Available online at www.sciencedirect.com
Electronic Notes in Discrete Mathematics 61 (2017) 925–931 www.elsevier.com/locate/endm
Packing and covering odd cycles in cubic plane graphs with small faces Diego Nicodemos 1,3 Col´egio Pedro II, COPPE/Sistemas Universidade Federal do Rio de Janeiro Rio de Janeiro, Brazil
Matˇej Stehl´ık 2,4 Laboratoire G-SCOP Univ. Grenoble Alpes Grenoble, France
Abstract We show that any 3-connected cubic plane graph on n vertices, with all faces of size at most 6, can be made bipartite by deleting no more than (p + 3t)n/5 edges, where p and t are the numbers of pentagonal and triangular faces, respectively. In particular, any such graph can be made bipartite by deleting at most 12n/5 edges. This bound is tight, and we characterise the extremal graphs. We deduce tight lower bounds on the size of a maximum cut and a maximum independent set for this class of graphs. This extends and sharpens the results of Faria, Klein and Stehl´ık [SIAM J. Discrete Math. 26 (2012) 1458–1469]. Keywords: Cubic plane graph, odd cycle, max-cut, independent set.
http://dx.doi.org/10.1016/j.endm.2017.07.055 1571-0653/© 2017 Elsevier B.V. All rights reserved.
926
1
D. Nicodemos, M. Stehlík / Electronic Notes in Discrete Mathematics 61 (2017) 925–931
Introduction
A set of edges intersecting every odd cycle in a graph is known as an odd cycle (edge) transversal, or odd cycle cover, and the minimum size of such a set is denoted by τodd . A set of edge-disjoint odd cycles in a graph is called a packing of odd cycles, and the maximum size of such a family is denoted by νodd . Clearly, τodd ≥ νodd . Dejter and Neumann-Lara [3] and independently Reed [9] showed that in general, τodd cannot be bounded by any function of νodd , i.e., they do not satisfy the Erd˝os–P´osa property. However, for planar graphs, Kr´al’ and Voss [8] proved the (tight) bound τodd ≤ 2νodd . In this paper we focus on packing and covering of odd cycles in 3-connected cubic plane graphs with all faces of size at most 6. Such graphs—and their dual triangulations—are a very natural class to consider, as they correspond to surfaces of genus 0 of non-negative curvature (see e.g. [12]). A much-studied subclass of cubic plane graphs with all faces of size at most 6 is the class of fullerene graphs, which only have faces of size 5 and 6. Faria, Klein and Stehl´ık [4] showed that any fullerene graph on n vertices has an odd cycle transversal with no more than 12n/5 edges, and characterised the extremal graphs. Our main result is the following extension and sharpening of their result to all 3-connected cubic plane graphs with all faces of size at most 6. Theorem 1.1 Let G be a 3-connected cubic plane graph on n vertices with all faces of size at most 6, with p pentagonal and t triangular faces. Then τodd (G) ≤ (p + 3t)n/5. In particular, τodd (G) ≤ 12n/5 always holds, with equality if and only if all faces have size 5 and 6, n = 60k 2 for some k ∈ N, and Aut(G) ∼ = Ih .
2
Outline of the proof
It is more convenient to consider the dual problem. An easy exercise shows that if G is a graph satisfying the hypothesis of Theorem 1.1, then the dual graph is a simple triangulation K of the sphere with all vertices of degree 1
Partially supported by CAPES and CNPq. Partially supported by ANR project Stint (ANR-13-BS02-0007), ANR project GATO (ANR-16-CE40-0009-01), and by LabEx PERSYVAL-Lab (ANR-11-LABX-0025). 3 Email:
[email protected] 4 Email:
[email protected] 2
D. Nicodemos, M. Stehlík / Electronic Notes in Discrete Mathematics 61 (2017) 925–931
927
Fig. 1. A 5-patch L of area 3 (shaded in dark grey) surrounded by a 5-moat Mt2 (L) of width 2 and area 16 (shaded in light grey).
at most 6. We denote the set of odd-degree vertices by T . An odd cycle transversal in G corresponds to a set of edges in K whose removal results in a graph with all vertices of even degree; such a set of edges is known as a T -join. More precisely, given a graph G = (V, E) with a distinguished set T of vertices of even cardinality, a T -join of G is a subset J ⊆ E such that T is equal to the set of odd-degree vertices in (V, J). The minimum size of a T -join of G is denoted by τ (G, T ). Clearly τodd (G) = τ (K, T ), so bounding τodd (G) is equivalent to bounding τ (K, T ). A closely related concept, a T -cut, is an edge cut δ(X) such that |T ∩ X| is odd. A packing of T -cuts is a disjoint collection δ(F) = {δ(X) | X ∈ F} of T -cuts of G; the maximum size of a packing of T -cuts is denoted by ν(G, T ). Since every T -join intersects every T -cut, the inequality τ (G, T ) ≥ ν(G, T ) always holds, but the complete graph on 4 vertices shows that τ (G, T ) may be strictly larger than ν(G, T ). A deep theorem of Seymour [10] asserts that τ (G, T ) = ν(G, T ) whenever G is bipartite. Given a subcomplex L ⊆ K, we denote the number of faces of L by area(L). An induced, contractible subcomplex L ⊆ K is called a c-patch, where c = u∈V (L) (6 − dK (u)); this quantity can be thought of as the ‘curvature’ of the patch. Given a c-patch L ⊆ K, a c-moat of width w is the set Mtw (L) of all faces in K \ L induced by vertices at distance at most w from L. With a slight abuse of notation, Mtw (L) will also denote the subcomplex of K formed by the faces in Mtw (L). The first key lemma in the proof of Theorem 1.1 gives the following tight lower bound on the area of a c-moat of width w surrounding a c-patch L. (1) area(Mtw (L)) ≥ (6 − c)w2 + 2w (6 − c) area(L). The next step is to prove the existence of a laminar packing M of 1-, 3-
928
D. Nicodemos, M. Stehlík / Electronic Notes in Discrete Mathematics 61 (2017) 925–931
and 5-moats in K of total width ν(K, T ). The idea is to take a certain minimal packing of inclusion-wise minimal T -cuts, and then show that it corresponds to a packing of 1-, 3- and 5-moats. Define the incidence vectors r, s, t ∈ R|T | as follows: for every u ∈ T , let ru , su , tu be the width of the 1-moat, 3-moat and 5-moat surrounding u, respectively. Furthermore, define the inner product ·, · on R|T | by x, y = u∈T (6 − d(u))xu yu and the norm · by x = x, x . With this inner product, the total width of 1-, 3- and 5-moats in M can be expressed as r, 1 , 13 s, 1 , and 15 t, 1 , respectively. Therefore, (2) ν(K, T ) = r + 13 s + 15 t, 1 . Let mc be the total area of c-moats of M, where c ∈ {1, 3, 5}. Using (1) and the Cauchy–Schwarz inequality, we obtain the following bounds on m1 , m3 and m5 . m1 = 5 r 2
(3)
√
m3 ≥ s 2 + 2 5r, s m5 ≥ 15 t 2 + 2r, t +
(4) (5)
√2 s, t . 5
The moats are pairwise disjoint, so by (3)–(5), √ area(K) ≥ m1 + m3 + m5 ≥ 5r + s +
2
√1 t 5
.
(6)
Inequalities (2) and (6), in conjunction with the Cauchy–Schwarz inequality, give the bound ν(K, T ) ≤ 15 u∈T (6 − d(u))area(K). (7) At this point we would like to apply Seymour’s theorem, but triangulations are clearly not bipartite. Luckily, this difficulty can be overcome with a simple trick. ˆ of K as follows. First, we subdivide each We construct the refinement K edge of K, that is, we replace it by an internally disjoint path of length 2, and then we add three new edges inside every face, incident to the three vertices of degree 2. (For an illustration, see Figure 2.) Therefore, every face of K is ˆ Observe that all the vertices in V (K) ˆ \ V (K) divided into four faces of K. ˆ so if T is the set of odd-degree vertices of K, then T is have degree 6 in K, ˆ also the set of odd-degree vertices of K.
D. Nicodemos, M. Stehlík / Electronic Notes in Discrete Mathematics 61 (2017) 925–931
929
Fig. 2. A face of a triangulation and its refinement.
An easy corollary of Seymour’s theorem proved in [4] is that ˆ T ). τ (K, T ) = 12 ν(K,
(8)
By combining (7) and (8) we easily get the desired bound τ (K, T ) ≤
1 5
u∈T (6
− d(u))area(K).
(9)
The bound in Theorem 1.1 follows from (9) by taking the planar dual. A careful analysis of the arguments shows that graphs attaining equality in the bound τodd (G) ≤ 12n/5 must be fullerene graphs on 60k 2 vertices with the full icosahedral automorphism group of a regular icosahedron. Conversely, if G is a fullerene graph with the full icosahedral automorphism group, then G can be parametrised using the Goldberg–Coxeter vector (see [1,5]). Using 2 the fact that G has 60k vertices, it easy to deduce that τodd (G) = 12n/5.
3
Consequences for max-cut and independence number
A classic problem in combinatorial optimisation, known as max-cut, asks for the maximum size of an edge-cut in a given graph. This problem is known to be NP-hard, even when restricted to triangle-free cubic graphs [13]. However, for the class of planar graphs, the problem can be solved in polynomial time using standard tools from combinatorial optimisation (namely T -joins), as observed by Hadlock [6]. Cui and Wang [2] proved that every planar, cubic graph on n vertices has a cut of size at least 39n/32 − 9/16, improving an earlier bound of Thomassen [11]. However, when the face size is bounded by 6, we get the following improved bound. Corollary 3.1 If G is a 3-connected cubic plane graph on n vertices with all faces of size at most 6, with p pentagonal and t triangular faces, then G has a cut of size at least 3n/2 − 12n/5.
930
D. Nicodemos, M. Stehlík / Electronic Notes in Discrete Mathematics 61 (2017) 925–931
Equality holds if and only if all faces have size 5 and 6, n = 60k 2 for some k ∈ N, and Aut(G) ∼ = Ih . A set of vertices in a graph is independent if there is no edge between any of its vertices, and the maximum size of an independent set in G is the independence number α(G). Heckman and Thomas [7] showed that every triangle-free, cubic, planar graph has an independent set of size at least 3n/8, and this bound is tight. Again, forbidding faces of size greater than 6 leads to a much better bound. Corollary 3.2 If G is a 3-connected cubic plane graph on n vertices with all faces of size at most 6, with p pentagonal and t triangular faces, then α(G) ≥ n/2 − 3n/5. Equality holds if and only if all faces have size 5 and 6, n = 60k 2 for some k ∈ N, and Aut(G) ∼ = Ih .
4
Concluding remarks
√ √ Clearly, a necessary condition for τodd = O( n) is that νodd = O( n). By the theorem of Kr´al’ and Voss [8] mentioned in the introduction, it is also a √ sufficient condition. It can be shown that νodd = O( n) is also a necessary √ and sufficient condition for having a max-cut of size at least √ 3n/2 − O( n), and for having an independent set of size at least n/2 − O( n). It is not hard to construct an infinite family of 3-connected cubic plane graphs with all faces of size at most 7 such that τodd ≥ εn, for a constant ε > 0. This shows that the condition on the size of faces in Theorem 1.1 and Corollaries 3.1 and 3.2 cannot be relaxed. Finally, we remark that bounding τodd is much simpler if the graph contains no pentagonalfaces. In this case, the bound in Theorem 1.1 can be improved to τodd (G) ≤ tn/3, where t is the number of triangular faces. In particular, τodd (G) ≤ 4n/3, with equality if and only if all faces have size 3 and 6, n = 12k 2 for some k ∈ N, and Aut(G) ∼ = Th . Corollaries 3.1 and 3.2 can be strengthened in the same way.
References [1] Coxeter, H. S. M., Virus macromolecules and geodesic domes, in: “A spectrum of mathematics (Essays presented to H. G. Forder),” Auckland Univ. Press,
D. Nicodemos, M. Stehlík / Electronic Notes in Discrete Mathematics 61 (2017) 925–931
931
Auckland, 1971, pp. 98–107. [2] Cui, Q. and J. Wang, Maximum bipartite subgraphs of cubic triangle-free planar graphs, Discrete Math. 309 (2009), 1091–1111. [3] Dejter, I. J. and V. Neumann-Lara, Unboundedness for generalized odd cyclic transversality, in: “Combinatorics (Eger, 1987),” Colloq. Math. Soc. J´anos Bolyai 52, North-Holland, Amsterdam, 1988, pp. 195–203. [4] Faria, L., S. Klein and M. Stehl´ık, Odd cycle transversals and independent sets in fullerene graphs, SIAM J. Discrete Math. 26 (2012), 1458–1469. [5] Goldberg, M., A class of multi-symmetric polyhedra, Tˆohoku Math. J. 43 (1937), 104–108. [6] Hadlock, F., Finding a maximum cut of a planar graph in polynomial time, SIAM J. Comput. 4 (1975), 221–225. [7] Heckman, C. C. and R. Thomas, Independent sets in triangle-free cubic planar graphs, J. Combin. Theory Ser. B 96 (2006), 253–275. [8] Kr´ al’, D. and H.-J. Voss, Edge-disjoint odd cycles in planar graphs, J. Combin. Theory Ser. B, 90 (2004), 107–120. [9] Reed, B, Mangoes and blueberries, Combinatorica 19 (1999), 267–296. [10] Seymour, P, On odd cuts and plane multicommodity flows, Proc. London Math. Soc. 42 (1981), 178–192. [11] Thomassen, C., On the max-cut problem for a planar, cubic, triangle-free graph, and the Chinese postman problem for a planar triangulation, J. Graph Theory 53 (2006), 261–269. [12] Thurston, W. P., Shapes of polyhedra and triangulations of the sphere, in: “The Epstein birthday schrift,” Geom. Topol. Monogr. 1, Geom. Topol. Publ., Coventry, 1998, pp. 511–549. [13] Yannakakis, M., Node- and edge-deletion NP-complete problems, in: “Conference Record of the Tenth Annual ACM Symposium on Theory of Computing (San Diego, Calif., 1978),” ACM, New York, 1978, pp. 253–264.