Note
Cyclically decomposing the complete graph into cycles
Hung-Lin Fu
a;1, Shung-Liang Wu
baDepartment of Applied Mathematics, National Chaio Tung University, Hsin Chu, Taiwan, ROC bNational United University, Miaoli, Taiwan, ROC
Received 9 July 2001; received in revised form 8 July 2003; accepted 28 December 2003
Abstract
Let m1; m2; : : : ; mk be positive integers not less than 3 and let n =ki=1mi. Then, it is proved that the complete graph
of order 2n + 1 can be cyclically decomposed into k(2n + 1) cycles such that, for each i = 1; 2; : : : ; k, the cycle of length mi occurs exactly 2n + 1 times.
c
2003 Elsevier B.V. All rights reserved.
Keywords: Complete graph; Cycle system; Skolem sequence; Hooked Skolem sequence; Near Skolem sequence
1. Introduction
A Steiner triple system (STS) is an ordered pair (V; B), where V is a <nite nonempty set of elements, and B is a collection of 3-element subsets of V called triples, such that each pair of distinct elements of V occurs together in exactly one triple of B. The order of a Steiner triple system (V; B) is the size of V , denoted by |V |.
From “graph decomposition” point of view, the existence of a Steiner triple system of order v (STS(v)) is equivalent to the existence of a decomposition of the complete graph Kvof order v into edge-disjoint triangles, denoted by C3. It is not
diBcult to see the necessary condition for such a decomposition to exist is that v ≡ 1 or 3 (mod 6). In fact, this condition was proved to be suBcient around 150 years ago by Kirkman [4]. An automorphism of a STS (V; B) is a bijection : V → V such that {x; y; z} ∈ B if and only if {(x); (y); (z)} ∈ B. A STS(v) is cyclic if it has an automorphism that is a permutation consisting of a single cycle of length v, for example (1; 2; 3; : : : v).
Cyclic Steiner triple systems do exist. In 1939, Peltesohn used the so-called diIerence method to settle the existence problem.
Theorem 1.1 (Peltesohn [7]). For all v ≡ 1 or 3 (mod 6) except v = 9, there exists a cyclic STS(v).
We move on to consider an analog of Steiner triple systems. An m-cycle system of order v is a pair (V; C), where V = V (Kv) and C is a collection of edge-disjoint m-cycles which partition the edge set of Kv. Let J be an automorphism
groupof the m-cycle system (V; C) (i.e., a groupof permutations on v vertices leaving the collection C of cycles invariant). If there is an automorphism ∈ J of order v, then the m-cycle system (V; C) is said to be cyclic. For an m-cycle system of Kv, the vertex set V can be identi<ed with Zv. It is easy to see the necessary conditions for such a decomposition are
(i) v is odd and (ii) m|v 2
.
The study of “existence problem” of m-cycle systems started around 40 years ago. Recently, Alspach and Gavlas [2] and LSajna [10] proved that an m-cycle system exists as long as the above conditions are met. Thus, we have all the m-cycle systems for each m ¿ 3.
1Research Supported by NSC 89-2115-M-009-041.
E-mail address:hlfu@math.nctu.edu.tw(H.-L. Fu).
0012-365X/$ - see front matter c 2003 Elsevier B.V. All rights reserved.
Conjecture. Let m1; m2; : : : ; mh be positive integers not less than 3 such that i=1mi = 2 for odd n (respectively,
h
i=1 mi=n2− n=2 for even n). Then Kn (respectively, Kn− F) can be decomposed into cycles C1; C2; : : : ; Ch such that
the length of Ci is mi for i = 1; 2; : : : ; h.
In this paper, we prove a special case of the conjecture, namely, we prove that if m1; m2; : : : ; mk are positive integers
all at least 3, then the complete graph K2n+1, where n =ki=1mi, has a cyclic decomposition into k(2n + 1) cycles such
that for each i = 1; 2; : : : ; k, there are exactly 2n + 1 cycles of length mi.
2. The main results
Throughout this paper, we shall use diIerence methods. The di5erence between two vertices x and y in the complete graph Kn with V (Kn) = Zn is |x − y| or n − |x − y|, whichever is smaller. We will say that the edge xy has diIerence
min{|x − y|; n − |x − y|}. Thus, the set of diIerences possible in Kn is {1; 2; : : : ; n=2} and each diIerence induces a
2-factor except the diIerence n=2 induces a 1-factor whenever n is even. For convenience, we shall use G[D] to denote the subgraph of G induced by the set of diIerences D ⊆ {1; 2; : : : ; ‘}. It is easy to check that K2‘+1[i] is a disjoint union
of cycles of length (2‘ + 1)=(2‘ + 1; i), where (2‘ + 1; i) denotes the greatest common divisor of 2‘ + 1 and i. Clearly, if (2‘ + 1; i) = 1, then K2‘+1[i] is a Hamiltonian cycle in K2‘+1. It should be mentioned that if cycles Ci (1 6 i 6 k) have
diIerence sets Ai which partition {1; 2; : : : ; ‘}, then there exists a cyclic decomposition of K2‘+1 into cycles Ci.
Notice that if H is a subgraph of K2‘+1such that each edge of H has a distinct diIerence, then the graph H +i obtained
from H by adding i (mod 2‘ + 1) to each vertex of H is an isomorphic copy of H. The following results are given in [13] and will be used in the proof of Theorem 2.5.
Lemma 2.1 (Wu [13]). For positive integers b and s, there exists a cycle C of length 4s with di5erence set {b; b + 1; : : : ; b + 4s − 1}
in Kn where n is odd with n ¿ 2(b + 4s − 1) + 1.
Lemma 2.2 (Wu [13]). Let b and s be positive integers. (1) There exists a cycle C of length 4s + 2 with di5erence set
{b; b + 1; : : : ; b + 4s; b + 4s + 2}
in Kn where n is odd with n ¿ 2(b + 4s + 2) + 1.
(2) There exists a cycle C of length 4s + 2 with di5erence set {b; b + 2; b + 3; : : : ; b + 4s + 2}
in Kn where n is odd with n ¿ 2(b + 4s + 2) + 1.
Note that one may use a consecutive block of integers to construct cycles of length congruent to 0 modulo 4 and/or an even number of cycles of length congruent to 2 modulo 4. For example, if m1= 4s + 2 and m2= 4t + 2, then applying (1)
and (2) of Lemma2.2give cycles C1 and C2 of lengths m1 and m2 with diIerence sets {b; b + 1; : : : ; b + 4s; b + 4s + 2}
and {b + 4s + 1; b + 4s + 3; : : : ; b + 4s + 4t + 3}, respectively, for any positive integer b. For convenience, in the following lemmas, we use a typical odd cycle as in Fig. 1.
Lemma 2.3. For positive integers a; b; c, and r, with c = a + b and r ¿ c, and a nonnegative integer s, there exists a cycle C of length 4s + 3 with di5erence set {a; b; c; r; r + 1; : : : ; r + 4s − 1} in Kn where n is odd and n ¿ 2(r + 4s − 1) + 1.
Fig. 1.
Proof. The proof is divided into two cases. Case 1: Either a or b is odd, say b.
The cycle C of length 4s + 3 is de<ned as the following:
An easy veri<cation shows that the vertices of the cycle C are: for i = 0; 1; : : : ; s, v2i+1= a + 2i, v2i+1 = c + 2i,
v2i= a − r − 2(i − 1), v2i = c + r + 4s − 2i + 1, where all indices are taken modulo n, and the diIerence set is
{a; b; c; r; r + 1; : : : ; r + 4s − 1}. Observe that since c = a + b and b is odd, it follows that a and c have opposite parity. Thus a; a + 2; : : : ; a + 2s and c; c + 2; : : : ; c + 2s have opposite parity and hence are distinct. Also, c + r + 4s − 1, c + r + 4s − 2; : : : ; c + r + 2s + 1 and a − r ; a − r − 2; : : : ; a − r − 2s − 2 have opposite parity when considered modulo n and thus are distinct. Therefore, the vertices of C are distinct.
Case 2: Both a and b are even.
(i) r is even: Let e1= a, e1= r + 4s − 2, e2= r + 4s − 1, e2= c, e2s+2= b, and for i = 3; 4; : : : ; 2s + 1, let ei= r +
4s − 2i + 3, e
i = r + 2i − 6. Now, we de<ne the vertices accordingly. Let v0= 0, v1= a, v1= r + 4s − 2 and for
i = 1; 2; : : : ; s; v2i= a + r + 4s − 2i + 1, v2i= c + r + 4s + 2i − 4, v2i+1= a + 2i, and v2i+1= c + 4s − 2i.
(ii) r is odd: In this subcase, we let e1= a, e1= r + 4s − 1, e2= r + 4s − 2, e2= c; e2s+2= b, and for i = 3; 4; : : : ; 2s +
1; ei= r + 4s − 2i + 2, ei = r + 2i − 5. Then according to the diIerences, we de<ne the vertices for the cycle. Let
v0= 0, v1= a, v1= r + 4s − 1, and for i = 1; 2; : : : ; s, v2i= a + r + 4s − 2i, v2i+1= a + 2i, v2i= c + r + 4s + 2i − 3,
and v
2i+1= c + 4s − 2i.
Remark. In Lemma 2.3, if c = a + b ± 2, then we can use a similar method mentioned above to construct a cycle of length 4s + 1 with diIerence set {a; b; c; r; r + 1; : : : ; r + 4s − 4; r + 4s − 2}. Notice that this construction will be used in Theorem2.5. See, for example, Fig.2, where a = 2; b = 3; c = 7; s = 3, and r = 9.
Notice that the value of a + b is even (odd) and c is odd (even). By routine computation, it follows that the diIerence set is {a; b; c; r; r + 1; : : : ; r + 4s − 3}, and the distinct vertices of C are: v1= a, v1= c, v2s= c + 2s − 3, v2s= c + r + 2s − 2,
and for i = 1; 2; : : : ; s − 1, v2i= c + 2i − 3, v2i= c + r + 4s − 1 − 2i, v2i+1= c − r − 2i − 1, and v2i+1 = c + 2i.
In order to prove the main theorem, we also need to use Skolem sequences, hooked Skolem sequences, and near Skolem sequences.
A Skolem sequence of order n is a sequence (s1; s2; : : : ; s2n) such that for each j ∈ {1; 2; : : : ; n}, there exists a unique
i ∈ {1; 2; : : : ; 2n} such that si= si+j= j. It is proved by Skolem [12] that such a sequence exists if and only if n ≡ 0 or 1 (mod 4).
A hooked Skolem sequence of order n is a sequence (s1; s2; : : : ; s2n+1) such that s2n= 0 and for each j ∈ {1; 2; : : : ; n},
there exists a unique i ∈ {1; 2; : : : ; 2n − 1; 2n + 1} such that si= si+j= j. These sequences are known to exist if and only
if n ≡ 2, 3 (mod 4) (see [6]).
An m-near Skolem sequence of order n (m 6 n) is a sequence (s1; s2; : : : ; s2n−2) of 2n − 2 integers which satis<es for
every j ∈ {1; 2; : : : ; n} \ {m}, there exists a unique i ∈ {1; 2; : : : ; 2n − 2} such that si= si+j= j. It is proved by Shalaby
[11] that an m-near Skolem sequence of order n exists if and only if either (1) m is odd and n ≡ 0 or 1 (mod 4) or (2) m is even and n ≡ 2 or 3 (mod 4).
Remark that a Skolem sequence (s1; s2; : : : ; s2n) of order n gives a partition of {1; 2; : : : ; 3n} into triples {j; n + i; n + i +
j|si= si+j= j} for j = 1; 2; : : : ; n. Similarly, a hooked Skolem sequence gives a partition of {1; 2; : : : ; 3n − 1; 3n + 1} into
triples {j; sj; tj} satisfying j+sj=tj (1 6 j 6 n) and an m-near Skolem sequence gives a partition of {1; 2; : : : ; 3n−2}\{m}
into triples {j; sj; tj} satisfying j + sj= tj for each j ∈ {1; 2; : : : ; n} \ {m}.
Now, we are ready for the proof of our main result.
Theorem 2.5. Let m1; m2; : : : ; mk be positive integers not less than 3 such that n =ki=1mi. Then there exists a cyclic
(m1; m2; : : : ; mk)-cycle system of order 2n + 1.
Proof. For convenience, let m1; m2; : : : ; mi1 denote the integers which are congruent to 3 modulo 4, mi1+1; mi1+2; : : : ; mi2
denote the integers which are congruent to 1 modulo 4, mi2+1; mi2+2; : : : ; mi3 denote the integers which are congruent to
0 modulo 4, and thus mi3+1; mi3+2; : : : ; mk will be the integers which are congruent to 2 modulo 4. It suBces to partition
the set {1; 2; : : : ; n} into sets A1; A2; : : : ; Ak such that:
• |Ai| = mi for each i with 1 6 i 6 k;
• each of the sets A1; A2; : : : ; Ai1 satis<es the conditions for the diIerence set given in Lemma2.3;
• each of the sets Ai1+1; Ai1+2; : : : ; Ai2 satis<es the conditions for the diIerence set given in Lemma2.4or the remark after Lemma 2.3;
• each of the sets Ai2+1; Ai2+2; : : : ; Ai3 satis<es the conditions for the diIerence set given in Lemma2.1; and
• each of the sets Ai3+1; Ai3+2; : : : ; Ak satis<es the conditions for the diIerence set given in either (1) or (2) in
Case 1: Suppose that i2≡ 0, 1 (mod 4). Clearly, if i2= 0, then it is easy to de<ne the sets A1; A2; : : : ; Ak by choosing
the diIerences for the cycles of length congruent to 0 modulo 4 <rst followed by choosing the diIerences for those cycles of length congruent to 2 modulo 4 last, using n + 1 for n as necessary. In fact, after de<ning the sets A1; A2; : : : ; Ai2,
if we left with a set {b; b + 1; : : : ; n} for some positive integer b, then we can easily choose the diIerences for sets Ai2+1; Ai2+2; : : : ; Ak.
Since i2 ≡ 0, 1 (mod 4), there exists a Skolem sequence of order i2 such that the set {1; 2; : : : ; 3i2} can be partitioned
into triples {i; si; ti} with i + si= ti for i = 1; 2; : : : ; i2. Suppose <rst that i2= i1 so that there are no cycles of length
congruent to 1 modulo 4. Then, the sets A1; A2; : : : ; Ai1 are de<ned as follows:
• 1, s1; t1∈ A1,
• 2, s2; t2∈ A2,
.. .
• i1, si1, ti1∈ Ai1, and
• starting with 3i1+ 1, assign the next m1− 3 consecutive integers to A1, the next m2− 3 consecutive integers to A2 and
so on until assigning mi1− 3 consecutive integers to Ai1.
Observe that the diIerences left arei1 i=1 mi+1,
i1
i=1 mi+2; : : : ; n, and as remarked earlier, the sets Ai2+1; Ai2+2; : : : ; Ak
are easily found.
Now suppose that i2¿ i1. Suppose <rst that i2− i1 is even. De<ne the sets A1; A2; : : : ; Ai1 as follows:
• 1, s1, t1∈ A1, • 2, s2, t2∈ A2, .. . • i1, si1, ti1∈ Ai1, • i1+ 2, si1+1, ti1+1∈ Ai1+1, • i1+ 1, si1+2, ti1+2∈ Ai1+2, .. . • i2, si2−1, ti2−1∈ Ai2−1, • i2− 1, si2, ti2∈ Ai2, and
• starting with 3i2+ 1, assign the next m1− 3 consecutive integers to A1, the next m2− 3 consecutive integers to A2 and
so on until assigning mi2− 3 consecutive integers to Ai2.
The diIerences remaining are i2
i=1 mi+ 1,
i2
i=1 mi+ 2; : : : ; n and the sets Ai2+1; Ai2+2; : : : ; Ak are easily found.
Now assume that i2 − i1 is odd. Then, by [11], there exists a 1-near Skolem sequence of order i2 so that the set
{2; 3; : : : ; 3i2− 2} can be partitioned into triples {i; si; ti} with i + si= ti for i = 2; : : : ; i2. If there are no cycles of length
congruent to 2 modulo 4, then we de<ne A1; A2; : : : ; Ai2 as follows:
• 2, s2, t2∈ A1, • 3, s3, t3∈ A2, .. . • i1+ 1, si1+1, ti1+1∈ Ai1, • i1+ 3, si1+2, ti1+2∈ Ai1+1, • i1+ 2, si1+3, ti1+3∈ Ai1+2, .. . • i2− 1, si2, ti2∈ Ai2−1, • 1, n − 1, n + 1 ∈ Ai2, and
• starting with 3i2− 1, assign the next m1− 3 consecutive integers to A1, the next m2− 3 consecutive integers to A2 and
• i1+ 2, si1+3, ti1+3∈ Ai1+2, .. . • i2− 1, si2, ti2∈ Ai2−1, • 1, 3i2− 1, 3i2+ 1 ∈ Ai2, • Ai3+1= {3i2; 3i2+ 2; 3i2+ 3; : : : ; 3i2+ mi3+ 1}, and
• starting with 3i2+ mi3+1+ 1, assign the next m1− 3 consecutive integers to A1, the next m2− 3 consecutive integers
to A2 and so on until assigning mi2− 3 consecutive integers to Ai2.
In either case, the diIerences remaining form a set of consecutive integers, and the sets Ai2+1; Ai2+2; : : : ; Ai3−1; Ai3+1; : : : ; Ak
are easily found.
Now, we have completed the partition of {1; 2; : : : ; n} into the sets A1; A2; : : : ; Ak. By Lemmas 2.1–2.4, we are able to obtain a cyclic cycle decomposition for K2n+1. This concludes the proof of the case when i2≡ 0 or 1 (mod 4).
Case 2: Suppose that i2≡ 2, 3 (mod 4). The proof can be obtained by using a procedure similar to those in Case 1,
and so we omit the details.
For clearness, we present two examples to show the idea of partition.
Example 1. n = 51. (m1; m2; : : : ; m10) = (3; 3; 7; 5; 9; 4; 4; 4; 6; 6), i2= 5. A1= {1; 6; 7}, A2= {2; 12; 14}, A3= {3; 8; 11; 16; 17, 18; 19}, A4 = {5; 9; 13; 20; 21}, A5 = {4; 10; 15; 22; 23; 24; 25; 26; 27}, A6 = {28; 29; 30; 31}, A7 = {32; 33; 34; 35}, A8 = {36; 37; 38; 39}, A9= {40; 41; 42; 43; 44; 46}, and A10= {45; 47; 48; 49; 50; 51}. Example 2. n = 51. (m1; m2; : : : ; m10) = (3; 3; 3; 7; 5; 5; 9; 4; 6; 6), i2= 7. A1= {1; 8; 9}, A2= {3; 14; 17}, A3= {4; 11; 15}, A4= {5; 13; 18; 22; 23; 24; 25}, A5= {7; 10; 16; 26; 27}, A6= {6; 12; 19; 28; 29}, A7= {2; 20; 21; 30; 31; 32; 33; 34; 35}, A8= {36; 37; 38; 39}, A9= {40; 41; 42; 43; 44; 46}, and A10= {45; 47; 48; 49; 50; 51}.
With the theorem we proved, the following known result can be obtained easily.
Corollary 2.6. For each m ¿ 3 there exists a cyclic m-cycle system of order v ≡ 1 (mod 2m).
Proof. Since v = 2km + 1 for some integer k ¿ 1, by Theorem2.5where we choose n = mk, we conclude the proof. Remark. By an independent eIort, we can also obtain the consequence, that is, for each odd prime p, there exists a cyclic p-cycle system.
Acknowledgements
The authors wish to thank the referees for their valuable comments and suggestions, and in particular, one of the referees for his/her considerable eIort to rewrite this paper into a more readable form.
References
[1] B. Alspach, Research problems, Problem 3, Discrete Math. 36 (1981) 333.
[3] M. Buratti, A.D. Fra, Existence of cyclic k-cycle systems of the complete graph, Discrete Math. 261 (2003) 113–125. [4] T.P. Kirkman, On a problem in combinations, Cambridge Dublin Math. J. 2 (1847) 191–204.
[5] A. Kotzig, On decompositions of the complete graph into 4k-gons, Mat.-Fyz. Cas. 15 (1965) 227–233. [6] E.S. O’Keefe, Veri<cation on a conjecture of Th. Skolem, Math. Scand. 9 (1961) 80–82.
[7] R. Peltesohn, Eine LRosung der beiden HeIterschen DiIerenzenprobleme, Compositio Math. 6 (1939) 251–257. [8] A. Rosa, On cyclic decompositions of the complete graph into (4m + 2)-gons, Mat.-Fyz. Cas. 16 (1966) 349–352.
[9] A. Rosa, On the cyclic decompositions of the complete graph into polygons with an odd number of edges, Casopis Pest. Math. 91 (1966) 53–63.
[10] M. LSajna, Cycle decompositions III: complete graphs and <xed length cycles, J. Combin. Des. 10 (2002) 27–78. [11] N. Shalaby, The existence of near Skolem and hooked near-Skolem sequences, Discrete Math. 135 (1994) 303–319. [12] Th. Skolem, On certain distributions of integers in pairs with given diIerences, Math. Scand. 5 (1957) 57–68. [13] S.L. Wu, Even (m1; m2; : : : ; mr)-cycle systems of the complete graph, Ars Combin., to appear.