Abstract
We construct face two-colourable embeddings of the complete graph \(K_{2n+1}\) in which every face is a cycle of length \(n\); equivalently biembeddings of pairs of symmetric \(n\)-cycle systems. We prove that the necessary and sufficient condition for the existence of such an embedding in an orientable surface is for \(n\ge 3\) to be odd, and in a nonorientable is for \(n\ge 4\).
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
An \(n\) -cycle system of order \(v\), \(n\)-CS\((v)\), is a partition of the edge set of the complete graph on \(v\) vertices into cycles of length \(n\). Consider a properly face two-coloured embedding of the complete graph \(K_v\) in which all the faces are cycles of length \(n\), then each colour class corresponds to a \(n\)-CS\((v)\); this pair of cycle systems is said to biembed in the surface.
In the case where \(n=3\), a \(3\)-CS(\(v\)) is more commonly referred to as a Steiner triple system of order \(v\), STS \((v)\). The spectrum for biembeddings of pairs of STS(\(v\)) in orientable surfaces is \(v\equiv 3\text { or }7 \,(\text {mod }12)\), [8, 9], while the spectrum for biembeddings of pairs of STS(\(v\)) in nonorientable surfaces is \(v\equiv 1\text { or }3 \,(\text {mod }6)\) and \(v\ge 9\), [5, 8].
When \(n=v\), an \(n\)-CS\((n)\) corresponds to a partition of the edge set of \(K_n\) into Hamilton cycles. In [2], Ellingham and Stephens established the spectrum for biembeddings of pairs of \(n\)-CS\((n)\) in nonorientable surfaces. The spectrum for orientable biembeddings of pairs of \(n\)-CS\((n)\) is yet to be resolved.
In this paper we provide both orientable and nonorientable biembeddings for \(n\)-cycle systems for \(n\ge 4\) in nonorientable surfaces and for odd \(n\ge 3\) in orientable surfaces. The family of \(n\)-cycle systems we consider are symmetric \(n\) -cycle systems, that is, \(n\)-cycle systems in which the number of cycles in the partition equals the number of vertices. As each cycle contains \(n\) edges such a system is the partition of the edge set of the complete graph \(K_{2n+1}\), i.e. an \(n\)-CS\((2n+1)\).
Suppose we have a biembedding of a pair of \(n\)-CS(\(2n+1\)) in some surface. Such a biembedding has \(2n+1\) vertices, \(n(2n+1)\) edges and \(2(2n+1)\) faces. If the biembedding is in an orientable surface, by Euler’s formula, \((3-n)(2n+1)=2-2g\), where \(g\) is the orientable genus, and hence \(n\) is odd. However if the biembedding is in a nonorientable surface, by Euler’s formula, \((3-n)(2n+1)=2-\gamma \), where \(\gamma \) is the nonorientable genus, and there is no restriction on the parity of \(n\). Further note that there is no triangulation of \(K_7\) in a nonorientable surface [3]. Hence necessary conditions for a biembedding of a pair of \(n\)-CS(\(2n+1\)) are,
-
for an orientable embedding, that \(n\ge 3\) is odd; and
-
for a nonorientable embedding, that \(n\ge 4\).
In Sects. 2 and 3 we prove that these necessary conditions are also sufficient.
We will use index 1 current graphs to construct the desired biembeddings of pairs of \(n\)-CS(\(2n+1\)). We follow the notation established in [7] and recommend this reference to the reader for a more in-depth discussion of current graphs. In particular, suppose that \(e\) is an edge in a current graph. If \(e\) is orientation preserving, i.e. Type 0, this will be indicated by a superscript 0; whereas, if \(e\) is orientation reversing, i.e. Type 1, this will be indicated by a superscript 1. The base graph that we will use is the complete graph on two vertices with edge multiplicity \(n\), that is \(nK_2\). Clearly \(nK_2\) is bipartite and hence, as observed in [4], an index 1 current graph \(\langle nK_2\rightarrow S, \mathbb {Z}_{2n+1}\rangle \) induces a proper face two-colouring to its lift. If the current graph \(\langle nK_2\rightarrow S, \mathbb {Z}_{2n+1}\rangle \) also satisfies Properties (i), (ii) and (iii) below, then the lift of the current graph is a biembedding of a pair of \(n\)-CS(\(2n+1\)) (see [6]).
-
(i)
The embedding \(nK_2\rightarrow S\) has one face.
-
(ii)
Each edge of \(nK_2\) is assigned a distinct current from \(\mathbb {Z}_{2n+1}\), moreover, if some edge is labelled \(i\), then no other edge is labelled \(-i\).
-
(iii)
Let \(e_1^{\epsilon _1}\ldots e_n^{\epsilon _n}\), where \(\epsilon _i\in \{0,1\}\) for \(1\le i\le n\), be the rotation at a vertex \(v\) of \(nK_2\rightarrow S\) with current \(c_i\) carried in the direction of edge \(e_i\) that has \(v\) as its initial vertex. Then \(c_1+\ldots +c_n\equiv 0\,(\text {mod }2n+1)\) and, for all \(1\le i\le n\), the sums \(\sum _{1\le j\le i}c_j\) are distinct modulo \(2n+1\).
Moreover, if \(S\) is orientable, then the biembedding is in an orientable surface, while if \(S\) is nonorientable, as the order of the current group is odd, the biembedding is in a nonorientable surface (see [7] for details).
Note that a necessary condition for Property (iii) to hold is that there exists a \(n\)-CS\((2n+1)\) with a cyclic automorphism; i.e. a cyclic \(n\) -CS \((2n+1)\). Consider a sequence
where \(-n\le a_i\le n\), the \(a_i\) are pairwise disjoint and, interpreting \(a_n\) as \(a_0\) and regarding \(k=a_i-a_{i-1}\) as its coset representative modulo \(2n+1\) in the range \([-n,n]\),
Then \(A\) is a block of a cyclic \(n\)-CS\((2n+1)\) and generates, under the automorphism \(a_i\rightarrow a_i+1\,(\text {mod }2n+1)\), a cyclic \(n\)-CS\((2n+1)\). We refer to \(A\) as a cyclic starter or simply as a starter of the cyclic \(n\)-CS\((2n+1)\). In [1], Buratti and Del Fra provide a construction for starters for cyclic \(n\)-CS\((2n+1)\) for all \(n\ge 3\); namely \(A=(a_0,a_1,\ldots ,a_{n-1})\), where
2 Orientable Biembeddings
In this section we will construct the required current graphs to prove the existence of orientable biembeddings of pairs of \(n\)-CS\((2n+1)\). Hence, for the remainder of this section we assume that \(n\) is odd and \(n\ge 3\).
Given \(A\) as above, the sequence
where the entries are adjusted modulo \(2n+1\) to be in the range \(-n\le a_0-a_i\le n\), is also a starter for a cyclic \(n\)-CS\((2n+1)\). We call this second starter the transpose of the first.
Consider an embedding \(nK_2\rightarrow S\) in which a clockwise orientation is assigned to one vertex, \(v_0\) say, and an anticlockwise orientation at the other vertex, \(v_1\) say. Suppose that all the edges are Type 0, i.e. orientation preserving, and further suppose that the rotation at \(v_0\) equals the rotation at \(v_1\). Then as \(n\) is odd, the embedding \(nK_2\rightarrow S\) has exactly one face and, as all the edges are Type 0, \(S\) is orientable.
Denote the rotation at \(v_0\) by \(e_0,e_1,\ldots ,e_{n-1}\). We direct all the edges from \(v_0\) to \(v_1\) and label edges \(e_i\), interpreting \(c_n\) as \(c_0\) and \(a_n\) as \(a_0\), with \(c_i=a_i-a_{i-1}\) for \(1\le i\le n\). Then the rotation of currents at \(v_1\) is \((-c_0,-c_1,\ldots ,-c_{n-1})\). Thus we have constructed a current graph \(\langle nK_2\rightarrow S, \mathbb {Z}_{2n+1}\rangle \) that satisfies Properties (i) to (iii). Hence, we have proved the following theorem.
Theorem 1
For all odd \(n\ge 3\) there exists a biembedding of a pair of \(n\)-CS\((2n+1)\) in an orientable surface.
Notice that in the proof of Theorem 1 it is not necessary for the embedding to be of a complete graph. Consider the partial \(n\)-cycle system \(n\)-PCS\((mn+1)\), where \(m\ge 2\), generated by a cyclic starter \(B=(b_0,b_1,\ldots ,b_{n-1})\). This gives a graph decomposition of the subgraph of \(K_{mn+1}\), on the vertex set \(\{0,1,\ldots ,mn\}\) composed of edges whose endpoints differ by, interpreting \(b_n\) as \(b_0\), elements of the set \(\{|b_i-b_{i-1}|\,(\text {mod }mn+1):1\le i\le n\}\). Then the transpose \(n\)-PCS\((mn+1)\) generated by \(B^T=(b_0-b_0,b_0-b_1,\ldots ,b_0-b_{n-1})\) is a graph decomposition of the same subgraph of \(K_{mn+1}\). Then the construction above yields the following theorem.
Theorem 2
For all odd \(n\ge 3\) and all \(m\ge 2\) a cyclic \(n\)-PCS\((mn+1)\) generated by a starter \(B\) biembeds with the cyclic \(n\)-PCS\((mn+1)\) generated by \(B^T\) in an orientable surface.
We illustrate the above theorem by an example. Let \(n=5\) and \(m=3\). Consider the cyclic starter \(B=(0,1,4,11,6)\), which has transpose \(B^T=(0,15,12,5,10)\). This yields a biembedding of a pair of 5-PCS(16) on the subgraph of \(K_{16}\) with vertex set \(\{0,1,\ldots , 15\}\) composed of edges whose endpoints differ, modulo 16, by elements in \(\{1, 3, 5, 6,7\}\).
3 Nonorientable Biembeddings
In the remainder of this paper, to aid the reader, we will denote the rotation at a vertex \(v\) as the first row of a table and indicate the label of the edge (taking it to be directed away from \(v\)) in the row below. Recall that we indicate a Type 1, i.e. orientation reversing, edge by a superscript 1, whereas in the following, an absence of a superscript indicates a Type 0, i.e. orientation preserving, edge. For example consider the current graph \(\langle 6K_2\rightarrow S,\mathbb {Z}_{13}\rangle \) in Fig. 1; the rotation at \(v_0\) is
and the rotation at \(v_1\) is
3.1 Even Cycles
In this subsection we will consider the case where \(n\) is even. We begin by describing an embedding \(nK_2\rightarrow S\) that satisfies Property (i) and in which \(S\) is a nonorientable surface.
Assign a clockwise orientation to one vertex, \(v_0\) say, and an anticlockwise orientation to the other vertex, \(v_1\) say. We will denote each of the \(n\) edges as \(e_i\) where \(1\le i\le n\). Now assign edges \(e_{n-3}\), \(e_{n-2}\) and \(e_n\) to be Type 1 and all the other edges to be Type 0. Finally set the rotation at \(v_0\) to be
and the rotation at \(v_1\) to be
This embedding results in precisely one face, namely
As the embedding, \(nK_2\rightarrow S\), has both Type 1 and Type 0 edges in parallel, the surface is nonorientable.
All that remains is to establish a labelling so that Properties (ii) and (iii) are satisfied. We consider two cases, where \(n\equiv 2\,(\text {mod }4)\) and where \(n\equiv 0\,(\text {mod }4)\).
First the case \(n\equiv 2\,(\text {mod }4)\). We use the Buratti and Del Fra solution to give the rotation at \(v_0\), namely
where a label is interpreted as its coset representative modulo \(2n+1\) in the range \([-n,n]\). Explicitly, this is the rotation
The rotation at \(v_1\) is then
All that remains to be checked is that the rotation at \(v_1\) does indeed yield a \(n-CS(2n+1)\). To do so consider the sequence of partial sums at \(v_1\), namely
The set of the first \(n/2-1\) partial sums is
the set of the partial sums from \(n/2\) to \(n-4\) is
the remaining partial sums are the elements of the set
If \(n-6>((n-2)/2+2)=n/2+1\), then \(n>14\) and \(|X\cup Y\cup Z|=n\). Hence for \(n>14\) and \(n\equiv 2\,(\text {mod }4)\) we have a biembedding of a pair of \(n\)-CS\((2n+1)\).
Next we consider the case \(n\equiv 0\,(\text {mod }4)\). Again we use the Buratti and Del Fra solution to establish the rotation at \(v_0\), namely
This time the rotation at \(v_1\) is
In this case the set of the first \(n/2-1\) partial sums is
the set of the partial sums from \(n/2\) to \(n-5\) is
the set of the remaining four partial sums is
If \(n-6>n/2+2\), then \(n>16\) and \(|X\cup Y\cup Z|=n\). Hence for \(n>16\) and \(n\equiv 0\,(\text {mod }4)\) we have a biembedding of a pair of \(n\)-CS\((2n+1)\).
3.2 Odd Cycles
We consider six cases, corresponding to residue classes modulo 12. In the cases where \(n\equiv 1,3,5,7,11\,(\text {mod }12)\) the rotation about \(v_0\) is obtained from the Buratti and Del Fra solution. When \(n\equiv 9\,(\text {mod }12)\) the rotation at \(v_0\) is slightly different, with two terms of the Buratti and Del Fra solution transposed. In all cases we will make repeated use of the following lemma.
Lemma 1
Let \(n\ge 5\) be an odd integer. Consider an embedding of the graph \(nK_2\) with vertices \(v_0\) and \(v_1\) in a surface \(S\). Suppose that there are rotations assigned to \(v_0\) and \(v_1\) and types assigned to the edges of \(nK_2\) so that \(nK_2\rightarrow S\) has only one face. Further suppose that there are edges \(a\), \(b\), \(c\) and \(d\) that are all Type 0 and the rotations at \(v_0\) and \(v_1\) are
Then by changing \(b\) and \(c\) to be Type 1 and the rotation at \(v_1\) to be
we obtain an embedding \(nK_2\rightarrow S'\) where \(S'\) is nonorientable and in which the embedding has only one face.
Proof
The embedding \(nK_2\rightarrow S\) has only one face, the facial walk of which is either
Changing \(b\) and \(c\) to be Type 1 and the rotation at \(v_1\) to be
the facial walk becomes one of either
As the embedding has Type 0 and Type 1 edges in parallel it is necessarilly nonorientable. \(\square \)
For each of the odd residue classes modulo 12 we start with a copy of \(nK_2\) in which one vertex, \(v_0\) say, is assigned a clockwise orientation and the other vertex, \(v_1\) say, is assigned an anticlockwise orientation and all the edges are assigned to be Type 0. We also set the rotations of both \(v_0\) and \(v_1\) to be \(e_1\;e_2\;\ldots \;e_{n-1}\;e_n\). As noted in Sect. 2, this embedding has precisely one face. To this embedding we will repeatedly apply Lemma 1 to obtain embeddings of \(nK_2\) in nonorientable surfaces that satisfy Property (i).
We will present the constructions in the following format. First we list the set of edges to which Lemma 1 is applied. When \(n\equiv 9\,(\text {mod }12)\) we give the rotations at \(v_0\) and \(v_1\). When \(n\not \equiv 9\,(\text {mod }12)\) the rotation at \(v_0\) corresponds to the Buratti and Del Fra solution and hence we just give the rotation at \(v_1\). We then form the set of partial sums to verify that \(v_1\) (and when \(n\equiv 9\,(\text {mod }12)\) also \(v_0\)) yields an \(n\)-CS\((2n+1)\).
3.2.1 \({\mathbf n=12k+1}\)
Apply Lemma 1 to the 4-tuples in the set
The rotation at \(v_1\) is
The set of partial sums, working modulo \(24k+3\), is
This set has cardinality \(n\) for all \(n\ge 13\).
3.2.2 \({\mathbf n=12k+3}\)
Apply Lemma 1 to the 4-tuples in the set
The rotation at \(v_1\) is
The set of partial sums, working modulo \(24k+7\), is
This set has cardinality \(n\) for all \(n\ge 15\).
3.2.3 \({\mathbf n=12k+5}\)
Apply Lemma 1 to the 4-tuples in the set
The rotation at \(v_1\) is
The set of partial sums, working modulo \(24k+11\), is
This set has cardinality \(n\) for all \(n\ge 17\).
3.2.4 \({\mathbf n=12k+7}\)
Apply Lemma 1 to the 4-tuples in the set
The rotation at \(v_1\) is
The set of partial sums, working modulo \(24k+15\), is
This set has cardinality \(n\) for all \(n\ge 7\).
3.2.5 \({\mathbf n=12k+9}\)
Apply Lemma 1 to the 4-tuples in the set
The rotation at \(v_0\) is
The set of partial sums for this rotation, working modulo \(24k+19\), is
The rotation at \(v_1\) is
The set of partial sums, working modulo \(24k+19\), is
Both sets of partial sums have cardinality \(n\) for all \(n\ge 21\).
3.2.6 \({\mathbf n=12k+11}\)
Apply Lemma 1 to the 4-tuples in the set
The rotation at \(v_1\) is
The set of partial sums, working modulo \(24k+23\), is
This set has cardinality \(n\) for all \(n\ge 11\).
3.3 Small Designs
The results in Sects. 3.1 and 3.2 establish for \(n\ge 4\) the existence of a biembedding of a pair of \(n\)-CS(\(2n+1\)) in a nonorientable surface except when \(n\in \{4,5,6,8,9,10,12,14,16\}\). In this subsection we deal with these values. Note that a current graph yielding a 6-CS(13) is given in Fig. 1.
The pair, \((A_4,B_4)\), of 4-CS(9)s biembed in a nonorientable surface as do the pair, \((A_5,B_5)\), of 5-CS(11)s.
\(A_5\) | \(B_5\) | ||
\(A_4\) | \(B_4\) | \((0,8,7,3,5)\) | \((7,0,4,10,5)\) |
\((0,1,5,3)\) | \((3,0,4,1)\) | \((1,9,8,4,6)\) | \((2,0,10,8,4)\) |
\((1,2,6,4)\) | \((8,0,2,4)\) | \((2,10,9,5,7)\) | \((3,0,5,4,1)\) |
\((2,3,7,5)\) | \((6,0,7,5)\) | \((3,0,10,6,8)\) | \((8,0,9,6,3)\) |
\((3,4,8,6)\) | \((5,0,1,2)\) | \((4,1,0,7,9)\) | \((6,0,1,10,2)\) |
\((4,5,0,7)\) | \((7,1,8,6)\) | \((5,2,1,8,10)\) | \((8,1,7,3,2)\) |
\((5,6,1,8)\) | \((6,1,5,4)\) | \((6,3,2,9,0)\) | \((5,1,6,7,8)\) |
\((6,7,2,0)\) | \((3,2,7,4)\) | \((7,4,3,10,1)\) | \((9,1,2,7,10)\) |
\((7,8,3,1)\) | \((6,2,8,3)\) | \((8,5,4,0,2)\) | \((9,2,5,3,4)\) |
\((8,0,4,2)\) | \((5,3,7,8)\) | \((9,6,5,1,3)\) | \((10,3,9,5,6)\) |
\((10,7,6,2,4)\) | \((6,4,7,9,8)\) |
Finally, we present rotations at \(v_0\) and \(v_1\) that provide current graphs that yield nonorientable biembeddings of pairs of \(n\)-CS\((2n+1)\) for \(n\in \{8,9,10,12,14,16\}\). It is straightforward to verify that each current graph has only one face.
\({\mathbf n=8}\)
\({\mathbf n=9}\)
\({\mathbf n=10}\)
\({\mathbf n=12}\)
\({\mathbf n=14}\)
\({\mathbf n=16}\)
Having dealt with these small cases we have completed the proof of the following theorem.
Theorem 3
For all \(n\ge 4\) there exists a biembedding of a pair of \(n\)-CS\((2n+1)\) in a nonorientable surface.
References
Buratti, M., Del Fra, A.: Existence of cyclic \(k\)-cycle systems of the complete graph. Discrete Math. 261, 113–125 (2003)
Ellingham, M.N., Stephens, C.: The nonorientable genus of joins of complete graphs with large edgeless graphs. J. Comb. Theory Ser. B. 97, 827–845 (2007)
Franklin, P.: A six color problem. J. Math. Phys. 16, 363–369 (1934)
Grannell, M.J., Griggs, T.S.: Designs and topology, in surveys in combinatorics. In: Hilton, A.J.W., Talbot, J. (Eds.), London Math. Soc. Lecture Note Series 346, pp. 121–174. Cambridge Univ. Press, Cambridge (2007)
Grannell, M.J., Korzhik, V.P.: Nonorientable biembeddings of Steiner triple systems. Discrete Math. 285, 121–126 (2004)
Gross, J.L., Alpert, S.R.: The topological theory of current graphs. J. Comb. Theory Ser. B 17, 218–233 (1974)
Gross, J.L., Tucker, T.W.: Topological graph theory. Wiley, New York (1987)
Ringel, G.: Map color theorem. Springer, New York (1974)
Youngs, J.W.T.: The mystery of the Heawood conjecture, in graph theory and its applications, pp. 17–50. Academic Press, New York (1970)
Acknowledgments
The authors wish to thank a referee for a very careful reading of the paper and comments which have made the exposition clearer.
Author information
Authors and Affiliations
Corresponding author
Rights and permissions
About this article
Cite this article
Griggs, T.S., McCourt, T.A. Biembeddings of Symmetric \(n\)-Cycle Systems. Graphs and Combinatorics 32, 147–160 (2016). https://doi.org/10.1007/s00373-015-1538-1
Received:
Revised:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00373-015-1538-1