A reduction of the spectrum problem for odd sun systems and the prime case
Marco Buratti
Dipartimento di Matematica e Informatica, Università di Perugia, Via Vanvitelli 1, I-06123 Perugia, Italy
[email protected]
,
Anita Pasotti
DICATAM - Sez. Matematica, Università degli Studi di Brescia, Via
Branze 43, I-25123 Brescia, Italy
[email protected]
and
Tommaso Traetta
DICATAM - Sez. Matematica, Università degli Studi di Brescia, Via
Branze 43, I-25123 Brescia, Italy
[email protected]
Abstract.
A k-cycle with a pendant edge attached to each vertex is called a k-sun.
The existence problem for k-sun decompositions of Kv, with k odd, has been solved only when
k=3 or 5.
By adapting a method used by Hoffmann, Lindner and Rodger to reduce the spectrum problem for odd cycle systems of the complete graph,
we show that if there is a k-sun system of Kv (k odd) whenever v lies in the range 2k<v<6k and
satisfies the obvious necessary conditions, then such a system exists for every admissible v≥6k.
Furthermore, we give a complete solution whenever k is an odd prime.
Key words and phrases:
Graph decompositions, Cycle systems, Sun systems, Crown graph, Partial mixed differences.
2010 Mathematics Subject Classification:
05B30, 05C51
1. Introduction
We denote by V(Γ) and E(Γ) the set of vertices and the list of edges of a graph Γ, respectively.
Also, we denote by Γ+w the graph obtained by adding to Γ an independent set
W={∞i∣1≤i≤w} of w≥0
vertices each adjacent to every vertex of Γ, namely,
[TABLE]
where
KV(Γ),W is the complete bipartite graph with parts V(Γ) and W. Denoting by Kv the
complete graph of order v, it is clear that Kv+1 is isomorphic to Kv+1.
We denote by x1∼x2∼…∼xk the path with edges {xi−1,xi}
for 2≤i≤k. By adding the edge {x1,xk} when k≥3, we obtain a cycle of length k
(briefly, a k-cycle)
denoted by (x1,x2,…,xk). A k-cycle with further v−k≥0 isolated vertices will be referred to as
a k-cycle of order v. By adding to (x1,x2,…,xk) an independent set of edges
\big{\{}\{x_{i},x^{\prime}_{i}\}\mid 1\leq{i}\leq{k}\big{\}}, we obtain the k-sun on 2k vertices
(sometimes referred to as k-crown graph) denoted by
[TABLE]
whose edge-set is therefore \big{\{}\{x_{i},x_{i+1}\},\{x_{i},x^{\prime}_{i}\}\mid 1\leq{i}\leq{k}\big{\}},
where xk+1=x1.
A decomposition of a graph K is a set {Γ1,Γ2,…,Γt} of subgraphs of K
whose edge-sets between them partition the edge-set of K; in this case, we briefly write
K=⊕i=1tΓi. If each Γi is isomorphic to Γ, we speak of a
Γ-decomposition of K. If Γ is a k-cycle (resp., k-sun), we also speak of a
k-cycle system (resp., k-sun system) of K.
In this paper we study the existence problem for k-sun systems of Kv (v>1).
Clearly, for such a system to exist we must have
[TABLE]
As far as we know, this problem has been completely settled only
when k=3,5 [8, 10], k=4,6,8 [12], and when k=10,14 or 2t≥4 [9].
It is important to notice that, as a consequence of a general result proved in [14],
condition (∗ ‣ 1) is sufficient whenever v is large enough with respect to k.
These results seem to suggest the following.
Conjecture 1**.**
Let k≥3 and v>1.
There exists a k-sun system of Kv if and only if (∗ ‣ 1) holds.
Our constructions rely on the existence of k-cycle systems of Kv,
a problem that has been completely settled in
[1, 4, 5, 11, 13]. More precisely,
[4] and [11]
reduce the problem to the orders v in the range k≤v<3k, with v odd. These cases are then
solved in [1, 13]. For odd k, an alternative proof based on 1-rotational constructions
is given in [5]. Further results on k-cycle systems of Kv with an automorphism group
acting sharply transitively on all but at most one vertex can be found in [2, 6, 7, 15].
The main results of this paper focus on the case where k is odd.
By adapting a method used in [11] to reduce the spectrum problem for odd cycle systems of the complete graph,
we show that if there is a k-sun system of Kv (k odd) whenever v lies in the range 2k<v<6k and
satisfies the obvious necessary conditions, then such a system exists for every admissible v≥6k.
In other words, we show the following.
Theorem 1.1**.**
Let k≥3 be an odd integer and v>1. Conjecture 1 is true if and only if there exists a k-sun system of Kv for all v satisfying the necessary conditions in (∗ ‣ 1) with 2k<v<6k.
We would like to point out that we strongly believe the reduction methods used in
[4, 11] could be further developed to reduce the spectrum problem of other types of graph decompositions of Kv.
In Section 6, we construct k-sun systems of Kv for every odd prime k whenever 2k<v<6k and (∗ ‣ 1) holds. Therefore, as a consequence of Theorem 1.1, we solve the existence problem for k-sun systems of Kv whenever k is an odd prime.
Theorem 1.2**.**
For every odd prime p there exists a p-sun system of Kv with v>1 if and only if
v≥2p and v(v−1)≡0(mod4p).
Both results rely on the difference methods described in Section 2. These methods are used in
Section 3 to construct specific k-cycle decompositions of some subgraphs of K2k+w,
which we then use in Section 4 to build k-sun systems of K4k+n. This is the last ingredient we need in Section 5 to prove Theorem 1.1.
Difference methods are finally used in Section 6 to construct k-sun systems of Kv for every odd prime k whenever 2k<v<6k and (∗ ‣ 1) holds.
2. Preliminaries
Henceforward, k≥3 is an odd integer, and ℓ=2k−1.
Also, given two integers a≤b, we denote by [a,b] the interval containing the integers {a,a+1,…,b}.
If a>b, then [a,b] is empty.
In our constructions we make extensive use of the method of partial mixed differences
which we now recall but limited to the scope of this paper.
Let G be an abelian group of odd order n in additive notation, let W={∞u∣1≤u≤w}, and denote by Γ a graph with vertices in V=(G×[0,m−1]) ∪ W.
For any permutation f of V, we denote
by f(Γ) the graph obtained by replacing each vertex of Γ, say x, with f(x).
Letting τg, with g∈G, be
the permutation of V fixing each ∞u∈W and mapping (x,i)∈G×[0,m−1] to (x+g,i),
we call τg the translation by g and τg(Γ) the related translate of Γ.
We denote by OrbG(Γ)={τg(Γ)∣g∈G}
the G-orbit of Γ, that is, the set of all distinct translates of Γ,
and by DevG(Γ)=⋃g∈Gτg(Γ) the graph union of all translates of Γ.
Further, by StabG(Γ)={g∈G∣τg(Γ)=Γ} we denote
the G-stabilizer of Γ, namely, the set of translations fixing Γ. We recall that StabG(Γ) is a subgroup of G,
hence s=∣StabG(Γ)∣ is a divisor of n=∣G∣.
Henceforward, when G=Zk, we will simply write Orb(Γ), Dev(Γ), and Stab(Γ).
Suppose now that Γ is either a k-cycle or a k-sun with vertices in V.
For every i,j∈[0,m−1], the list of (i,j)-differences of Γ is the multiset ΔijΓ
defined as follows:
- (1)
if Γ=(x1,x2,…,xk), then
[TABLE]
2. (2)
if
\Gamma=\left(\begin{array}[]{cccc}x_{1}&x_{2}&\ldots&x_{k}\\
x^{\prime}_{1}&x^{\prime}_{2}&\ldots&x^{\prime}_{k}\end{array}\right), then
[TABLE]
We notice that when s=1 we find the classic concept of list of differences.
Usually, one speaks of pure or mixed differences according to whether i=j or not, and when
m=1 we simply write ΔΓ. This concept naturally extends to a family F of graphs with vertices
in V by setting ΔijF=⋃Γ∈FΔijΓ.
Clearly, ΔijΓ=−ΔjiΓ, hence ΔijF=−ΔjiF, for
every i,j∈[0,m−1].
We also need to define the list of neighbours of ∞u in F, that is,
the multiset NF(∞u) of the vertices in V adjacent to ∞u
in some graph Γ∈F.
Finally, we introduce a special class of subgraphs of Kmn. To this purpose,
we take V(Kmn)=G×[0,m−1]. Letting Dii⊆G∖{0}
for every 0≤i≤m−1, and Dij⊆G for every 0≤i<j≤m−1,
we denote by
[TABLE]
the spanning subgraph of Kmn
containing exactly the edges
\big{\{}(g,i),(g+d,j)\big{\}} for every g∈G, d∈Dij,
and 0≤i≤j≤m−1. The reader can easily check that this graph remains unchanged if we replace any set Dii with ±Dii.
The following result, standard in the context of difference families,
provides us with a method to construct
Γ-decompositions for subgraphs of Kmn+w.
Proposition 2.1**.**
Let G be an abelian group of odd order n, let m and w be non-negative integers,
and denote by F a family of k-cycles
(resp., k-suns) with vertices in (G×[0,m−1]) ∪ {∞u∣u∈Zw}
satisfying the following conditions:
- (1)
ΔijF* has no repeated elements, for every 0≤i≤j<m;*
2. (2)
N_{\mathcal{F}}(\infty_{u})=\big{\{}(g_{u,i},i)\mid 0\leq i<m,g_{u,i}\in G\big{\}}*
for every 1≤u≤w.*
Then ⋃Γ∈FOrbG(Γ)={τg(Γ)∣g∈G,Γ∈F} is a k-cycle (resp., k-sun) system of
⟨ΔijF∣0≤i≤j≤m−1⟩+w.
Proof.
Let F∗=⋃Γ∈FOrbG(Γ), K=⟨ΔijF∣0≤i≤j≤m−1⟩, and let ϵ be an edge of K+w.
We are going to show that ϵ belongs to exactly one graph of F∗.
If ϵ∈E(K), by recalling the definition of K we have that ϵ={(g,i),(g+d,j)} for some g∈G and d∈ΔijF, with 0≤i≤j<m.
Hence, there is a graph Γ∈F such that d∈ΔijΓ.
This means that Γ contains the edge ϵ′={(g′,i),(g′+d,j)} for some g′∈G,
therefore ϵ=τg−g′(ϵ′)∈τg−g′(Γ)∈F∗.
To prove that ϵ only belongs to τg−g′(Γ), let Γ′ be any graph in F such that ϵ∈τx(Γ′), for some x∈G.
Since translations preserve differences, we have that
d∈Δijτx(Γ′)=ΔijΓ′.
Considering that d∈ΔijΓ ∩ ΔijΓ′ and,
by condition (1),
ΔijF has no repeated elements, we necessarily have that Γ′=Γ, hence τ−x(ϵ)∈Γ. Again, since ΔijΓ has no repeated elements
(condition (1)),
and considering that ϵ′ and τ−x(ϵ) are edges of Γ that yield the same
differences, then τ−x(ϵ)=ϵ′=τg′−g(ϵ), that is,
τg′−g+x(ϵ)=ϵ. Since G has odd order, it has no element of order 2,
hence g′−g+x=0, that is, x=g−g′, therefore τg−g′(Γ) is the only graph of F∗ containing ϵ.
Similarly, we show that every edge of (K+w)∖K belongs to exactly one graph of F∗.
Let ϵ={∞u,(g,i)} for some u∈Zw and (g,i)∈G×[0,m−1].
By assumption, there is a graph Γ∈F∗ containing the edge
ϵ′={∞u,(gu,i,i)} with gu,i∈G.
Hence, ϵ=τg−gu,i(ϵ′)∈τg−gu,i(Γ).
Finally, if ϵ∈τx(Γ′) for some x∈G and Γ′∈F, then
{∞u,(g−x,i)}=τ−x(ϵ)∈Γ′.
Since condition (2) implies that
NF(∞u) contains exactly one pair from G×{i},
we necessarily have that Γ=Γ′ and x=g−gu,i; therefore,
there is exactly one graph of F∗ containing ϵ.
Condition (2) also implies that NF(∞u) is disjoint from
{∞u∣u∈Zw}, and this guarantees that no graph in F∗
contains edges joining two infinities.
Therefore, F∗ is the desired decomposition of K+w.
∎
Considering that Kmn=⟨Dij∣0≤i≤j≤m−1⟩ if and only if
±Dii=G∖{0} for every i∈[0,m−1], and Dij=G for every 0≤i<j≤m−1,
the proof of the following corollary to Proposition 2.1 is straightforward.
Corollary 2.2**.**
Let G be an abelian group of odd order n, let m and w be non-negative integers,
and denote by F a family of k-cycles
(resp., k-suns) with vertices in (G×[0,m−1]) ∪ {∞u∣u∈Zw}
satisfying the following conditions:
- (1)
\Delta_{ij}\mathcal{F}=\begin{cases}G\setminus\{0\}&\text{if 0\leq i=j\leq m-1};\\
G&\text{if 0\leq i<j\leq m-1};\end{cases}**
2. (2)
N_{\mathcal{F}}(\infty_{u})=\big{\{}(g_{u,i},i)\mid 0\leq i<m,g_{u,i}\in G\big{\}}*
for every 1≤u≤w.*
Then ⋃Γ∈FOrbG(Γ) is a k-cycle (resp., k-sun) system of Kmn+w.
3. Constructing k-cycle systems of ⟨D00,D01,D11⟩+w
In this section, we recall and generalize some results from [11] in order to provide conditions on
D00,D01,D11⊆Zk that guarantee the existence of
a k-cycle system for the subgraph ⟨D00,D01,D11⟩+w of
K2k+w, where V(K2k)=Zk×{0,1}.
We recall that every connected 4-regular Cayley graph over an abelian group has a Hamilton cycle system [3]
and show the following.
Lemma 3.1**.**
Let [a,b],[c,d]⊆[1,ℓ].
The graph \big{\langle}\left[a,b\right],\varnothing,\left[c,d\right]\big{\rangle}
has a k-cycle system whenever both [a,b] and [c,d] satisfy the following condition:
the interval has even size or contains an integer coprime with k.
Proof.
The graph \big{\langle}\left[a,b\right],\varnothing,\left[c,d\right]\big{\rangle} decomposes into \big{\langle}\left[a,b\right],\varnothing,\varnothing\big{\rangle}
and \big{\langle}\varnothing,\varnothing,\left[c,d\right]\big{\rangle}.
The first one is the Cayley graph
Γ=Cay(Zk,[a,b]) with further k isolated vertices, while the second one
is isomorphic to \big{\langle}\left[c,d\right],\varnothing,\varnothing\big{\rangle}.
Therefore, it is enough to show that Γ has a k-cycle system.
Note that Γ decomposes into the subgraphs Cay(Zk,Di), for 0≤i≤t, whenever
the sets Di between them partition [a,b]. By assumption,
[a,b] has even size or contains an integer coprime with k. Therefore, we can assume that
for every i>0 the set Di is a pair of integers at distance 1 or 2, and
D0 is either empty or contains exactly one integer coprime with k.
Clearly, Cay(Zk,D0) is either the empty graph or a k-cycle, and
the remaining Cay(Zk,Di)
are 4-regular Cayley graphs. Also, for every i>0 we have that
Di is a generating set of Zk
(since k is odd and Di contains integers at distance 1 or 2), hence
the graph Cay(Zk,Di) is connected. It follows that each Cay(Zk,Di), with i>0, decomposes into two k-cycles, thus the assertion is proven.
∎
Lemma 3.2**.**
Let S⊆{2i−1∣1≤i≤ℓ}.
Then
there exist k-cycle systems for the graphs
\big{\langle}\{\ell\},S\ \cup\ (S+1),\varnothing\big{\rangle}
and \big{\langle}\{\ell\},(S+1)\ \cup\ (S+2),\varnothing\big{\rangle}.
Proof.
We note that the result is trivial when S=∅, since
\big{\langle}\{\ell\},\varnothing,\varnothing\big{\rangle} is a k-cycle.
The existence of a k-cycle system of
\Gamma=\big{\langle}\{\ell\},S\ \cup\ (S+1),\varnothing\big{\rangle} has been proven in
[11, Lemma 3] when S⊆{2i−1∣1≤i≤ℓ}. Consider now the permutation
f of Zk×{0,1} fixing Zk×{0} pointwise, and mapping (i,1) to
(i+1,1) for every i∈Zk. It is not difficult to check that f(\Gamma)=\big{\langle}\{\ell\},(S+1)\ \cup\ (S+2),\varnothing\big{\rangle}
which is therefore isomorphic to Γ, and hence it has a k-cycle system.
∎
Lemma 3.3**.**
Let r,s and s′ be integers such that
1≤s≤s′≤min{s+1,ℓ}, and 0<r≡s+s′(mod2). Also, let
D⊆[0,k−1] be a non-empty interval of size k−(s+s′+2r). Then
there is a cycle C=(x1,x2,…,xk) of
\Gamma=\big{\langle}[1+\epsilon,s+\epsilon],D,[1+\epsilon,s^{\prime}+\epsilon]\big{\rangle}+r, for every
ϵ∈{0,1},
such that Orb(C) is a k-cycle system of Γ.
Furthermore, if u=0 or u=1−ϵ=1≤s−1, then
- (1)
Dev\big{(}\{x_{2-u},x_{3-u}\}\big{)}* is a k-cycle with vertices in Zk×{0};*
2. (2)
Dev\big{(}\{x_{4+u},x_{5+u}\}\big{)}* is a k-cycle with vertices in Zk×{1}.*
Proof.
Set t=k−(s+s′+2r) and let
\Omega=\big{\langle}[1+\epsilon,s+\epsilon],[0,t-1],[1+\epsilon,s^{\prime}+\epsilon]\big{\rangle}+r.
For i∈[0,s+s′+1] and j∈[0,t+r−1], let ai and bj be the elements of Zk×{0,1} defined
as follows:
[TABLE]
Since the elements ai and bj are pairwise distinct, except for a0=b0 and as+s′+1=bt+r−1,
then the union F of the following two paths is a k-cycle:
[TABLE]
Since ΔijF=ΔijP ∪ ΔijQ, for i,j∈{0,1}, where
[TABLE]
and considering that NF(∞h)=NQ(∞h)={bt+h−2,bt+h−1} for every
h∈[1,r],
Proposition 2.1
guarantees that Orb(F) is a k-cycle system of Ω. Furthermore,
if u=0 or u=1−ϵ=1≤s−1, then
[TABLE]
Since k is odd, we have that
Dev({as−u−1,as−u}) and Dev({as+u+2,as+u+1}) are k-cycles with
vertices in Zk×{0} and Zk×{1}, respectively.
If D=[g,g+t−1] is any interval of [0,k−1] of size t, and f is the permutation of Zk×{0,1}
fixing Zk×{0} pointwise, and mapping (i,1) to (i+g,1) for every i∈Zk, one can check that C=f(F) is the desired k-cycle of Γ=f(Ω).
∎
Lemma 3.4**.**
- (1)
Let ℓ be odd.
If Γ is a 1-factor of K2k, then Γ+ℓ decomposes into k cycles of length k,
each of which contains exactly one edge of Γ.
Furthermore, if \Gamma=\big{\langle}\varnothing,\{d\},\varnothing\big{\rangle}, then there exists a
k-cycle C=(c1,c2,…,ck) of Γ+ℓ, with c1∈Zk×{0} and
c2∈Zk×{1}, such that
[TABLE]
2. (2)
Let ℓ be even.
If Γ is a k-cycle of order 2k,
then Γ+ℓ decomposes into k cycles of length k, each of which contains exactly one edge of Γ.
Furthermore, if \Gamma=\big{\langle}\{d\},\varnothing,\varnothing\big{\rangle} and d is coprime with k,
then there exists a k-cycle C=(c1,c2,…,ck) of Γ+ℓ,
with c1,c2∈Zk×{0}, such that
[TABLE]
Proof.
Permuting the vertices of K2k if necessary, we can assume that
Γ is the 1-factor \Gamma_{0}=\big{\langle}\varnothing,\{0\},\varnothing\big{\rangle} when ℓ is odd,
and the k-cycle \Gamma_{1}=\big{\langle}\{\ell\},\varnothing,\varnothing\big{\rangle} (of order 2k) when ℓ is even.
For h∈{0,1},
let Ch=(ch,1,ch,2,∞1,c3,∞2,c4,…,∞ℓ−1,cℓ+1,∞ℓ) be
the k-cycle of Γh+ℓ, where
[TABLE]
Note that the sets ΔijCh are empty, except for Δ01C0={0}
and Δ00C1={±ℓ}. Also, the two neighbours of
∞u in Ch belong to Zk×{0} and Zk×{1}, respectively.
Hence, Proposition 2.1 guarantees that
Orb(Ch) is a k-cycle system of Γh+ℓ, for h∈{0,1}. We finally notice that
Dev({ch,1,ch,2})=Γh (up to isolated vertices)
and this completes the proof.
∎
The following result has been proven in [11].
Lemma 3.5**.**
Let D⊆[1,ℓ].
The subgraph ⟨D,{0},D⟩ of K2k has a 1-factorization.
Remark 3.6**.**
Considering the permutation f of Zk×{0,1} such that f(i,j)=(i,1−j), and
a graph \Gamma=\big{\langle}D_{0},D_{1},D_{2}\big{\rangle}, we have that f(\Gamma)=\big{\langle}D_{2},-D_{1},D_{0}\big{\rangle}. Therefore,
Lemmas 3.1 – 3.5 continue to hold when we replace Γ by f(Γ).
4. k-sun systems of K4k+n
In this section we provide sufficient conditions for a k-sun system of
K4k+n to exist, when n≡0,1(mod4). More precisely, we show the following.
Theorem 4.1**.**
Let k≥7 be an odd integer and let n≡0,1(mod4) with 2k<n<10k, then there exists a
k-sun system of K4k+n, except possibly when
k=7* and n=20,21,32,33,44,45,56,57,64,65,68,69,*
k=11* and n=100,101,112,113.*
To prove Theorem 4.1, we start by introducing some notions and prove some preliminary results.
Let M be a positive integer and
take V(K2iM)=ZM×[0,2i−1] and
V(K_{2^{i}M}+w)=V(K_{2^{i}M})\ \cup\ \big{\{}\infty_{h}\mid h\in\mathbb{Z}_{w}\big{\}},
for i∈{1,2} and w>0.
Now assume that w=2u, and let x↦x be the permutation of V(K4M+2u) defined as follows:
[TABLE]
For any subgraph Γ of K4M+2u,
we denote by Γ the graph (isomorphic to Γ)
obtained by replacing each vertex x of Γ with x.
Given a subgraph Γ of K2M+u, we denote by Γ[2] the spanning subgraph of K4M+2u
whose edge set is
[TABLE]
and let Γ∗[2]=Γ[2]⊕I be the graph obtained by adding to Γ[2]
the 1-factor
[TABLE]
Note that, up to isolated vertices, Γ[2] is the lexicographic product of
Γ with the empty graph on two vertices.
The proof of the following elementary lemma is left to the reader.
Lemma 4.2**.**
Let Γ=⊕i=1nΓi and let w=∑i=1nwi with wi≥0.
If Γ and the Γis have the same vertex set (possibly with isolated vertices), then
- (1)
Γ+w=⊕i=1n(Γi+wi);
2. (2)
Γ[2]=⊕i=1nΓi[2];
3. (3)
(Γ+w)[2]=Γ[2]+2w.
We start showing that if C is a k-cycle, then C[2] decomposes into two k-suns.
Lemma 4.3**.**
Let C=(c1,c2,…,ck) be a cycle with vertices in
\big{(}\mathbb{Z}_{M}\times\{0,1\}\big{)}\ \cup\ \{\infty_{h}\mid h\in\mathbb{Z}_{u}\} and let S be the k-sun defined as follows:
[TABLE]
where si∈{ci,ci} for every i∈[1,k].
Then C[2]=S⊕S.
Proof.
It is enough to notice that S contains the edges
{si,si+1} and {si,si+1},
while S contains {si,si+1} and {si,si+1}, for every i∈[1,k], where sk+1=s1 and sk+1=s1.
∎
Example 4.4**.**
In Figure 1 we have the graph C7[2] which can be decomposed into two 7-suns S and S.
The non-dashed edges are those of S, while the dashed edges are those of S.
For every cycle C=(c1,c2,…,ck) with vertices in ZM×{0,1}, we set
[TABLE]
Clearly, C[2]=σ(C)⊕σ(C) by Lemma 4.3.
Lemma 4.5**.**
If C={C1,C2,…,Ct} is a k-cycle system of Γ+u, where Γ is a subgraph of
K2M, and
Si is a k-sun obtained from Ci as in Lemma 4.3, then
{\mathcal{S}}=\big{\{}S_{i},\overline{S_{i}}\mid i\in[1,t]\big{\}} is a k-sun system of Γ[2]+2u. In particular,
if \mathcal{C}=Orb\big{(}C_{1}\big{)}, then
Orb(S_{1})\ \cup\ Orb\big{(}\overline{S_{1}}\big{)}
is a k-sun system of Γ[2]+2u.
Proof.
By assumption Γ+u=⊕i=1tCi, where each Ci is a k-cycle. Also, by Lemma 4.2,
we have that Γ[2]+2u=(Γ+u)[2]=⊕i=1tCi[2]. Since Ci[2]=Si⊕Si by Lemma
4.3, then S is a k-sun system of Γ[2]+2u.
The second part easily follows by noticing that
if Ci=τg(C1) for some g∈ZM, then
C_{i}[2]=\tau_{g}(C_{1}[2])=\tau_{g}\big{(}S_{1}\big{)}\oplus\tau_{g}\big{(}\overline{S_{1}}\big{)}.
∎
The following lemma describes the general method we use to construct k-sun systems of K4k+n.
We point out that throughout the rest of this section we take
V(K2k)=Zk×{0,1} and V(K4k)=Zk×[0,3].
Lemma 4.6**.**
Let K2k=Γ1⊕Γ2 with V(Γ1)=V(Γ2)=V(K2k).
If Γ1+w1 has a k-cycle system and Γ2∗[2]+w2 has a k-sun system,
then K4k+(2w1+w2) has a k-sun system.
Proof.
The result follows by Lemma 4.2. In fact, noting that
K4k=K2k[2]⊕I, where I=\big{\{}\{z,\overline{z}\}\mid z\in\mathbb{Z}_{k}\times\{0,1\}\big{\}}, we have that
[TABLE]
The result then follows by Lemma 4.5.
∎
We are now ready to prove the main result of this section, Theorem 4.1.
The case k≡1(mod4) is proven in Theorem 4.7, while the
case k≡3(mod4) is dealt with in Theorems
4.9, 4.10, 4.11 and 4.12.
Theorem 4.7**.**
If k≡1(mod4)≥9 and n≡0,1(mod4) with 2k<n<10k, then there exists a
k-sun system of K4k+n.
Proof.
Let n=2(qℓ+r)+ν with 1≤r≤ℓ and ν∈{2,3}.
Note that ℓ≥4 is even and r is odd, since n≡0,1(mod4)≥9 and k≡1(mod4).
Considering also that 2k<n<10k, we have that 2≤q≤10≤k+2r−1. Furthermore,
let V(K_{4k}+n)=\big{(}\mathbb{Z}_{k}\times[0,3]\big{)}\cup\{\infty_{h}\mid h\in\mathbb{Z}_{n-\nu}\}\cup\{\infty^{\prime}_{1},\infty^{\prime}_{2},\infty^{\prime}_{\nu}\}.
We start decomposing K2k into the following two graphs:
[TABLE]
We notice that Γ1 further decomposes into the following graphs:
[TABLE]
each of which decomposes into k-cycles by Lemmas 3.1 and 3.2; hence Γ1 has a k-cycle system
{C1,C2,…,Cγ}, where γ=k+2r−2.
Note that this system is non-empty, since 1≤q−1≤γ.
Without loss of generality, we can assume that each cycle Ci has order 2k and
[TABLE]
Now set Ω1=Γ1∖C1 and Ω2=Γ2⊕C1.
Letting w1=(q−2)ℓ=∑j=2γw1,j, where
w1,j=ℓ when j<q, and w1,j=0 otherwise,
by Lemma 4.2 we have that Ω1+w1=⊕i=2γ(Ci+w1,i).
Therefore, Ω1+w1 has a k-cycle system, since each Ci+w1,i decomposes into k-cycles
by Lemma 3.4.
Setting w2=n−2w1=2(2ℓ+r)+ν and considering that K2k=Γ1⊕Γ2=Ω1⊕Ω2,
by Lemma 4.6 it is left to show that
Ω2∗[2]+w2 has a k-sun system.
Set Γ3=C1, and recall that Ω2∗[2]=Ω2[2]⊕I=Γ2[2]⊕Γ3[2]⊕I,
where I denotes the 1-factor \big{\{}\{z,\overline{z}\}\mid z\in\mathbb{Z}_{k}\times\{0,1\}\big{\}} of K4k.
Hence,
[TABLE]
by Lemma 4.2.
Clearly, Γ2=Γ2,1⊕Γ2,2 where
\Gamma_{2,1}=\big{\langle}\{1\},[0,k-2r-3],\{1\}\big{\rangle} and
\Gamma_{2,2}=\big{\langle}\varnothing,\varnothing,\{\ell\}\big{\rangle},
hence Γ2+(ℓ+r)=(Γ2,1+r)⊕(Γ2,2+ℓ).
By Lemmas 3.3 and 3.4, there exist a k-cycle A=(x1,x2,y3,y4,a5,…,ak) of Γ2,1+r and a k-cycle B=(y1,y2,b3,…,bk) of
Γ2,2+ℓ satisfying the following properties:
[TABLE]
Furthermore, denoted by (c1,c2,…,ck) the cycle in Γ3, Lemma 3.4 guarantees that
[TABLE]
Let S={S1,S2,S3,S4}
and S′={S3+2j,S4+2j∣j∈[1,k]}, where
[TABLE]
By Lemma 4.5 we have that
⋃S∈SOrb(S) is a k-sun system of
\big{(}\Gamma_{2}+(\ell+r)\big{)}[2], and
S′ is a k-sun system of
\big{(}\Gamma_{3}+\ell\big{)}[2].
It follows by (3) that ⋃S∈SOrb(S) ∪ S′ decomposes (Ω2∗[2]+w2)∖(I+ν).
To construct a k-sun system of Ω2∗[2]+w2, we first modify the k-suns in S ∪ S′
by replacing some of their vertices with ∞1′,∞2′, and possibly ∞3′ when ν=3.
More precisely, following Table 1, we obtain Ti from Si by replacing the ordered set Vi of vertices
of Si with Vi′. This yields a set Mi of ‘missing’ edges no longer covered by Ti after this substitution,
but replaced by those in Ni, namely
[TABLE]
We point out that T3+2j=S3+2j,
and T4+2j=S4+2j when ν=2, for every j∈[1,k]. The remaining graphs Ti are explicitly
given below, where the elements in bold are the replaced vertices.
[TABLE]
We notice that \displaystyle\bigcup_{i=1}^{4}Dev(N_{i})\cup\bigcup_{i=5}^{2k+4}N_{i}=\big{\{}\{\infty^{\prime}_{j},x\}\mid j\in[1,\nu],x\in\mathbb{Z}_{k}\times[0,3]\big{\}}.
We finally build the following 2ν+1 graphs:
[TABLE]
By recalling (2) and
(4)–(6), it is not difficult to check that
G1,G2,…,G2ν+1 are k-suns. Furthermore,
[TABLE]
where, we recall, I denotes the 1-factor \big{\{}\{z,\overline{z}\}\mid z\in\mathbb{Z}_{k}\times\{0,1\}\big{\}} of K4k.
Therefore, ⋃i=14Orb(Ti) ∪ {T5,T6,…,T2k+4} ∪ {G1,G2,…,G2ν+1}
is a k-sun system of Ω2∗[2]+w2, and this concludes the proof.
∎
Example 4.8**.**
By following the proof of Theorem 4.7, we construct a
k-sun system of K4k+n when (k,n)=(9,21); hence
(ℓ,q,r,ν)=(4,2,1,3).
The graphs \Gamma_{1}=\big{\langle}[2,4],[5,8],[2,3]\big{\rangle} and
\Gamma_{2}=\big{\langle}\{1\},[0,4],\{1,4\}\big{\rangle} decompose the complete graph K18
with vertex-set Z9×{0,1}. Also Γ1 decomposes into the following
9-cycles of order 18, where i=0,1:
[TABLE]
Clearly, K18=Ω1⊕Ω2, where Ω1=Γ1∖C1 and Ω2=Γ2⊕C1.
Let V(K36)=Z9×[0,3], and
denote by I the 1-factor of K36 containing all edges of the form {(a,i),(a,i+2)},
with a∈Z9 and i∈{0,1}. Then,
[TABLE]
Considering that (Ω2+9)[2]=Ω2[2]+18, we have
[TABLE]
Since the set {σ(Ci),σ(Ci)∣i∈[2,9]} is a 9-sun system of
Ω1[2], it is left to build a 9-sun system of Ω2∗[2]+21=(Ω2[2]+18)⊕(I+3).
We start by decomposing Ω2+9 into 9-cycles. Since
Ω2=Γ2,1⊕Γ2,2⊕Γ3 with
\Gamma_{2,1}=\big{\langle}\{1\},[0,4],\{1\}\big{\rangle},
\Gamma_{2,2}=\big{\langle}\varnothing,\varnothing,\{4\}\big{\rangle}
and Γ3=C1, then
[TABLE]
Let A=(x1,x2,y3,y4,a5,…,a9) and B=(y1,y2,b3,…,b9)
be the 9-cycles defined as follows:
[TABLE]
One can easily check that Orb(A) (resp., Orb(B)) decomposes Γ2,1+1
(resp., Γ2,2+4).
Also, for every edge {cj,cj+1} of C1,
with j∈[1,9] and c10=c1, we construct the cycle Fj=(cj,cj+1,fj,3,fj,4,…,fj,9), where
[TABLE]
One can check that {F1,F2,…,F9} is a 9-cycle system of Γ3+4.
Therefore, U1=Orb(A) ∪ Orb(B) ∪ {F1,F2,…,F9} provides a
9-cycle system of Ω2+9.
Since the set {C[2]∣C∈U1} decomposes (Ω2+9)[2],
and each C[2] decomposes into two 9-suns, we can easily obtain a
9-sun system of (Ω2+9)[2]. Indeed, letting
[TABLE]
we have that A[2]=S1⊕S2, B[2]=S3⊕S4, and
Fj[2]=S3+2j⊕S4+2j, for every j∈[1,9]. Therefore
U2=⋃i=14Orb(Si)∪{S5,S6,…,S22}
is a 9-sun system of Ω2[2]+18.
We finally use U2 to build a 9-sun system of Ω2∗[2]+21=(Ω2[2]+18)⊕(I+3). By replacing the vertices of each Si, as outlined in
Table 1, we obtain the 9-sun Ti.
The new 22 graphs, T1,T2,…,T22, are built in such a way that
[TABLE]
This way we obtain a 9-sun system of Ω2∗[2]+21, and hence
the desired 9-sun system of K36+21.
Theorem 4.9**.**
Let k≡3(mod4)≥7 and n≡0,1(mod4) with 2k<n<10k.
If n≡2,3(modk−1) and ⌊k−1n−4⌋ is even, then there exists a k-sun system of K4k+n
except possibly when (k,n)∈{(7,64),(7,65)}.
Proof.
First, k≡3(mod4)≥7 implies that ℓ≥3 is odd.
Now, let n=2(qℓ+r)+ν with 1≤r≤ℓ and ν∈{2,3}.
Note that q=⌊k−1n−4⌋, hence q is even.
Also, since 2k<n<10k, we have 2≤q≤10.
By q even and n≡0,1(mod4) it follows that r is odd,
and n≡2,3(modk−1) implies that r=ℓ.
To sum up,
[TABLE]
As in the previous theorem,
let V(K_{4k}+n)=\big{(}\mathbb{Z}_{k}\times[0,3]\big{)}\cup\{\infty_{h}\mid h\in\mathbb{Z}_{n-\nu}\}\cup\{\infty^{\prime}_{1},\infty^{\prime}_{2},\infty^{\prime}_{\nu}\}.
We split the proof into two cases.
Case 1) q≤2r+4.
We start decomposing K2k into the following two graphs:
[TABLE]
Since q≤2r+4, the graph Γ1 can be further decomposed into the following graphs:
[TABLE]
[TABLE]
The first two graphs have a k-cycle system
by Lemmas 3.2 and 3.1, while Γ1,3 decomposes into (q−1) 1-factors, say J1,J2,…,Jq−1.
Setting w1=(q−1)ℓ, by Lemma 4.2 we have that:
[TABLE]
Hence Γ1+(q−1)ℓ has a k-cycle system
since each Ji+ℓ decomposes into k-cycles by Lemma 3.4.
Letting w2=n−2w1=2(ℓ+r)+ν and recalling that K2k=Γ1⊕Γ2, by Lemma 4.6
it remains to construct a k-sun system of Γ2∗[2]+w2.
We start decomposing Γ2 into the following graphs:
[TABLE]
Recalling that Γ2∗[2]=Γ2[2]⊕I,
where I denotes the 1-factor \big{\{}\{z,\overline{z}\}\mid z\in\mathbb{Z}_{k}\times\{0,1\}\big{\}} of K4k,
by Lemma 4.2 we have that
[TABLE]
By Lemmas 3.3 and 3.4 there exist a k-cycle A=(x1,x2,x3,y4,y5,y6,a7,…,ak) of Γ2,0+r
and a k-cycle B=(y,x,b3,…,bk) of
Γ2,1+ℓ, satisfying the following properties:
[TABLE]
Set A′=(x1,x2,x3,y4,y5,y6,a7,…,ak)
and B′=(y,x,b3,…,bk)
and let S={σ(A′),σ(A′),σ(B′),σ(B′)}.
By Lemma 4.5, we have that
⋃S∈SOrb(S) is a k-sun system of
\big{(}\Gamma_{2}+(\ell+r)\big{)}[2]=\Gamma_{2}[2]+2(\ell+r)=(\Gamma_{2}^{*}[2]+w_{2})\setminus(I+\nu).
To construct a k-sun system of Γ2∗[2]+w2 we proceed as in Theorem 4.7.
We modify the graphs in S and obtain four k-suns T1,T2,T3,T4
whose translates between them cover all edges incident with
∞1′,∞2′, and possibly ∞3′ when ν=3.
Then we construct further 2ν+1 k-suns G1,…,G2ν+1 to cover the missing edges.
The reader can check that ⋃i=14Orb(Ti)∪{G1,…,G2ν+1} is a
k-sun system of Γ2∗[2]+w2.
The graphs Ti are the following, where the elements in bold are the replaced vertices:
[TABLE]
The graphs Gi, for i=[1,2ν+1], are so defined:
[TABLE]
Case 2) q≥2r+6. Note that this implies r=1 and q=8,10. As before K2k=Γ1⊕Γ2 where
[TABLE]
Since (k,n)=(7,64),(7,65) then (ℓ,q)=(3,10), hence
the graph Γ1 can be decomposed into the following graphs:
[TABLE]
[TABLE]
The graph Γ1,1 decomposes into five 1-factors J1,…,J5, while by Lemma 3.5 Γ1,2 decomposes
into (q−5) 1-factors J1′,…,Jq−5′.
Letting w1=qℓ, by Lemma 4.2 we have that
[TABLE]
By Lemmas 3.4 and 3.1, each Ji+ℓ, each Ji′+ℓ and Γ1,3 decompose into k-cycles.
Hence Γ1+qℓ has a k-cycle system.
Let now w2=n−2w1=2+ν. Note that a k-sun system of Γ2∗[2]+w2 can be obtained as in Case 1, where Γ2,1 is empty.
∎
Theorem 4.10**.**
Let k≡3(mod4)≥11 and n≡0,1(mod4) with 2k<n<10k.
If ⌊k−1n−4⌋ is even, and n≡2,3(modk−1), then there is a
k-sun system of K4k+n, except possibly when (k,n)∈{(11,112), (11,113)}.
Proof.
Let n=2(qℓ+r)+ν with 1≤r≤ℓ and ν∈{2,3}. Clearly,
q=⌊k−1n−4⌋, hence q is even. Since k≥11, 2k<n<10k and n≡2,3(mod2ℓ),
we have that
[TABLE]
As before,
let V(K_{4k}+n)=\big{(}\mathbb{Z}_{k}\times[0,3]\big{)}\cup\{\infty_{h}\mid h\in\mathbb{Z}_{n-\nu}\}\cup\{\infty^{\prime}_{1},\infty^{\prime}_{2},\infty^{\prime}_{\nu}\}.
We start decomposing K2k into the following two graphs:
[TABLE]
If q=2,4, Γ1 can be further decomposed into
[TABLE]
[TABLE]
The graph Γ1,1 decomposes into q 1-factors, say J1,…,Jq.
Letting w1=qℓ, by Lemma 4.2 we have that
[TABLE]
Lemmas 3.4, 3.2 and 3.1 guarantee that each Ji+ℓ, Γ1,2 and Γ1,3
decompose into k-cycles, hence Γ1+w1 has a k-cycle system.
Suppose now q≥6. By (k,n)∈{(11,112),(11,113)},
we have (ℓ,q)=(5,10).
In this case Γ1 can be further decomposed into
[TABLE]
[TABLE]
The graph Γ1,1 can be decomposed into three 1-factors say J1,J2,J3, also by Lemma 3.5
the graph Γ1,2 can be decomposed into (q−3) 1-factors say J1′,…,Jq−3′.
Set again w1=qℓ, by Lemma 4.2 we have that
[TABLE]
By Lemmas 3.4 and 3.1 we have that each Ji+ℓ, each Jj′+ℓ and Γ1,3 decompose into k-cycles,
hence Γ1+w1 has a k-cycle system.
Hence for any value of q we have proved that Γ1+w1 has a k-cycle system.
Now, setting w2=n−2w1=2ℓ+ν and recalling that K2k=Γ1⊕Γ2,
by Lemma 4.6 it is left to show that
Γ2∗[2]+w2 has a k-sun system.
Let r1 and r2≥2 be an odd and an even integer, respectively, such that r1+r2=r=ℓ.
Note that Γ2 can be further decomposed into
[TABLE]
Recalling that Γ2∗[2]=Γ2[2]⊕I,
where I denotes the 1-factor \big{\{}\{z,\overline{z}\}\mid z\in\mathbb{Z}_{k}\times\{0,1\}\big{\}} of K4k,
by Lemma 4.2 we have that
[TABLE]
By Lemma 3.3 there is a k-cycle A=(y1,y2,x3,x4,a5,…,ak) of Γ2,1+r1
and a k-cycle B=(x1,x2,y3,y4,b5,…,bk) of Γ2,2+r2
such that
[TABLE]
Set A′=(y1,y2,x3,x4,a5,…,ak)
and B′=(x1,x2,y3,y4,b5,…,bk).
Let S={σ(A′),σ(A′),σ(B′),σ(B′)}, by Lemma 4.5,
we have that ⋃S∈SOrb(S) is a k-sun system of
\big{(}\Gamma_{2}+\ell\big{)}[2]=\Gamma_{2}[2]+2\ell=(\Gamma_{2}^{*}[2]+w_{2})\setminus(I+\nu).
To construct a k-sun system of Γ2∗[2]+w2, we build a family T={T1,T2,T3,T4} of k-suns
by modifying
the graphs in S so that ⋃T∈TOrb(T) covers all the edges incident with
∞1′,∞2′, and possibly ∞3′ when ν=3.
We then construct further
(2ν+1) k-suns G1,G2,…,G2ν+1 which cover the remaining edges exactly once. Hence,
⋃T∈TOrb(T)∪{G1,G2,…,G2ν+1} is a
k-sun system of Γ2∗[2]+w2.
The graphs T1,…,T4 and G1,…,G2ν+1
are the following, where as before the elements in bold are the replaced vertices.
[TABLE]
[TABLE]
By recalling (LABEL:thm:holek=3,a2), it is not difficult to check that
the graphs Gh are k-suns.
∎
Theorem 4.11**.**
*Let k≡3(mod4)≥7 and n≡0,1(mod4) with 2k<n<10k.
If ⌊k−1n−4⌋ is odd and n≡0,1(modk−1), then there is a
k-sun system of K4k+n.
*
Proof.
Let n=2(qℓ+r)+ν with 1≤r≤ℓ and ν∈{2,3}. Clearly,
q=⌊k−1n−4⌋.
Also, we have that q and ℓ≥3 are odd,
and n≡0,1(mod4); hence r is even.
Furthermore, we have that 2≤q≤10, since by assumption
2k<n<10k. Considering now the hypothesis that
n≡0,1(mod2ℓ), it follows that r=ℓ−1. To sum up,
[TABLE]
As before,
let V(K_{4k}+n)=\big{(}\mathbb{Z}_{k}\times[0,3]\big{)}\cup\{\infty_{h}\mid h\in\mathbb{Z}_{n-\nu}\}\cup\{\infty^{\prime}_{1},\infty^{\prime}_{2},\infty^{\prime}_{\nu}\}.
We start decomposing K2k into the following two graphs:
[TABLE]
Considering that 3≤q≤9≤2r+5, the graph Γ1 can be further decomposed into the following graphs:
[TABLE]
[TABLE]
The first two have a k-cycle system by Lemmas 3.1 and 3.2, while
Γ1,3 decomposes into (q−3) 1-factors, say J1,J2,…,Jq−3.
Letting w1=(q−3)ℓ,
by Lemma 4.2 we have that
[TABLE]
Therefore, Γ1+w1 has a k-cycle system, since each Ji+ℓ decomposes into k-cycles
by Lemma 3.4. Setting w2=n−2w1=2(3ℓ+r)+ν and recalling that K2k=Γ1⊕Γ2,
by Lemma 4.6 it is left to show that
Γ2∗[2]+w2 has a k-sun system.
We start decomposing Γ2 into the following graphs:
[TABLE]
Recalling that Γ2∗[2]=Γ2[2]⊕I,
where I denotes the 1-factor \big{\{}\{z,\overline{z}\}\mid z\in\mathbb{Z}_{k}\times\{0,1\}\big{\}} of K4k,
by Lemma 4.2 we have that
[TABLE]
By Lemmas 3.3 and 3.4 there exist a k-cycle A=(x1,x2,x3,y4,y5,y6,a7,…,ak) of Γ2,0+r,
a k-cycle B1=(x1,0,y1,1,b1,2,…,b1,k−1) of Γ2,1+ℓ,
and a k-cycle Bi=(yi,0,xi,1,bi,2,…,bi,k−1) of
Γ2,i+ℓ, for 2≤i≤3, satisfying the following properties:
[TABLE]
Set A′=(x1,x2,x3,y4,y5,y6,a7,a8,…,ak−1,ak) and
let S={σ(A′),σ(A′)} ∪ {σ(Bi),σ(Bi)∣1≤i≤3}.
By Lemma 4.5, we have that
⋃S∈SOrb(S) is a k-sun system of
\big{(}\Gamma_{2}+(3\ell+r)\big{)}[2]=\Gamma_{2}[2]+2(3\ell+r)=(\Gamma_{2}^{*}[2]+w_{2})\setminus(I+\nu).
To construct a k-sun system of Γ2∗[2]+w2, we build a family T={T0,T1,…,T7} of k-suns
by modifying
the graphs in S so that ⋃T∈TOrb(T) covers all the edges incident with
∞1′,∞2′, and possibly ∞3′ when ν=3.
We then construct further
(2ν+1) k-suns G1,G2,…,G2ν+1 which cover the remaining edges exactly once. Hence,
⋃T∈TOrb(T)∪{G1,G2,…,G2ν+1} is a
k-sun system of Γ2∗[2]+w2.
The graphs T0,…,T7 and G1,…,G2ν+1
are the following, where as before the elements in bold are the replaced vertices.
[TABLE]
[TABLE]
By recalling (LABEL:thm:holek=3,b1:cond2)–(11), it is not difficult to check that the graphs Gh are k-suns.
∎
Theorem 4.12**.**
Let k≡3(mod4)≥7 and n≡0,1(mod4) with 2k<n<10k.
If ⌊k−1n−4⌋ is odd, and n≡0,1(modk−1), then there is a
k-sun system of K4k+n except possibly when (k,n)∈{(11,100),(11,101)}.
Proof.
Let n=2(qℓ+r)+ν with 1≤r≤ℓ and ν∈{2,3}.
Reasoning as in the proof of Theorem 4.11 and
considering that n≡0,1(mod2ℓ) and (k,n)∈{(11,100),(11,101)},
we have that
[TABLE]
As before,
let V(K_{4k}+n)=\big{(}\mathbb{Z}_{k}\times[0,3]\big{)}\cup\{\infty_{h}\mid h\in\mathbb{Z}_{n-\nu}\}\cup\{\infty^{\prime}_{1},\infty^{\prime}_{2},\infty^{\prime}_{\nu}\}.
We start decomposing K2k into the following two graphs
[TABLE]
Considering (12),
we can further decompose Γ1 into the following two graphs:
[TABLE]
By Lemma 3.5, the graph Γ1,1 decomposes into q 1-factors,
say J1,J2,…,Jq. Letting w1=qℓ, by Lemma 4.2 we have that
[TABLE]
Lemmas 3.4 and 3.1 guarantee that each Ji+ℓ and Γ1,2 decompose into k-cycles,
hence Γ1+w1 has a k-cycle system.
Let r1 and r2 be odd positive integers such that r=ℓ−1=r1+r2.
Then, setting w2=n−2w1=2(r1+r2)+ν and recalling that K2k=Γ1⊕Γ2,
by Lemma 4.6 it is left to show that
Γ2∗[2]+w2 has a k-sun system.
We start decomposing Γ2 into the following graphs:
[TABLE]
Recalling that Γ2∗[2]=Γ2[2]⊕I,
where I denotes the 1-factor \big{\{}\{z,\overline{z}\}\mid z\in\mathbb{Z}_{k}\times\{0,1\}\big{\}} of K4k,
by Lemma 4.2 we have that
[TABLE]
By Lemma 3.3 there is a k-cycle A=(y1,y2,x3,x4,a5,…,ak) of Γ2,1+r1 and
a k-cycle
B=(x1,x2,y3,y4,b5,…,bk) of Γ2,2+r2 such that
[TABLE]
Set A′=(y1,y2,x3,x4,a5,…,ak),
B′=(x1,x2,y3,y4,b5,…,bk) and
let
S={σ(A′),σ(A′),σ(B′),σ(B′)}.
By Lemma 4.5, we have that
⋃S∈SOrb(S) is a k-sun system of
(Γ2∗[2]+w2)∖(I+ν).
To construct a k-sun system of Γ2∗[2]+w2, we build a family T={T1,T2,T3,T4} of
four k-suns, each of which is obtained from a graph in S by replacing some of their vertices
with ∞1′,∞2′, and possibly ∞3′ when ν=3.
Then we construct further
(2ν+1) k-suns G1,G2,…,G2ν+1 so that
⋃T∈TOrb(T) ∪ {G1,G2,…,G2ν+1} is a
k-sun system of Γ2∗[2]+w2.
[TABLE]
[TABLE]
By (13), it is not difficult to check that
the graphs Gh are k-suns.
∎
5. It is sufficient to solve 2k<v<6k
In this section we show that if the necessary conditions in (∗ ‣ 1), for the existence of
a k-sun system of Kv, are sufficient for all v satisfying 2k<v<6k, then they are sufficient for all v. In other words, we prove Theorem 1.1.
We start by showing how to construct k-sun systems of Kg×h
(i.e., the complete multipartite graph with g parts each of size h) when h=4k.
Theorem 5.1**.**
For any odd integer k≥3 and any integer g≥3,
there exists a k-sun system of Kg×4k.
Proof.
Set V(Kg×2k)=Zgk×[0,1] and let
Kg×4k=Kg×2k[2].
In [11, Theorem 2] the authors proved the existence of a k-cycle system of Kg×2k.
By applying Lemma 4.5 (with Γ=Kg×2k and u=0)
we obtain the existence of a k-sun system of Kg×4k.
∎
The following result exploits Theorem 5.1 and shows how to construct
k-sun systems of K4kg+n, for g=2, starting from a k-sun system of K4k+n and a k-sun system of
either Kn or K4k+n.
Theorem 5.2**.**
Let k≥3 be an odd integer and assume that both the following conditions hold:
- (1)
there exists a k-sun system of either Kn or K4k+n;
2. (2)
there exists a k-sun system of K4k+n.
Then there is a k-sun system of K4kg+n for all positive g=2.
Proof.
Suppose there exists a k-sun system S1 of Kn, also, by (2), there exists a k-sun system S2 of K4k+n. Clearly, S1 ∪ S2 is a k-sun system of
Kn+4k=Kn⊕(K4k+n). Hence we can suppose g≥3.
Let V, H and G be sets of size n, 4k and g, respectively, such that V∩(H×G)=∅.
Let S be a k-sun system of Kn (resp., Kn+4k)
with vertex set V (resp., V∪(H×{x0}) for some x0∈G).
By assumption, for each x∈G, there is a k-sun system, say Bx, of K4k+n
with vertex set V∪(H×{x}), where V(K4k)=H×{x}.
Also, by Theorem 5.1 there is a k-sun system C of Kg×4k whose parts are
H×{x} with x∈G.
Hence the k-suns of Bx with x∈G (resp., x∈G∖{x0}), S and
C form a k-sun system of Kn+4kg with vertex set V∪(H×G).
∎
We are now ready to prove Theorem 1.1 whose statement is recalled below.
Theorem 1.1.
*
Let k≥3 be an odd integer and v>1. Conjecture 1 is true if and only if there exists a k-sun system of Kv for all v satisfying the necessary conditions in (∗ ‣ 1) with 2k<v<6k.*
Proof.
The existence of 3-sun systems and 5-sun systems has been solved in [10] and in [8], respectively. Hence we can suppose k≥7 and 2k<v<6k.
We first deal with the case where (k,v)=(7,21).
By assumption there exists a k-sun system of Kv,
which implies v(v−1)≡0(mod4), hence
Theorem 4.1 guarantees the existence of
a k-sun system of K4k+v.
Therefore, by Theorem 5.2 there is
a k-sun decomposition of K4kg+v whenever g=2.
To decompose K8k+v into k-suns, we first decompose
K8k+v into K4k+v and K4k+(4k+v).
By Theorem 5.2 (with g=1), there is a k-sun system of K4k+v.
Furthermore, Theorem 4.1 guarantees the existence of
a k-sun system of K4k+(4k+v), except possibly when
(k,4k+v)∈{(7,56),(7,57),(7,64),(11,100)}.
Therefore, by Theorem 5.2,
there is a k-sun decomposition of K8k+v
whenever (k,4k+v)∈{(7,56),(7,57),(7,64),(11,100)}.
For each of these four cases we construct k-sun systems of K8k+v as follows.
If k=7 and 4k+v=56, set V(K84)=Z83∪{∞}. We consider the following 7-suns
[TABLE]
One can easily check that ⋃i=13OrbZ83(Ti) is a 7-sun system of K84.
If k=7 and 4k+v=57, set V(K85)=Z85. Let T1 and T2 be defined as above, and let T3′ be the graph obtained from T3
replacing ∞ with 60. It is immediate that ⋃i=12OrbZ85(Ti)∪OrbZ85(T3′) is a 7-sun system of K85.
If k=7 and 4k+v=64, set V(K92)=(Z7×Z13)∪{∞}. We consider the following 7-suns
[TABLE]
[TABLE]
One can easily check that
⋃i=13OrbZ7×Z13(Ti) ∪ ⋃i=45Orb{0}×Z13(Ti) is a 7-sun system of K92.
If k=11 and 4k+v=100, set V(K144)=(Z11×Z13)∪{∞}.
We consider the following 11-suns
[TABLE]
[TABLE]
One can check that
⋃i=13OrbZ11×Z13(Ti) ∪ ⋃i=46Orb{0}×Z13(Ti)
is an 11-sun system of K144.
It is left to prove the
existence of a k-sun system of K4kg+v when (k,v)=(7,21) and for every g≥1.
If g=1, a 7-sun system of K49 can be obtained as a particular case of the following construction. Let p be a prime, q=pn≡1(mod4) and r be a primitive root of Fq.
Setting
S=Dev⟨r⟩(0∼r∼r+1) where ⟨r⟩={jr∣1≤j≤p},
we have that ⋃i=04q−5OrbFq(r2iS) is a p-sun system of Kq.
If g≥2, we notice that K28g+21=K28(g−1)+49. Considering the
7-sun system of K49 just built, and recalling that by Theorem 4.1
there is a 7-sun system of K28+49, then Theorem 5.2 guarantees the existence
of a 7-sun system of K28(g−1)+49 whenever g=3.
When g=3, a 7-sun system of K105 is constructed as follows.
Set V(K105)=Z7×Z15.
Let
Si,j and T be the 7-suns defined below, where
(i,j)∈X=([1,3]×[1,7])∖{(1,3),(1,6)}:
[TABLE]
One can check that
(i,j)∈X⋃Orb{0}×Z15(Si,j) ∪ OrbZ7×Z15(T) is a 7-sun system of K105.
∎
6. Construction of p-sun systems, p prime
In this section we prove Theorem 1.2.
Clearly in view of Theorem 1.1 it is sufficient to construct a p-sun system of Kv for any admissible v with 2p<v<6p.
Hence, we are going to prove the following result.
Theorem 6.1**.**
Let p be an odd prime and let v(v−1)≡0(mod4p) with 2p<v<6p. Then there exists a p-sun system of Kv.
Since the existence of p-sun systems with p=3,5 has been proved in [10] and in [8], respectively, here we can assume p≥7.
It is immediate to see that by the necessary conditions for the existence of a p-sun system of Kv,
it follows that v lies in one of the following congruence classes modulo 4p:
v≡0,1(mod4p);
v≡p,3p+1(mod4p) if p≡1(mod4);
v≡p+1,3p(mod4p) if p≡3(mod4).
If v≡0,1(mod4p) we present a direct construction which holds more in general for p=k, where
k is an odd integer and not necessarily a prime.
Theorem 6.2**.**
For any k=2t+1≥7 there exists a k-sun system of K4k+1
and a k-sun system of K4k.
Proof.
Let C be the k-cycle with vertices in Z so defined:
[TABLE]
Note that the list D1 of the positive differences in Z of C is D1=[1,2t]∪{3t}.
Consider now the ordered k-set D2={d1,d2,…,dk}
so defined:
[TABLE]
Obviously D1∪D2=[1,2k].
Let {c1,c2,…,ck}
be the increasing order of the vertices of the cycle C
and set ℓr=cr+dr for every r∈[1,k],
with r=2t+1, and
ℓ2t+1=c2t+1−d2t+1 when t is odd.
It is not hard to see that V={c1,c2,…,ck,ℓ1,ℓ2,…,ℓk} is a set.
Note also that V⊆{−3t−1}∪[−t,5t]∪{6t+2}.
Let S be the sun obtainable from C by adding the pendant edges {ci,ℓi}
for i∈[1,k].
Clearly, ΔS=±(D1 ∪ D2)=±[1,2k].
So we can conclude that if we consider the vertices of S as elements of Z4k+1,
the vertices are still pairwise distinct and ΔS=Z4k+1∖{0}.
Then, by applying Corollary 2.2 (with G=Z4k+1,m=1,w=0),
it follows that
OrbZ4k+1S is a k-sun system of K4k+1.
Now we construct a k-sun system of K4k.
Let S be defined as above and
note that dk=2k. Let S∗
be the sun obtained by S setting ℓk=∞.
It is immediate that if we consider the vertices of S∗ as elements of Z4k−1∪{∞},
then Corollary 2.2 (with G=Z4k−1,m=1,w=1) guarantees that OrbZ4k−1S∗ is a k-sun system of K4k.
∎
Example 6.3**.**
Let k=2t+1=9, hence t=4.
By following the proof of Theorem 6.2, we construct a 9-sun system of K37.
Taking C=(0,−1,1,−2,2,−3,3,−4,8), we have that
[TABLE]
Hence {ℓ1,ℓ2,…,ℓ9}={5,7,9,12,14,16,18,20,26}
and we obtain the following 9-sun S with vertices in Z37:
[TABLE]
such that ΔS=Z37∖{0}.
Therefore, OrbZ37S is a 9-sun system of K37.
From now on,
we assume that p is an odd prime number and denote by Σ the following p-sun:
[TABLE]
Lemma 6.4**.**
Let p be an odd prime. For any x,y∈Zp with x=0 and any i,j∈Zm with i=j
there exists a p-sun S such that ΔiiS=±x, ΔijS=y, ΔjiS=−y and ΔhkS=∅ for any (h,k)∈(Zm×Zm)∖{(i,i),(i,j),(j,i)}.
Proof.
It is easy to see that S=DevZp×{0}((0,i)∼(x,i)∼(y+x,j)) is the required p-sun.
∎
We will call such a p-sun a sun of type (i,j).
For the following it is important to note that if S is a p-sun of type (i,j),
then ∣ΔiiS∣=2, ∣ΔjjS∣=0 and ∣ΔijS∣=∣ΔjiS∣=1.
The following two propositions provide us p-sun systems of Kmp+1 whenever m∈{3,5}
and p≡m−2(mod4).
Proposition 6.5**.**
Let p≡1(mod4)≥13 be a prime. Then there exists a p-sun system of K3p+1.
Proof.
We have to distinguish two cases according to the congruence of p modulo 12.
Case 1. Let p≡1(mod12).
If p=13, we construct a 13-sun system of K40 as follows.
Let S be the following 13-sun whose vertices are labelled with elements of (Z13×Z3)∪{∞}:
[TABLE]
We have:
[TABLE]
Now it remains to construct a set T of edge-disjoint 13-suns such that
[TABLE]
In order to do this it is sufficient to take, T={T01i∣i∈[1,4]}∪{T02i∣i∈[1,2]}∪{T10i∣i∈[1,3]}∪{T12i∣i∈[1,2]}∪{T20i∣i∈[1,3]}∪{T21i∣i∈[1,3]}, where:
[TABLE]
We have that T∪OrbZ13×{0}S is a 13-sun system of K40.
Suppose now that p≥37. We proceed in a very similar way to the previous case.
Let r be a primitive root of Zp.
Consider the ((Zp×Z3)∪{∞})-labeling B of Σ so defined:
[TABLE]
except for 4p−9 values of i≡1(mod3) for which we set B(ℓi)=(ri+1,i).
Letting S=B(Σ),
it is immediate that the labels of the vertices of S are pairwise distinct.
Note that
[TABLE]
[TABLE]
Hence, reasoning as in the previous case, we have to construct a set T of p-suns
such that if i=j then ΔijT=Zp∖ΔijS is a set and also
ΔiiT=Zp∗∖ΔiiS is a set.
In particular, this implies that for any T,T′∈T we have ΔijT∩ΔijT′=∅
and that ∣Δ00T∣=∣Δ22T∣=p−1,
∣Δ11T∣=2p+7,
∣ΔijT∣=3p+2 for (i,j)∈{(0,2),(1,2),(2,0),(2,1)},
and ∣Δ01T∣=∣Δ10T∣=127p−7.
In order to do this it is sufficient to take T as a set consisting of
2p−1 suns of type (0,1),
12p−1 suns of type (1,0),
6p+11 suns of type (1,2),
3p+2 suns of type (2,0),
6p−7 suns of type (2,1),
which exist in view of Lemma 6.4.
We have that OrbZp×{0}S∪T is a p-sun system of K3p+1.
Case 2. Let p≡5(mod12). Let r be a primitive root of Zp.
Consider the ((Zp×Z3)∪{∞})-labeling B of Σ so defined:
[TABLE]
except for 6p−17 values of i≡0(mod3) with i∈[3,2p−1] for which we set
B(ℓi)=(ri−1,i) and
12p−5 values of i≡0(mod3) with i∈[2p+1,p−5] for which we set
B(ℓi)=(ri+1,i).
Letting S=B(Σ), it is easy to see that the labels of the vertices of S are pairwise distinct.
Note that
[TABLE]
Hence, we have to construct a set T of p-suns
such that ∣Δ11T∣=∣Δ22T∣=p−1,
∣Δ00T∣=2p+7,
∣Δ01T∣=∣Δ10T∣=2p−1,
∣Δ02T∣=∣Δ20T∣=125p−1, and
∣Δ12T∣=∣Δ21T∣=3p+4.
In order to do this it is sufficient to take T as a set consisting of
4p+7 suns of type (0,1),
4p−9 suns of type (1,0),
4p+7 suns of type (1,2),
125p−1 suns of type (2,0), and
12p−5 suns of type (2,1)
which exist in view of Lemma 6.4.
We have that OrbZpS∪T is a p-sun system of K3p+1.
∎
Proposition 6.6**.**
For any prime p≡3(mod4) there exists a p-sun system of K5p+1.
Proof.
Set p=4n+3, and let Y=[1,n] and X=[n+1,2n+1].
Consider the following (Zp×Z5)∪{∞}-labeling B of Σ defined as follows:
[TABLE]
One can directly check that the vertices of S=B(Σ) are pairwise distinct. Also,
it is not hard to verify that ΔS does not have repetitions
and that its complement in (Zp×Z5)∖{(0,0)} is the set
[TABLE]
Clearly, D can be partitioned into n+1 quadruples of the form
Dx={±(2x,0), ±(rx,sx)} with x∈X and sx=0.
Letting
[TABLE]
for x∈X,
it is clear that ΔSx=Dx, hence Δ{Sx∣x∈X}=D.
Therefore, Corollary 2.2 guarantees that
⋃x∈XOrb{0}×Z5(Sx) ∪ OrbZp×Z5(S)
is a p-sun system of K5p+1.
∎
Example 6.7**.**
Here, we construct a 7-sun system of K36 following the proof of Proposition 6.6.
In this case, Y={1} and X={2,3}. Now consider the 7-sun S defined below, whose vertices lie in (Z7×Z5)∪{∞}:
[TABLE]
We have
[TABLE]
Hence ΔS does not have repetitions
and its complement in (Z7×Z5)∖{(0,0)} is the set
[TABLE]
Now it is sufficient to take
[TABLE]
One can check that
⋃x∈XOrb{0}×Z5(Sx) ∪ OrbZ7×Z5S
is a 7-sun system of K36.
We finally construct p-sun systems of Kmp whenever p≡m(mod4).
Proposition 6.8**.**
Let m and p be odd prime numbers
with m≤p and m≡p(mod4). Then there exists a p-sun system of Kmp.
Proof.
For each pair (r,s)∈Zp∗×Zm, let Br,s:V(Σ)→Zp×Zm
be the labeling of the vertices of Σ defined as follows:
[TABLE]
Since Br,s is injective,
for every h∈Zm the graph S_{r,s}^{h}=\tau_{(0,h)}\big{(}B_{r,s}(\Sigma)\big{)} is a p-sun.
For i,j∈Zm, we also notice that
Δij{Sr,sh∣h∈Zm}={± r} whenever
i−j=±s, otherwise it is empty.
Letting S be the union of the following two sets of p-suns:
[TABLE]
it is not difficult to see that for every i,j∈Zm
[TABLE]
It is left to construct a set T of p-suns
such that ΔijT=Zp∖ΔijS whenever i=j,
and ΔiiT=Zp∗∖ΔiiS=Zp∗. Therefore,
[TABLE]
It is enough to take T as a set consisting of
one sun of type (h,h+x)
and 2p−m suns of type (h,h+1),
for every h∈Zm and x∈[1,2m−1].
These p-suns exist by Lemma 6.4, therefore
S∪T is the desired p-sun system of Kmp.
∎
Example 6.9**.**
Let (m,p)=(3,11). Following the proof of Proposition 6.8, we construct
an 11-sun system of K33. For every h∈Z3 and r∈[1,3], let Sr,1h be the 11-sun
defined below:
[TABLE]
One can check that Δij{Sr,10,Sr,11,Sr,12}={±r} if i=j,
otherwise it is empty. Therefore,
letting S={Sr,1h∣h∈Z3,r∈[1,3]},
we have that ΔijS is non-empty only when i=j, in which case
we have ΔijS=±[1,3].
Now let T={Thg∣h∈Z3,g∈[1,5]} where Thg is the 11-sun defined as follows:
[TABLE]
Note that each Thg is an 11-sun of type (h,h+1). Therefore we have that
[TABLE]
By Corollary 2.2, it follows that S∪T
is an 11-sun system of K33.
We are now ready to show that the necessary conditions for the existence of a p-sun system of
Kv are also sufficient whenever p is an odd prime. In other words,
we end this section by proving Theorem 6.1.
Proof of Theorem 6.1.
If p=3,5 the result can be found in [10] and in [8], respectively.
For p≥7, the result follows from Propositions 6.5, 6.6 and 6.8.
Acknowledgements
The authors gratefully acknowledge support from GNSAGA of Istituto Nazionale di Alta Matematica.