9**.**
If H is a uniform central graph with C=⟨Z(H)⟩, ⟨CP(H)⟩=P, r=r(H) and κ=covA(P), then ∣I(H)∣≥κ(r−1).
Proof.
Let {P1,…,Pk′′} be a subcover of the induced cover on P from H from lemmas 2.2 and 2.1. By definition of κ and corollary 2.6 it follows that
[TABLE]
Since
[TABLE]
the result follows.
∎
Proposition 2.9 is used to obtain bounds on the radius of a UCG.
Corollary 2.10**.**
If H is a uniform central graph with ⟨CP(H)⟩=P, κ=covA(P) and ∣I(H)∣≤sκ+t where s,t∈N and t<κ, then r(H)≤s+1.
Proof.
Observing (s+1)κ>sκ+t the result follows from the contrapositive of proposition 2.9.
∎
We now construct a graph with center C and centered periphery P. Let C and P be graphs, and ρ∈N. Suppose P={P1,…,Pk} is a covering of P. Define a graph G=G(C,P,P,ρ) as follows:
[TABLE]
and ab is an edge in G if and only if one of the following occurs
-
ab is an edge of C.
2. 2.
ab is an edge of P.
3. 3.
a is a vertex of C and b=xi,1 for some i with 0≤i≤k.
4. 4.
a∈Pi and b=xi,ρ for some i with 1≤i≤k.
5. 5.
a=xi,j and b=xi,j+1 for i and j with 0≤i≤k,1≤j≤ρ−1
Proposition 2.11**.**
For two given graphs C and P, let P={P1,…,Pk} be a covering of P satisfying condition A, and ρ≥min{diam(C),2}. Then G=G(C,P,P,ρ) is a UCG with radius ρ+1, ⟨Z(G)⟩=C and ⟨CP(G)⟩=P.
Proof.
From the construction of G, we can show the following:
- i)
For all j=1,…,ρ and any vertex p in P, d(x0,j,p)≥ρ+2, and so e(x0,j)≥ρ+2 and e(p)≥ρ+2.
2. ii)
For all i and j with 1≤i≤k and 2≤j≤ρ, d(xi,j,x0,ρ)≥ρ+2, and so e(xi,j)≥ρ+2.
3. iii)
Since the covering P satisfies condition A, for each i with 1≤i≤k there exists a vertex p in Pj for some j=i satisfying dP(Pi,p)≥2. Thus d(xi,1,p)≥ρ+2 and e(xi,1)≥ρ+2.
4. iv)
For a vertex c in C, d(c,xi,j)≤ρ, d(c,p)=ρ+1 for all p in P, and d(c,c′)≤2 for any vertex c′ in C. This implies e(c)=ρ+1 and EC(c)=V(P).
Therefore, G=G(C,P,P,ρ) is a UCG with radius ρ+1, ⟨Z(G)⟩=C and ⟨CP(G)⟩=P.
∎
We end this section with a result about conditions when a spanning subgraph of a UCG is still a UCG.
Lemma 2.12**.**
Let H be a UCG with V(C)=Z(H) and let G be a spanning subgraph of H. Then G is a UCG with V(C)=Z(G) and CP(G)=CP(H) if for all c∈V(C) and x∈V(G), dG(c,x)=dH(c,x).
Proof.
Since E(G)⊂E(H), the eccentricity of a vertex v in G are greater than or equal to the eccentricity of v in H. The assumption that dG(c,x)=dH(c,x) for all c∈V(C) and x∈V(G) implies Z(G)=V(C) and EC(c) in G is CP(H). Hence, G is a UCG with V(C)=Z(G) and CP(G)=CP(H).
∎
3 When C=Kn with n≥2
In this section we compute appendage numbers when the center C is a complete graph Kn with n≥2 in terms of the size of a smallest covering for the centered periphery.
We introduce a new condition on coverings. Let P be a graph with covering P={P1,…,Pk}. We say (P,P) satisfies
Condition B:
if for every 1≤i≤k and for each p∈Pi either
- 1.
there is a vertex p′∈Pi with d(p,p′)≥3, or
2. 2.
there is a j=i satisfying d(p,Pj)≥2.
Once again, we say the covering P satisfies condition B when there is no confusion about P. Also define covAB(P) to be the smallest size of a covering of P satisfying both condition A and condition B.
Before determining appendage numbers it is necessary to prove the existence of coverings satisfying conditions A and B.
Lemma 3.1**.**
Let P be a graph with r(P)≥2. Then there exists a covering of P satisfying conditions A and B.
Proof.
Let V(P)={p1,…,pk}. Then P={{p1},…,{pk}} is a covering of P and P satisfies conditions A and B because r(P)≥2.
∎
We now find the appendage numbers in terms of κ=covA(P).
Proposition 3.2**.**
κ≤Aucg(Kn,P)≤κ+1* where κ=covA(P).*
Proof.
By lemma 3.1 there is a covering of P satisfying condition A, so let P={P1,P2,…,Pκ} be a smallest covering of P satisfying condition A. By proposition 2.11, the graph G=G(Kn,P,P,1) is a UCG with radius 2 and ∣I(G)∣=κ+1, and thus Aucg(Kn,P)≤κ+1. The lower bound is from proposition 2.9.
∎
Theorem 3.3**.**
Let κ=covA(P). Then Aucg(Kn,P)=κ if and only if covAB(P)=κ.
Proof.
Suppose covAB(P)=κ and let P={P1,…,Pκ} be a smallestl covering of P with respect to conditions A and B. Also, let G=G(Kn,P,P,1)−{x0,1}. Since P satisfies condition B, one can verify that for G, e(p)≥3 for all p∈V(P). Using a similar argument to proposition 2.11 we can show G is a UCG with r(G)=2, ⟨CP(G)⟩=P, ⟨Z(G)⟩=Kn and ∣I(G)∣=κ.
Now assume Aucg(Kn,P)=κ. Let H be a UCG with P=⟨CP(H)⟩,⟨Z(H)⟩=Kn and ∣I(H)∣=κ. From corollary 2.10, r(H)=2.
Let {P1,…,Pk} be the induced covering of P as in lemma 2.2. We know
[TABLE]
and so k=κ. Let D1={x1,…,xk}, where xi is associated to Pi.
Suppose p is a vertex in Pi. Since e(p)≥3 there is an x∈V(H) with dH(p,x)≥3. Note, x∈Pi, because all vertices in Pi are adjacent to xi. If x∈Pj for some j=i, then condition B-1 is satisfied by definition. Finally, if x=xj∈D1 for some j=i, then dP(p,Pj)≥2 and so P satisfies condition B-2. ∎
In section 5 we determine when covA(P)=covAB(P).
4 When C=Kn.
For this section assume that C is not a complete graph and so diam(C)≥2. The results in this section mirror those of when C=Kn, however conditions on the coverings of P, as well as the proofs, are more technical.
Proposition 4.1**.**
For a given pair of graphs (C,P) , 2κ≤Aucg(C,P)≤2κ+2 where κ=covA(P).
Proof.
For a given pair of graphs C and P, suppose H is a UCG with ⟨Z(H)⟩=C and ⟨CP(H)⟩=P. Then r(H)≥diam(C)+1≥3. From corollary 2.9 we obtain a lower bound Aucg(C,P)≥2κ.
For the upper bound consider G=G(C,P,P,2) where P is a smallest covering of P satisfying condition A. By proposition 2.11 G is a UCG with ∣I(G)∣=2κ+2 and so Aucg(C,P)≤2κ+2.
∎
In the proof of proposition 4.1 we use the fact that G(C,P,P,2) is a UCG so long as P satisfies condition A. We now consider a modification of this graph, and determine conditions that this new graph is a UCG.
Let P={P1,…,Pk} be a covering of P and let
[TABLE]
By construction, d(c,p)=3 for all c∈V(C), p∈V(P), and d(c,x)≤2 for x∈V(G)−V(P). So G is a UCG with center C and centred periphery P if and only if for all u∈V(G)−V(C), e(u)≥4.
Consider the case u=xi,1. If G is a UCG, P must satisfy condition A. This implies e(xi,1)≥4 for all 1≤i≤k.
Next, consider the case u=xi,2 for some 1≤i≤k. Since e(u)≥4 there exists a v∈V(G) such that d(u,v)≥4. Because v∈V(C) and d(u,xj,1)≤3, either v∈V(P) or v=xj,2 for some j=i. If p∈V(P) then v∈Pi and d(v,Pi)≥3. If v=xj,2, then d(Pi,Pj)≥2.
Finally, consider the case u∈Pi for some i. If G is a UCG, then e(u)≥4, and so there is a v∈V(G) such that d(u,v)≥4. Since v∈V(C) and v∈Pi because d(p,p′)≤2 for all p,p′∈Pi, it follows that either v=xj,1, v=xj,2 or v∈Pj for some j=i. If v=xj,1 then d(u,Pj)≥2 for j=i. If v=xj,2 then d(u,Pj)≥3. Finally, if v∈Pj for some j=i, then for all p∈Pj
[TABLE]
We conclude d(u,Pj)≥2. This analysis gives rise to two new conditions on P.
Let P={P1,…,Pk} be a covering of a graph P. We say (P,P) satisfies if for all 1≤i≤k
Condition A′:**
if either
- 1.
there is a p∈Pi satisfying d(Pi,p)≥3, or
2. 2.
there is a j=i satisfying d(Pi,Pj)≥2.
Condition B′:**
if for each p∈Pi there is a j=i satisfying d(p,Pj)≥2.
Condition A*′* arises from e(xi,2)≥4 and condition B*′* from e(p)≥4, for p∈V(P). Note condition A*′* implies condition A, and condition B*′* implies condition B. We often abuse notation and say the covering P satisfies a specified condition.
Let covA′(P) be the smallest size of the covering of P satisfying condition A*′* and covA′B′(P) the smallest size of the covering of P satisfying conditions A*′* and B*′*.
The arguments used to determine conditions A*′* and B*′* are bi-directional. This implies the following proposition.
Proposition 4.2**.**
Let C and P be graphs, P a covering of P, and G=G(C,P,P,2)−{x0,1,x0,2}. Then G is a UCG with center C and centered periphery P if and only if P satisfies conditions A′ and B′.
We are now ready to relate appendage numbers to coverings.
Proposition 4.3**.**
Let κ=covA(P). For a graph P with r(P)>1, covA′B′(P)=κ if and only if Aucg(C,P)=2κ.
Proof.
First, assume covA′B′(P)=κ. Let P={P1,…,Pκ} be a covering of P satisfying conditions A*′* and B*′*. Then by proposition 4.2 the graph
[TABLE]
is a UCG and ∣I(G)∣=2κ and so Aucg(C,P)≤2κ. However, by proposition 4.1, Aucg(C,P)≥2κ and the result follows.
Next, assume Aucg(C,P)=2κ. Let H be a UCG with P=⟨CP(H)⟩, C=⟨Z(H)⟩ and ∣I(H)∣=2κ. Note that r(H)=3 by corollary 2.10.
Let {P1,…,Pk} be the induced covering of P as in lemma 2.2. There exists a subcover P={P1,…,Pk′′} and an associated set of vertices {p~1,…,p~k′′} as in lemma 2.1. Assume D1′′={x1,…,xk′′}, and by corollary 2.6 ∣D2′′∣≥k′′.
Because κ=covA(P) and P satisfies condition A we know
[TABLE]
and so k′′=∣D1′′∣=∣D2′′∣=κ. This implies D1=D1′′,D2=D2′′ and κ=k′′=k.
We now show H contains a spanning subgraph isomorphic to
[TABLE]
First, for each p~i defined above and each central vertex c there is a (c,p~i)-radial path. By construction of p~i this path must contain xi, and thus xi is adjacent to every c in the center by lemma 2.8.
By lemma 2.5 there is an enumeration of D2={y1,…,yk} so that yi is adjacent to both xi and each vertex in Pi. Therefore, G=G(C,P,P,2)−{x0,1,x0,2} is isomorphic to a spanning subgraph of H. By lemma 2.12, G is a UCG and by proposition 4.2, P satisfies conditions A*′* and B*′*.
∎
We now move on to understand when Aucg(C,P)=2κ+1. Let H be a UCG such that ⟨Z(H)⟩=C, ⟨CP(H)⟩=P and ∣I(H)∣=2κ+1. By proposition 4.3, κ=covA′B′(P). Also, by corollary 2.10 r(H)=3, and hence ∣D1∣+∣D2∣=2κ+1. Let {P1,…,Pk} be the induced covering of P through D1={x1,…,xk} as in lemma 2.2. Without loss of generality let P={P1,…,Pk′′}, with k′′≤k, be a subcover with an associated set of vertices {p~1,…,p~k′′} as in lemma 2.1. Then
[TABLE]
and hence κ=k′′.
Since
[TABLE]
either ∣D1∣=κ+1 or ∣D2∣=κ+1.
We first address when ∣D1∣=κ+1 and ∣D2∣=κ.
Proposition 4.4**.**
Assume Aucg(C,P)=2κ+1, and let H be a UCG such that ⟨Z(H)⟩=C, ⟨CP(H)⟩=P, and ∣I(H)∣=2κ+1 where κ=covA(P). Furthermore, assume ∣D1∣=κ+1. Then covA′(P)=κ.
Proof.
By corollary 2.6 it follows that
[TABLE]
Therefore D2=D2′′ and ∣D1′′∣=κ. Since k=∣D1∣=κ+1, xk∈D1 but xk∈D1′′.
Next, we prove xk∈D1′ by showing if xk∈D1′, then P satisfies both conditions A*′* and B*′*. Hence Aucg(C,P)=2κ by proposition 4.3.
Assume xk∈D1′, that is D1=D1′. Because ∣D2∣=∣D2′′∣=κ=∣D1′′∣ we may assume D2={y1,…,yκ} such that each yi is adjacent to xi and all vertices of Pi by lemma 2.7. To understand the structure of Pk define the indexing set
[TABLE]
From the definition of D1′ and I,
[TABLE]
For each j∈I every radial path to p~j must contain xj. Hence, each central vertex c∈V(C) is adjacent to xj∈D1 for j∈I. A similar argument shows that for each c∈V(C) and each i∈I, c is adjacent to either xk or xi.
We now show P satisfies conditions A*′* and B*′*.
First, assume condition B*′* fails. Then there exists an ι and a vertex p∈Pι such that dP(p,Pj)≤1 for all j=ι. Therefore, there is a pj∈Pj such that d(p,pj)≤1, and so the following hold:
- i)
d(p,xj)≤d(p,pj)+d(pj,xj)≤3.
2. ii)
d(p,yj)≤d(p,pj)+d(pj,yj)≤2.
3. iii)
d(p,p′)≤d(p,pj)+d(pj,p′)≤3 for all p′∈Pj.
4. iv)
d(p,xι)=2.
5. v)
d(p,yι)=1.
6. vi)
d(c,p)=3 for c∈V(C).
If ι∈I, then d(p,xk)=2. If ι∈I, then for a j∈I
[TABLE]
since d(pj,xk)=2. Therefore e(p)=3, a contradiction, and so P satisifies condition B*′*.
Next, assume condition A*′* fails. Then there exists an ι such that dP(Pι,p′)≤2 for all p′∈V(P)−Pi and dP(Pι,Pj)≤1 for all j=ι. Then d(yι,p)≤3 for all p∈V(P) and d(yι,yj)≤3 for all j=ι. We obtain a contradiction by showing yι is in the center.
If ι∈I then for each xj∈D1 there is a c∈V(C) that is adjacent to xj. Then yι−xι−c−xj is a path and d(yι,xj)≤3. Similarly when ι∈I and j∈I, d(yι,xj)≤3. If ι,j∈I then yι−xk−yj−xj is a path and so d(yι,xj)≤3. Finally, d(yι,c)=2 for all c∈V(C), and so e(yι)=3, a contradiction and P satisfies condition A*′*.
Since P satisfies conditions A*′* and B*′*, Aucg(C,P)=2κ by proposition 4.3. This contradicts the assumptions of the proposition and hence xk∈D1′ and so D1′=D1′′.
Then xk is not on a radial path because xk∈D1′. Therefore, vertices adjacent to xk are in C or D1′′. Furthermore each yi∈D2 satisfies e(yi)≥4 from H being a UCG with r(H)=3 . Therefore, there exists a u∈V(G) with d(yi,u)≥4. Because d(c,yi)=2 for all c∈V(C), we know u∈V(C).
Because D1′=D1′′, each (c,p~i)-radial path contains xi, and hence each c is adjacent to xi. It follows that for a c∈V(C) adjacent to xj∈D1 that
[TABLE]
which means u∈D1.
If u∈V(P) then d(u,Pi)≥3 and condition A*′-1 is satisfied for i. If u=yj∈D2′′ then dP(Pi,Pj)≥2 and condition A′-2 is satisfied for i. Since these hold for each i, P satisfies condition A′* and covA′(P)=κ.
∎
A weak converse of proposition 4.4 also holds.
Proposition 4.5**.**
If P is a graph with r(P)>1, κ=covA(P)=covA′B′(P) and covA′(P)=covA(P), then Aucg(C,P)=2κ+1.
Proof.
Assume covA′B′(P)=κ but covA′(P)=κ. Let P={P1,…Pκ} be a smallest covering with respect to condition A*′* and
[TABLE]
We claim G is a UCG with C=⟨Z(G)⟩, P=⟨CP(G)⟩ and ∣I(G)∣=2κ+1.
By proposition 4.2 the graph G′=G−{x0,1} is a UCG if and only if P satisfies conditions A*′* and B*′. In G′, if P satisfies conditions A′*, then e(xi,1)≥4 and e(xi,2)≥4. This is still holds in G. Furthermore, for each vertex p∈V(P), d(x0,1,p)=4 in G, and so G is a UCG. Therefore, Aucg(C,P)≤2κ+1, but by proposition 4.3 Aucg(C,P)>2κ.
∎
The case when ∣D2∣=κ+1 is more complicated. To understand this case we consider a new graph G, and determine new conditions for when G is a UCG.
For two graphs C and P, let P={P1,…,Pk} be a covering of P, and Q={Q0,Q1,P2,…,Pk} be a covering such that Q0∪Q1=P1.
Define a graph G=G′(C,P,Q) as follows (see figure 3):
[TABLE]
and ab is an edge in G if and only if one of the following occurs
- i)
ab is an edge of C.
2. ii)
ab is an edge of P.
3. iii)
a is a vertex of C and b=xi for some i with 1≤i≤k.
4. iv)
a∈Pj and b=yj for some j with 2≤j≤k.
5. v)
a∈Ql and b=yl for some l=0,1.
6. vi)
a=xi and b=yi for 1≤i≤k
7. vii)
a=x1 and b=y0.
We now determine conditions on Q for G to be a UCG with C=⟨Z(G)⟩ and P=⟨CP(G)⟩. By construction for all c∈V(C) and p∈V(P), d(c,p)=3, and d(c,x)≤2 for all x∈V(G)−V(P). So G is a UCG with center C and centred periphery P if and only if for all u∈V(G)−V(C), e(u)≥4. Then there exists a vertex v with d(u,v)≥4.
First consider the case u=xi. In every UCG the induced covering satisfies condition A. This implies v∈V(P) and e(xi)≥4 for all 1≤i≤k. Note that, in G the induced covering is P, not Q.
Next, consider the case u=yi for some 2≤i≤k. Since v∈V(C) and d(v,xj)≤3 for 1≤j≤k, either v∈V(P) or v=yj for 0≤j≤k and j=i. The one of the following holds.
- i)
If v∈V(P) then v∈Pi and d(Pi,v)≥3.
2. ii)
If v=yj, 2≤j≤k and j=i, then d(Pi,Pj)≥2.
3. iii)
If v=yl for l=0 or 1, then d(Pi,Ql)≥2.
Now, consider the case u=y0 or y1. Without loss of generality we assume u=y0. Since v∈V(C), v=xj for 1≤j≤k, v=y1 and v∈Q0∪Q1=P1, the following must hold.
- i)
If v∈V(P) then v∈P1 and d(Q0,v)≥3.
2. ii)
If v=yj, 2≤j≤k and j=i, then d(Q0,Pj)≥2.
Now, consider the case u=p∈Pi for some 2≤i≤k. Once again, if G is a UCG, then e(p)≥4, and so there exists a v∈V(G) such that d(u,v)≥4. We know v∈V(C). Also, v∈Pi since d(p,p′)≤2 for all p′∈Pi. Also note, that for j≥2 and j=i, if d(u,yj)≥4 then d(u,xj)≥4. This implies that we do not need to determine the conditions for d(u,yj)≥4. Given this, one of the following must hold.
- i)
If v=xj for j=i and 2≤j≤k, then d(u,Pj)≥2.
2. ii)
If v=x1, then d(u,Q0)≥2 and d(u,Q1)≥2.
3. iii)
If v=yl for l=0 or 1, then d(u,Ql)≥3.
4. iv)
If v=p′∈Pj−Pi for 2≤j≤k and j=i, then d(u,p′)≥4 and d(u,Pj)≥2.
5. v)
If v=p′∈Ql−Pi for l=0 or 1, then d(p,Ql)≥2 and d(u,p′)≥4.
The last case to consider, without loss of generality, is u=p∈Q0. Since d(u,v)≤3 if v∈V(C), v=x1,y0 or y1 or v∈Q0, the following must hold.
- i)
If v=xj, then 2≤j≤k and d(u,Pj)≥2.
2. ii)
If v=yj for 2≤j≤k, then d(u,Pj)≥3.
3. iii)
If v=p′∈Pj−Q0 for 2≤j≤k, then d(u,Pj)≥2 and d(u,p′)≥4.
4. iv)
If v=p′∈Q1−Q0, then d(u,p′)≥4 and d(u,Q1)≥2.
The above discussion is in terms of Q, however the rest of the paper is in terms of P. For this reason now summarize the discussion in terms of two technical conditions on P.
Let P={P1,…,Pk} be a covering of a graph P. For a given ι and two sets Q0 and Q1 such that Q0∪Q1=Pι, let
[TABLE]
We say (P,P,Q) satisfies
Condition A′′:**
- 1.
if for each i=ι one of the following holds
- (a)
there is a p∈Pi satisfying d(Pi,p)≥3, or
2. (b)
there is a j=ι satisfying d(Pi,Pj)≥2, or
3. (c)
there is an l=0 or 1 such that d(Pi,Ql)≥2,
2. 2.
and if for all l=0,1 either
- (a)
there is a p∈Pι satisfying d(Ql,p)≥3, or
2. (b)
there is a j=ι satisfying d(Ql,Pj)≥2.
Condition B′′:**
- 1.
if for each p∈Pi, i=ι one of the following holds
- (a)
there is a j=ι satisfying d(p,Pj)≥2, or
2. (b)
d(p,Q0)≥2 and d(p,Q1)≥2, or
3. (c)
there is an l=0 or 1 such that d(p,Ql)≥3, or
4. (d)
there l=0 or 1 such that d(p,Ql)≥2 and a p′∈Ql so that d(p,p′)≥4.
2. 2.
and if for each l=0,1 and each p∈Ql, either
- (a)
there is a j=ι satisfying d(p,Pj)≥2, or
2. (b)
there exists p′∈Pι−Ql such that d(p,p′)≥4 and d(p,Ql′)≥2, where l′=0 or 1 but l′=l.
When there exists an ι, Q0 and Q1 such that (P,P,Q) satisfies condition A*′′* (resp. condition B*′′), we say (P,P) or simply P satisfies condition A′′* (condition B*′′). Without loss of generality we may renumber P so that ι=1. Note that A′* and B*′* imply A*′′* and B*′′* by taking ι=1 and letting Q0=P1 and Q1=∅. However, condition A*′′* does not imply condition A. So let covAA′′B′′(P) be the smallest size of the covering P of P satisfying conditions A, A*′′* and B*′′*.
Similar to the discussion of conditions A*′* and B*′, the arguments used to determine conditions A′′* and B*′′* from the graph G(C,P,Q) are bi-directional. We summarize the discussion in the following proposition.
Proposition 4.6**.**
For graphs C and P, and a triple (P,P,Q), G=G′(C,P,Q) is a UCG if and only if P satisfies condition A and (P,P,Q) satisfies conditions A′′ and B′′.
We are now ready to prove analogues to propositions 4.4 and 4.5.
Proposition 4.7**.**
Assume Aucg(C,P)=2κ+1, and let H be a UCG with ⟨Z(H)⟩=C, ⟨CP(H)⟩=P, and ∣I(H)∣=2κ+1 where κ=covA(P). Furthermore, assume ∣D2∣=κ+1. Then covAA′′B′′(P)=κ.
Proof.
Let H be a UCG with C=⟨Z(H)⟩, P=⟨CP(H)⟩, ∣I(H)∣=2κ+1 and ∣D2∣=κ+1. Since ∣D1∣=κ and κ≤∣D1′′∣≤∣D1∣, it follows that ∣D1∣=∣D1′′∣. We prove this proposition by studying the structure of a spanning subgraph.
Let D1={x1,…,xκ} and {p~1,…,p~κ} be a set of vertices associated to the induced cover P. By proposition 2.5 there exists an enumeration {y0,…,yκ} of D2 such that the vertex yj is adjacent to xj and p~j for each j, 1≤j≤κ. We may also assume y0 is adjacent to x1.
We now define a different cover of P. For each i, 0≤i≤κ, let
[TABLE]
Let P1′=Q0∪Q1 and Pi′=Qi for 2≤i≤κ, P′={P1′,…,Pκ′} and Q={Q0,…,Qκ}.
Since ∣D1∣=∣D1′′∣ every vertex of C is adjacent to vertex in D1 by lemma 2.8. Therefore, G=G′(C,P,Q) is isomorphic to a spanning subgraph of H, and is a UCG by lemma 2.12. By proposition 4.6 P′ satisfies conditions A, A*′′* and B*′′* which means
[TABLE]
We conclude covAA′′B′′(P)=κ.
∎
Proposition 4.8**.**
If P is a graph with r(P)>1, κ=covA(P)=covA′B′(P) and covAA′′B′′(P)=covA(P), then Aucg(C,P)=2κ+1.
Proof.
Assume covA′B′(P)=κ but covAA′′B′′(P)=κ. By proposition 4.3 Aucg(C,P)>2κ, so we need to show Aucg(C,P)≤2κ+1. Let P be a smallest covering of P with respect to condition A such that there is a refined cover Q where the pair (P,Q) is smallest with respect to conditions A*′′* and B*′′*. Without loss of generality we may assume ι=1. By proposition 4.6 G=G(C,P,Q) is a UCG with C=⟨Z(G)⟩, P=⟨CP(G)⟩ and ∣I(G)∣=2κ+1, which implies Aucg(C,P)≤2κ+1.
∎
The following theorem summarizes propositions 4.1, 4.3, 4.4, 4.5, 4.7, and 4.8.
Theorem 4.9**.**
If P is a graph with r(P)>1 and covA(P)=κ, then following holds:
-
Aucg(C,P)=2κ* if and only if covA′B′(P)=κ.*
2. 2.
Aucg(C,P)=2κ+1* if and only if κ=covA′B′(P) and either*
- (a)
covA′(P)=κ* or*
2. (b)
covAA′′B′′(P)=κ.
3. 3.
Aucg(C,P)=2κ+2* otherwise.*
5 Coverings
In this section we determine when κ=covA(P) is the size of smallest coverings with respect to the other conditions described in sections 3 and 4. To do this we introduce one last set of notation. For a graph G, a vertex V in G and s∈N let
[TABLE]
be the closed s neighborhood of v. When s=1 we simply let N1[x]=N[x].
Proposition 3.2 and theorem 3.3 determine Aucg(Kn,P) up to knowing when covA(P)=covAB(P). We now determine conditions for a graph P to satisfy covA(P)=covAB(P).
Proposition 5.1**.**
If P is a graph with diam(P)≥3, then covA(P)=2.
Proof.
Suppose x is a vertex in P satisfying e(x)≥3. Let P1=N[x] and P2=V(P)−P1. Then d(x,P2)≥2. Since e(x)≥3, P2 contains a vertex y satisfying d(x,y)≥3, and so d(y,P1)≥2. Thus, {P1,P2} is a covering of P satisfying condition A.
∎
Proposition 5.2**.**
If P is a graph satisfying diam(P)≥4 and r(P)≥3, then covAB(P)=2.
Proof.
Let x and y be vertices in P satisfying d(x,y)=4. We construct P1 and P2 recursively. Initialize P1:=N[x] and P2:=N[y]. Note that d(x,N[y])=d(y,N[x])=3. For a vertex z in V(P)−(P1∪P2), update P1 and P2 as follows.
- i)
If there is a vertex p∈P2 satisfying d(z,p)≥3, then let P1:=P1∪{z}.
2. ii)
Else if there is a vertex p∈P1 satisfying d(z,p)≥3, then let P2:=P2∪{z}.
3. iii)
Else there is a vertex p∈V(P)−(P1∪P2) satisfying d(z,p)=3, since r(P)=3. Let P1:=P1∪{z} and P2:=P2∪{p}.
Continue until all vertices of P have been accounted for. By construction {P1,P2} is a covering of P satisfying condition B. Since N[x]⊂P1, d(x,P2)≥2 and similarly d(y,P1)≥2. Thus {P1,P2} satisfies condition A.
∎
Proposition 5.3**.**
If P is a graph with r(P)=2, then covAB(P)=2.
Proof.
Suppose there is a covering {P1,P2} of P satisfying conditions A and B. Assume P1 contains a vertex z with e(z)=2. By condition B, d(z,P2)=2 and so N(z)∩P2=∅. If p is a vertex in P2, then d(z,p)=2 and so d(P1,p)=1, a contradiction to condition A.
∎
Proposition 5.4**.**
For every α∈N, there exists a graph P with r(P)=diam(P)=2 and covA(P)=2α.
Proof.
Construct Pα=(Vα,Eα) as follows. Let
[TABLE]
and
[TABLE]
Then r(Pα)=diam(Pα)=2.
We now show covA(Pα)=2α. By lemma 3.1 there exists a covering
[TABLE]
of Pα satisfying condition A. Suppose e1∈P1α. If P1α contains a vertex other than e1, then d(P1α,x)=1 for all x∈Vα−P1α, a contradiction to condition A. Therefore P1α={e1}. Similarly ∣Piα∣=1 for all i=1,2,…,k and covA(Pα)=2α.
∎
Propositions 5.1 through 5.3 determine when covA(P)=covAB(P) for all graphs P with r(P)>1 except those with r(P)=diam(P)=2 and those with r(P)=diam(P)=3. Proposition 5.4 gives insight into richness of the case r(P)=diam(P)=2. However, we do not have any definitive results for r(P)=diam(P)=3. This is discussed further in section 6.
For a non-complete graph C, theorem 4.9 relates Aucg(C,P) to covA′B′(P), covA′(P), and covAA′′B′′(P). Because of proposition 5.4, we only consider P with diam(P)>2. By proposition 5.1 we need to understand when the smallest coverings are of size 2.
Proposition 5.5**.**
P* is a graph with covA′B′(P)=2 if and only if P is disconnected.*
Proof.
Assume P is disconnected. Let P1 be the vertices of a connected component of P and let P2=V(P)−P1. Since, for all u∈P1 and v∈P2, d(u,v)=∞ it follows that {P1,P2} satisfies conditions A*′* and B*′*.
Next, assume P is connected but P={P1,P2} is a covering. Since P is connected, d(P1,P2)=1 and there is a p1∈P1 and p2∈P2 with d(p1,p2)=1. Therefore d(p1,P2)=1 and d(p2,P1)=1 and condition B*′* fails, and so covA′B′(P)>2.
∎
Proposition 5.6**.**
For a graph P, covA′(P)=2 if and only if diam(P)≥5.
Proof.
Assume diam(P)≥5. Then there exist u,v∈V(P) with d(u,v)≥5. Let P1={p∈V(P):d(u,p)≤2} and P2=V(P)−P1. Then d(u,P2)≥3. Also d(v,P1)≥3 because if there is a p∈P1 with d(v,p)≤2, then d(u,v)≤d(u,p)+d(p,v)≤4, a contradiction. So {P1,P2} satisfies condition A*′*.
Next, assume there is a covering {P1,P2} of P satisfying condition A*′*. Then there is a u∈P1 such that d(u,P2)≥3. Then N2[u]⊂P1 and N2[u]∩P2=∅. Similarly, P2 contains a vertex v such that d(v,P1)≥3, and so N2[v]⊂P1 and N2[v]∩P1=∅. Therefore d(u,v)≥5.
∎
Proposition 5.7**.**
If P is a graph with diam(P)≤3, then covAA′′B′′(P)=2.
Proof.
Let P be a graph with diam(P)≤3. Suppose there is a covering P={P1,P2} of P satisfying condition A such that for ι=1 the triple (P,P,Q) satisfies conditions A*′′* and B*′′, where Q={Q0,Q1,P2} with P1=Q0∪Q1. For each p∈P1, d(p,P2)≥2 by condition B′′*-2 because diam(P)≤3. This implies d(P1,P2)≥2. However, d(P1,P2)≤1 since P is connected, a contradiction.
∎
Proposition 5.8**.**
If P is a graph with r(P)=2, then covAA′′B′′(P)=2.
Proof.
Let P be a graph with r(P)=2. Suppose there is a covering P={P1,P2} of P satisfying condition A such that for ι=1 the triple (P,P,Q) satisfies conditions A*′′* and B*′′*, where Q={Q0,Q1,P2} with P1=Q0∪Q1.
For a central vertex c∈V(P) either c∈P1 or c∈P2.
Suppose c∈P1. Without loss of generality assume c∈Q0. By condition B*′′*-2, d(c,P2)≥2. In fact, d(c,P2)=2 because c is central, and so there exists a p∈P2 with d(c,p)=2. Let c−x−p be a dmpath. Then x∈P1, e(x)≤3 and d(x,P2)=1. Therefore, {P1,P2} does not satisfy condition B′′-2 for x, a contradiction.
Next, suppose c∈P2. By the hypotheses c does not satisfy B*′′-1a, B′′-1c, and B′′-1d. Therefore, d(c,Q0)≥2 and d(c,Q1)≥2. Since c is a central vertex, d(c,Q0)=2 and d(c,Q1)=2. Then there exist q0∈Q0 and q1∈Q1, satisfying d(c,q0)=2 and d(c,q1)=2. Let c−x−q0 and c−y−q1 be paths. Then both x and y are in P2 and d(P2,Qi)≤1 for i=0,1. Therefore {P1,P2} does not satisfy condition A′′*-1.
∎
To fully understand Aucg(C,P) there are still two cases left to consider, when diam(P)=4 and r(P)=3, and diam(P)=4 and r(P)=4. In these cases covA′B′(P)=2 and covA′(P)=2. When diam(P)=r(P)=4, covAA′′B′′(P)=2 depends on P. To show that there exist graphs P with covAA′′B′′(P)=2, we introduce the following lemma.
Lemma 5.9**.**
If P is a graph with diam(P)=4 and covAA′′B′′(P)=2, then there exist three vertices x1,x2,x3 such that
[TABLE]
Proof.
Suppose there is a covering P={P1,P2} of P satisfying condition A such that for ι=1 the triple (P,P,Q) satisfies conditions A*′′* and B*′′*, where Q={Q0,Q1,P2} with P1=Q0∪Q1.
Since P is connected without loss of generality we may assume that d(P2,Q1)≤1. Then A*′′-2b is not satisfied for l=1 and so, from A′′-2a, there exists a p1∈P2 such that d(Q1,p1)≥3. For l=0 there are two cases for condition A′′*-2, either part 2a is met or part 2b is met.
For l=0, either there exists a p0∈P2 such that d(Q0,p0)≥3 or d(P2,Q0)≥2 by condition A*′′*-2b.
Suppose d(P2,Q0)≥2, by connectivity of P we know d(P2,Q1)≤1 and d(Q0,Q1)≤1. Since d(p1,Q1)≥3, N2[p1]∩Q1=∅ and so
[TABLE]
Since p1∈P2 and d(Q0,P2)≥2, N1[p1]∩Q0=∅ and N1[p1]⊂P2. From
[TABLE]
it follows that N2[p1]⊂P2. Therefore, d(N2[p1],Q0)≥2 and d(p1,Q0)≥4. Since diam(P)=4, d(p1,Q0)=4.
Let p1−x−y−z−q0 be a dmpath for some q0∈Q0. Then z∈Q0 since d(z,p1)=3, and z∈P2 since d(q0,z)=1. Hence z∈Q1. Furthermore, y∈N2[p1]⊂P2, so d(z,P2)≤1 and d(z,Q0)≤1. This implies condition B*′′*-2 is not met for z∈Q1, a contradiction. Therefore d(P2,Q0)≤1.
Then, for l=0 condition A*′′*-2a is satisfied and there is a p0∈P2 such that d(Q0,p0)≥3. Therefore, N2[p0]∩Q0=∅. Since N2[p1]∩Q1=∅, it follows that N2[p0]∩N2[p1]⊂P2.
From condition A*′′*-1 there exists a q∈Q0 such that d(q,P2)≥3. Then N2[q]∩P2=∅ and
[TABLE]
∎
Proposition 5.10**.**
There exists a graph P with r(P)=diam(P)=4 and covAA′′B′′(P)=2.
Proof.
Let P be the graph of a hexagonal prism as in figure 4. Then r(P)=diam(P)=4. If covAA′′B′′(P)=2 then by lemma 5.9 there exist vertices x1,x2,x3 such that
[TABLE]
or equivalently,
[TABLE]
For any vertex x of P, the compliment of N2[x] is a star with three pendant vertices as in figure 4. Because the prism has twelve vertices, a set of three stars would cover the prism with no overlap. However, one can check this is not possible and so covAA′′B′′(P)=2
∎
Proposition 5.11**.**
There exists a graph P with r(P)=diam(P)=4 and covAA′′B′′(P)=2.
Proof.
Let P be the graph of a heptagonal prism as in figure 5. Then r(P)=diam(P)=4. Consider the covering P={P1,P2}, where P1=Q0∪Q1 as in figure 5. Here vertices of Q0 are represented by the open circles, Q1 by the open squares, and P2 by the filled circles. One can verify this covering satisfies conditions A, A*′′* and B*′′*, and so covAA′′B′′(P)=2.
∎
6 Appendage Numbers
In this section we determine Aucg(C,P) for most pairs of graphs (C,P) based on the structure of C. Conjectures are also given for the two remaining cases.
Theorem 6.1**.**
Let C={v} and P be any graph with r(P)≥2. Then Aucg(C,P)=0.
Proof.
Let H be the graph with vertex set
[TABLE]
and edge set
[TABLE]
We claim H is a UCG with ⟨Z(H)⟩=C, ⟨CP(H)⟩=P and ∣I(H)∣=0.
Since v is adjacent to all other vertices in H, e(v)=1. If p is a vertex in P
there is a p′∈V(P) satisfying dP(p,p′)=2. Since p and p′ are not adjacent in P, they are not adjacent in H and e(p)≥2. This implies ⟨Z(H)⟩={v} and H is UCG with ⟨CP(H)⟩=P.
∎
Theorem 6.2**.**
When n≥2
[TABLE]
Furthermore, for all t∈N there is a graph P with diam(P)=r(P)=2 and Aucg(Kn,P)≥t.
Proof.
This follows directly from proposition 3.2, theorem 3.3 and propositions 5.1, 5.2, 5.3, and 5.4.
∎
Conjecture 1**.**
If diam(P)=r(P)=3, then Aucg(Kn,P)=2.
When diam(P)=r(P)=3, Aucg(C,P)=2 or 3. However, all our examples show Aucg(Kn,P)=2, but we have not been able to prove this is always true.
Theorem 6.3**.**
If C is a graph with diam(C)>1 then
[TABLE]
Furthermore, for all t∈Z there is a graph P with diam(P)=r(P)=2 and Aucg(C,P)≥t.
Proof.
This follows directly from theorem 4.9, proposition 5.1 and propositions 5.4, through 5.11.
∎
The only case not accounted for is when diam(P)=4 and r(P)=3. In this case we know covA′B′(P)=2, so Aucg(C,P)=5 or 6. We also know covA′(P)=2. Therefore Aucg(C,P)=5 if and only if covAA′′B′′(P)=2. However, we have not found this to be the case for any such P. We also have been unable to show it is impossible, so we are left with the following conjecture.
Conjecture 2**.**
If C is a graph with diam(C)>1 and P is a graph with diam(P)=4 and r(P)=3, then Aucg(C,P)=6.
The diam(P)=r(P)=4 case also warrants further discussion. Propositions 5.10 and 5.11 show there are examples of P when Aucg(C,P)=6 and with Aucg(C,P)=5. It will be necessary to find another metric invariant other than diameter and radius to refine the results of this case. At this point we are unsure what a suitable invariant may be.
Finally, independent of C, there is a major difference between possible appendage numbers when diam(P)>2 and when diam(P)=2. When diam(P)>2, theorems 6.2 and 6.3 show there are only finitely many possible appendage numbers, and which are independent of the size of V(P). On the other hand, for diam(P)=r(P)=2 the graph Pα in proposition 5.4 gives
[TABLE]
This may suggest that appendage numbers are related to ∣V(P)∣ when diam(P)=r(P)=2, however the following proposition shows this is not the case.
Proposition 6.4**.**
For every α,β∈N there is a graph P such that V(P)=2α+β and Aucg(Kn,P)=2α.
Proof.
In this proof we modify the construction of Pα from proposition 5.4. For α,β∈N define Pα,β=(Vα,β,Eα,β) as follows. Let
[TABLE]
and
[TABLE]
That is, Pα,β is the graph from proposition 5.4 with β new vertices, gi′s, that are adjacent to every vertex except themselves, e1 and f1. Observe that diam(Pα,β)=r(Pα,β)=2 if α≥2.
We now show covA(Pα,β)=covAB(Pα,β)=2α. Let Pα,β={P1,…,Pk} be a covering of Pα,β satisfying condition A. For 2≤i≤α, {ei},{fi}∈Pα,β as in proposition 5.4. If
[TABLE]
then Pα,β fails to satisfy condition A. Therefore e1, f1 and the gls must be contained in least two elements of Pα,β, and so covA(Pα,β)≥2α.
Let P1α,β={e1} and Q1α,β={f1,g1…,gβ}, and for 2≤i≤α let Piα,β={ei} and Qiα,β={fi}. Then {P1α,β,Q1α,β,…,Pαα,β,Qαα,β} is a covering of Pα,β that satisfies both conditions A and B. So Aucg(Kn,Pα,β)=2α and V(Pα,β)=2α+β.
∎
7 Other Appendage Numbers
In the paper [3] Gu defines Aucg(C) to be the minimum number of vertices needed to be added to C in order to create a uniform central graph G with ⟨Z(G)⟩=C. To match notation we let Aucg(C,−)=Aucg(C). Gu’s main theorem is the following.
Theorem 7.1** (Gu).**
If C is a connected graph, then
[TABLE]
We use our results to give an alternative proof of Gu’s result which is also true without the condition that C is connected.
Proof.
Observe that Aucg(C,−)=min{Aucg(C,P)+∣V(P)∣} where the minimum is taken over all graphs P. By proposition 2.4 r(P)>1 and so we may assume ∣V(P)∣≥2. Since covA(P)≥2 for any P, by proposition 3.2 Aucg(Kn,P)≥2 for n≥2, and by proposition 4.1 Aucg(C,P)≥4 for any non-complete graph C. Hence Aucg(Kn,−)≥4 and Aucg(C,−)≥6.
Let P2 be a graph with two isolated vertices {u,v} and P2={{u},{v}} a covering. Note P2 satisfies conditions A, B, A*′* and B*′*. Let
[TABLE]
and
[TABLE]
The graph G1 is the UCG in the first half of the proof of theorem 3.3 and G2 is a UCG by proposition 4.2. Furthermore, ∣I(G1)∣=2 and ∣I(G1)∣=4, so Aucg(Kn,−)=4 and Aucg(C,−)=6.
Finally, Aucg({v},−)=2 by theorem 6.1.
∎
We can also consider Aucg(−,P), the minimum number of vertices needed to be added to P in order to construct a uniform central graph G with ⟨CP(G)⟩=P. From propositions 2.4 and 6.1, it follows that Aucg(−,P)=∞ if r(P)=1, and Aucg(−,P)=1 otherwise.