Circular Chromatic Number of Even-Faced Projective Plane Graphs DRAFT 2: NOT FOR DISTRIBUTION Luis Goddyn* Department of Mathematics Simon Fraser University Burnaby, BC, Canada E-mail: [email protected] We derive an exact formula for the circular chromatic number of any graph embeddable on the projective plane in such a way that all of it faces have even length. 1. INTRODUCTION The circular chromatic number χc (G) of a graph G, is an invariant which refines the chromatic number χ(G). More precisely, if G is finite and loopless, then χc (G) is a rational number in [2, ∞) satisfying χ(G) = dχc (G)e. Of the numerous equivalent definitions of χc (G) [37], perhaps the easiest to state is the following. For any loopless graph G, χc (G) is the least real number c such that there exists a function from V (G) to the set of unitlength open arcs of a circle of circumference c, such that adjacent vertices map to disjoint arcs. If G has a loop, then we define χc (G) = ∞. It is NP-hard to calculate χc (G), even if G is planar and χ(G) is part of the input [?]. In this paper we give an exact, and quickly computable, formula for χc (G) which applies to a particular class of graphs. An even-faced projective plane graph is a graph which has been embedded in the projective plane in such a way that a walk around the boundary of any face has even length. A surface subgraph of an embedded graph G is any embedded graph which may be obtained from G by successively deleting edges which bound distinct faces. We make these definitions pre- * Supported in part by the National Sciences and Engineering Research Council of Canada, and the Pacific Institute for the Mathematical Sciences 1 2 DRAFT: NOT FOR DISTRIBUTION cise in Section 2. Let maxfl(G) denote the maximum length of a face of an embedded graph G. We may now state our main result. Theorem 1.1. Let G be an even-faced projective plane graph. Then either G is bipartite (in which case χc (G) = 2), or χc (G) = 2 + 2 r−1 where 2r is the least value of maxfl(H) among all surface subgraphs H of G. In particular, every even-faced projective plane graph has circular chromatic number 2 + 2/s for some s ∈ {0, 1, . . . , ∞}. (We interpret 2/0 = ∞ and 2/∞ = 0.) The case k = 1 of Theorem 1.1 corresponds to G having a loop, which is necessarily noncontractible. That loop induces a surface subgraph H for which maxfl(H) = 2. The case k = 2 strengthens a result of Youngs [35]. He shows that loopless quadrangulations G of the projective plane satisfy χc (G) ∈ {2} ∪ (3, 4] whereas our result implies that χc (G) ∈ {2, 4}. An application of this regards the well-known Mycielski transformation [23]: given a graph G, we join a new copy v 0 of each v ∈ V (G) to the neighbours of v, then we join a new vertex x to all vertices of the form v 0 . This constructs, from any simple graph G, a new graph denoted µ(G) having the same clique number as G, and such that χ(µ(G)) = χ(G) + 1. For any odd circuit C2k+1 , the graph µ(C2k+1 ) embeds on the projective plane as a nonbipartite quadrangulation. (Here µ(C5 ) is the famous Grötzsch graph.) Our results imply that χc (µ(C2k+1 )) = 4 even though χc (C2k+1 ) = 2+1/k. This result already appears in [4]. We add that a similar statement holds true for the ‘generalized Mycielski construction’, a variation which which has recently gained attention with regard to graph homomorphisms [29]. When applied to an odd circuit, the generalized Mycielski construction also results in a nonbipartite quadrangulation whose circular chromatic number is exactly 4. The case k = 3 implies that any loopless nonbipartite even-faced projective plane graph G containing no quadrangulation has circular chromatic number 3. Every hexangulation of the projective plane satisfies χc (G) ∈ {2, 3, 4, ∞}, whereas octangulations satisfy χc (G) ∈ {2, 8/3, 3, 4, ∞}. A related result of Gimbel and Thomassen [10] states that all projective plane graphs with girth at least 5 satisfy χc (G) ≤ 3. Devos et. al. [6] consider even-faced graphs G embedded on an arbitrary surface. They show that, provided G has large edge-width, χc (G) is either very close to 2, or χc (G) ≥ 2 + 2/(k − 1) where 2k = maxfl(G). This paper shows that their lower bound is exact, in a sense, in case the surface is the projective plane. DRAFT: NOT FOR DISTRIBUTION 3 An efficient way of determining the value k in Theorem 1.1 is described in the next section. 2. DEFINITIONS We follow the definitions as in chapter 3 of [22], with exceptions as noted. A surface X is a compact connected Hausdorff space in which each point has a neighbourhood homeomorphic to the complex plane C. Thus every surface is homeomorphic to either a sphere with h handles, denoted Sh , for some nonnegative integer h, or a sphere with k crosscaps, denoted Nk for some positive integer k. The surfaces S0 , S1 , N1 , N2 are respectively called the sphere, the torus, the projective plane and the Klien bottle. A homeomorphic image of C on X is called an open disc. A homeomorphic image of R on X is called an open arc For X ⊆ σ, we denote by X, int(X) and δX, the (topological) closure, interior and boundary of X. A (closed) bordered surface with k holes is the topological space X(k) := X − ∪ki=1 Di where X is a surface and D1 , . . . Dk are open discs in X whose closures are (1) (2) pairwise disjoint. Thus S0 is a closed disc. The bordered surfaces S0 (1) and N1 are respectively called the closed cylinder and Möbius band . An embedding of a graph G on a bordered surface X is function φ on V (G) ∪ E(G) mapping vertices to distinct points on X and edges to disjoint open arcs in X such that for any uv ∈ E(G) we have δφ(uv) = {φ(u), φ(v)}, and such that each arcwise connected component of X − φ(V ∪ E) is an open disc (called a face). Such a pair (G, φ) is called an embedded graph or an X-embedded graph. We often write G for (G, φ), and |G| for X when the embedding is understood. Accordingly, we often identify the vertices and edges of G with their images under φ. We denote the set of faces of an embedded graph G by F (G). An embedded graph G = (V, E, F ) can be combinatorially described by specifying the graph (V, E) and, for each f ∈ F (G), a closed walk Wf in (V, E) specifying the boundary of f . The walk Wf is called a boundary walk of f . It is easy to see that a triple (V, E, {Wf : f ∈ F }) specifies an embedded graph if and only if each e ∈ E appears exactly twice among all the boundary walks, and for each v ∈ V there is a cyclic list (e0 , e1 , . . . , ed = e0 ) of incident edges such that ei and ei+1 appear as consecutive edges in some boundary walk for 0 ≤ i < d. The cyclic list is called the rotation at v, and d is the degree of v. Any e ∈ E appears exactly twice among all rotations, and e may appear twice in a single rotation or a single boundary walk. For f ∈ F , any edge appearing in Wf is incident with f . The length of f is the length of Wf . By interchanging the roles of V and F in G = (V, E, F ) in an obvious way, one obtains a new embedded graph G∗ = (V ∗ , E ∗ , F ∗ ) = (F, E, V ) called the surface dual of G. There is an obvious bijection between E and E ∗ . 4 DRAFT: NOT FOR DISTRIBUTION A walk in G∗ corresponds to a sequence of edges and faces of G which is called a dual walk in G. A dual walk is simple it corresponds to a path or circuit in G∗ . If e ∈ E(G) bounds distinct faces, then an new embedded graph, denoted G\e may be obtained by deleting e and identifying its two incident faces in an obvious way. If e ∈ E(G) is not a loop, then a new embedded graph, denoted G/e may be obtained by contracting e and identifying its two incident vertices in an obvious way. Any embedded graph obtained from G by successively deleting edges and contracting edges is called a surface minor of G. A b-contraction of G is a contraction G/e such that e is incident to a vertex of degree one. Any embedded graph obtained from G by successively deleting edges and applying b-contractions is called a surface subgraph of G. If G is an embedded graph, then the notation H ⊆ G indicates that H is a surface subgraph of G. An embedded graph G is even-faced if all its faces have even length. This is equivalent to saying G∗ is Eulerian. We briefly review some notions of surface topology. This development essencially follows [27]. A curve on a bordered surface X is a continuous function C : [0, 1] → X. The curve is closed if C(0) = C(1), open if C(0) 6= C(1), and simple if C is injective on [0, 1). Let C, C 0 be closed curves on X. A transformation from C to C 0 on X is a continuous function T : [0, 1] × [0, 1] → X such that for each t ∈ [0, 1] the projection x 7→ T (x, t) is a closed curve on X, and in particular C(x) = T (x, 0), and C 0 (x) = T (x, 1). If such a transformation exists, then C and C 0 are freely homotopic. This is an equivalence relation whose equivalence classes are called free homotopy classes. A curve C is contractible on X, if C is free-homotopy equivalent to a constant function. If two curves C, D : [0, 1] → X satisfy C(1) = D(0), then their composition C · D is the curve is defined by (C · D)(x) = C(2x) if 0 ≤ x ≤ 1/2 D(2x − 1) if 1/2 < x ≤ 1. The inverse of C is defined by C −1 : x 7→ C(1−x). These definitions extend without difficulty to the free homotopy classes of X. The free homotopy classes of a bordered surface X forms a group under composition called the (first) fundamental group π1 (X) of X. The fundamental groups of S0 and C are trivial. The group π1 (N1 ) is the two-element group, whose nonidentity element corresponds to the noncontractible curves in N1 . The fundamental group of a cylinder is the free abelian group generated by a simple closed curve along one boundary component. A covering map is a surjective map f : X → Y preserving open sets which restricts to a homeomorphism in a neighbourhood of each x ∈ X. Each preimage f −1 (y) is a discrete subset of X whose cardinality is independent DRAFT: NOT FOR DISTRIBUTION 5 of X. We say X is simply connected if it has trivial fundamental group. A universal cover of Y is a simply connected space X for which there exists a universal covering map f : X → Y . A universal cover is unique up to homeomorphism. Two curves C, C 0 are homotopic if there exists a transformation T : [0, 1] × [0, 1] → X such that T (0, t) and T (1, t) are constant functions of t. This is an equivalence relation, where equivalent paths have common endpoints, and equalence classes are called homotopy classes. Fix a point p ∈ X, let Hom(p) denote the set of homotopy classes of curves C where C(0) = p. Each λ ∈ Hom(p) consists of curves C with a common terminus C(1). The universal covering surface of a bordered surface (with respect to p) is the space U (X) whose point set is Hom(p). A subset T ⊆ X0 is open if for any λ ∈ T , there exists an open disc N in X containing the terminus q of λ such that for each curve D in Hom(q) ∩ N , the composition C · D belongs to T . It turns out that U (X) is essencially independent of p. The continuous function φ : U (X) → X which maps any point λ to its terminus, is called the projection of X. The preimage of q ∈ X under φ is a discrete subset of U (X) called the fibre of q. In particular, U (N1 ) is the sphere, and the covering map is 2 to 1. For h ≥ 1, U (Sh ) is homeomorphic to C. For any simple closed dual walk ω = f0 , e1 , f2 , . . . , ek , f0 on an embedded graph G, the union of the (open) 1-cells and 2-cells corresponding to edges and faces of ω is a 2-manifold homeomorphic to either a cylinder or a Möbius band without boundary. We say that ω is two-sided if it is a cylinder, and one-sided if it is a Möbius band. If |G| = P, then ω is twosided if and only if the circuit in G∗ corresponding to ω is contractible. Let C 0 (x) = e2πix be the curve in C bounding the open set D 0 = {z ∈ C : |z| < 1}. Let X be a 2-manifold. A set D ⊆ X is an open disc if there is a homeomorphism from D 0 to D. Let D ⊆ X be an open disc. A curve C in X is called a boundary curve of D if C(x) = h(C 0 (x)) for some continuous extension h of a homeomorphism h0 : D0 → D. Let X ⊆ X. A closed curve C surrounds X on X, if there exists an open disc D ⊆ X containing X such that there exists a transformation on X − D from C to a boundary curve of D. We often identify a walk W on an embedded graph G with a corresponding curve CW on |G|. Each edge in W corresponds to a subinterval of [0, 1], which is mapped injectively by CW onto the arc in |G| representing that edge. A closed walk W on G is a boundary walk of an open disc D ⊆ |G| if it corresponds to a boundary curve CW of D. If W is a boundary walk of an open disc D in X, then it is possible to form a surface subgraph of G by successively deleting or b-contracting all edges whose interiors are contained in D. We denote this surface subgraph by G − D. We have D ∈ F (G − D). If |G| 6= SS, then a closed walk W in G is the boundary walk of at most one open disc D ⊆ X. If D is the unique open disc with boundary walk 6 DRAFT: NOT FOR DISTRIBUTION W , then D is called the interior of W , written D = int(W ). For example, int(C) is well defined for any contractible circuit in G, provided |G| 6= SS. If int(W ) exists, then we can not conclude that W is a circuit of G. For example, if G consists of a single vertex with a single noncontractible loop on the projective plane, then the unique face of G has length 2 whose boundary a walk traverses that loop exactly twice. The edge-width of G, written edgewidth(G), is the length of a shortest noncontractible closed walk in G. Let maxfl(G) denote the maximum length of a face of G. Let minmaxfl(G) denote the minimum possible value of maxfl(H) among the surface subgraphs H of G. That is, minmaxfl(G) := min max length(f ). H⊆G f ∈F (H) (1) This is the invariant of interest, according to Theorem 1.1. Let f be a face of an embedded graph G. We denote by W(f ) = WG (f ) the set of closed walks in G which surround f . A family F of subsets of a set X is laminar if, for any x, y ∈ F, at least one of the sets x − y, y − x, x ∩ y is empty. 3. PRELIMINARIES Here we derive an alternative formulation of minmaxfl(G) which is better suited for proofs and for its efficient computation. The proofs of the following propositions have been omitted since they are not difficult, although they are somewhat tedious to describe. [[Should I do it?]] Proposition 3.1. Let H be a S0 -embedded graph, let v∞ ∈ V (H). For each v ∈ V (H) − {v∞ } let Xv ⊆ V (H) be a minimal set of vertices such that v ∈ Xv and δXv is a minimum size edge cut separating v from v∞ . Then {Xv : v ∈ V (H) − {v∞ }} is a laminar family. Proof. Let u, v ∈ V (H) − {v∞ }. Suppose that each of the sets U := Xu − Xv V := Xv − Xu W := Xu ∩ Xv is nonempty. Then the set Z := V (H) − U − V − W is nonempty since v∞ ∈ Z. By submodularity of edge-cut sizes we have the following. |δU | + |δV | ≤ |δXu | + |δXv | |δW | + |δZ| ≤ |δXu | + |δXv | (2) (3) Suppose that u ∈ U and v ∈ V . By choice of Xu and Xv we have |δU | > |δXu | and |δV | > |δXv |. These inequalities contradict (2), so either u or v belongs to W . We may assume u ∈ W . By choice of Xu we have that |δZ| ≥ |δXu |. DRAFT: NOT FOR DISTRIBUTION 7 If v ∈ W , then by choice of Xv we have |δW | > |δXv |, and the last two inequalities contradict (3). Thus v ∈ V and u ∈ W . By choice of Xv and Xu we have |δZ| ≥ |δXv | and |δW | > |δXu |. Adding these inequalities, we get a contradiction to (3) which proves the proposition. 8 DRAFT: NOT FOR DISTRIBUTION Proposition 3.2. Let G be an embedded graph where |G| is not the sphere. For each f ∈ F (G), let Wf be a shortest walk in W(f ) which surrounds as few faces of G as possible. Then {int(Wf ) | f ∈ F (G)} is a laminar family of open discs in |G|. Additionally, if f, f 0 ∈ F (G) and int(Wf ) ⊆ int(Wf 0 ), then length(Wf ) ≤ length(Wf 0 ). Proof. Let G0 be the universal covering graph of G. If |G| = N1 then |G0 | = S0 and G0 is a 2-fold cover of G. Otherwise |G0 | is homeomorphic to C (by a theorem of Poincaré), and G0 is an infinite graph. Let π : |G0 | → |G| be the projection function. For each f 0 ∈ F (G0 ), let Uf 0 be the unique closed walk in G0 surrounding f 0 such that π(Uf 0 ) = Wπ(f 0 ) . [[We assume 0 0 that π is injective on such int(f 0 )]] It suffices to prove that {int(Uf 0 ) | f ∈ F (G )} 0 is a laminar family of open discs in |G |. We consider the case |G| = N1 . Here π is 2 to 1 and |G0 | = S0 . The two preimages of f ∈ F (G) are said to be opposite on G0 . Let H be the surface dual of G0 . Suppose {int(Uf 0 ), int(Ug0 )} is not laminar for some f 0 , g 0 ∈ V (H). The vertices of H opposite to f 0 and g 0 on H are connected by a path whose edges are disjoint from int(Uf 0 ) ∪ int(Ug0 ). By contracting these edges, we get a new surface subgraph H 0 in which f∞ denotes the new vertex corresponding to the contracted path. Since Wf surrounds f , deleting the edge set E(Uf 0 ) from H 0 separates f 0 from f∞ . By hypothesis, E(Uf 0 ) is a minimal f 0 , f∞ -separating edge cut in H 0 which has the form δXf for some minimal subset Xf 0 of V (H 0 ) where f 0 ∈ Xf 0 . A similar statement holds for the edges of Ug0 . The sets Xf 0 and Xg0 are not laminar. this contradicts the statement of Proposition 3.1, and proves the case that |G| is the projective plane. Now suppose |G| is not the projective plane. Let H be the infinite dual graph of G0 , and suppose that {int(Uf 0 ), int(Ug0 )} is not laminar for some f 0 , g 0 ∈ V (H). YIKES! Let F ⊆ F (G0 ) be such that for each f ∈ F (G) there S exists a unique 0 f ∈ F such that π(int(f 00 )) = int(f ), and the closure of {int(f 0 ) | f 0 ∈ F} is a closed disc in |G0 |. It suffices to prove that {int(Uf 0 ) | f 0 ∈ F} is a laminar family of open discs in |G0 |. Select f 0 ∈ F (G0 ) such that f 0 is an arcwise connected component of −1 π (f ). Proof. OLD ATTEMPT: Let f, f 0 ∈ F (G). Let W ∈ W(f ), W 0 ∈ W(f 0 ) be chosen as in the hypothesis. Let X = int(W ), X 0 = int(W 0 ). Suppose each of X − X 0 , X 0 − X, X ∩ X 0 is nonempty. Each connected component of X − W 0 is an open disc. At least one of these discs, say D, has a boundary walk δD which consists of a subwalk of S of W , followed 9 DRAFT: NOT FOR DISTRIBUTION by a subwalk S 0 of W 0 . We consider two cases, depending on whether D ⊆ X − X 0 or D ⊆ X ∩ X 0 . Case 1 Suppose D ⊆ X − X 0 . We claim that int(f ) ⊆ D. Suppose not. Let W −S+S 0 denote the closed walk obtained by replacing S by S 0 in W . Then W −S +S 0 belongs to W(f ) and surrounds strictly fewer faces than does W . By choice of W , W −S +S 0 is strictly longer than W , so length(S 0 ) > length(S). Now W 0 − S 0 + S belongs to W(f 0 ) and is stictly shorter than W 0 . This contradicts the choice of W 0 and proves the claim. We have f ⊆ D ⊂ X. Therefore S + S 0 ∈ W(f ), so the choice of W implies that length(W − S) < length(S 0 ). (4) Suppose f 0 ⊆ X 0 − X. Then W 0 − S 0 + (W − S) ∈ W(f 0 ) and is strictly shorter than W 0 by (4), contradicting the choice of W 0 . Therefore f 0 ⊆ X 0 ∩ X. It follows that W − S + S 0 ∈ W(f 0 ). By choice of W 0 we have length(W 0 − S 0 ) ≤ length(W − S). On the other hand, W 0 − S 0 + S ∈ W(f ) so, by choice of W , we have length(W − S) ≤ length(W 0 − S 0 ). These two inequalities imply length(W − S) = length(W 0 − S 0 ). We have that W 0 − S 0 + S ∈ W(f 0 ) so, by choice of W 0 , length(S 0 ) ≤ length(S). Case 2 Suppose now D ⊆ X ∩ X 0. 10 DRAFT: NOT FOR DISTRIBUTION Lemma 3.3. Let f be a face of a graph G embedded on C or on P. Then the length of a shortest circuit in W(F ) can be computed in polynomial time. Proof. Suppose G is embedded on C. Since C − int(f ) is a topological cylinder, its homotopy group is the free cyclic group. A closed walk in C − Int(f ) belongs to W(f ) if and only if its homotopy class generates that group. If a vertex appears twice in a closed walk W ∈ W(f ), then, W has a proper subwalk whose homotopy class also generates the homotopy group. It follows that every shortest walk in W is a simple circuit in G. Let f ∗ and g ∗ denote the vertices of G∗ corresponding to f and the infinite face of G. Then the simple circuits in G correspond to minimal f ∗ , g ∗ separating edge cuts in G∗ . It follows that a shortest circuit in W(f ) can be found by generating G∗ and applying a max-flow min-cut algortithm to find a minimum weight f ∗ , g ∗ -separating edge cut G∗ . This is a polynomial time procedure. Suppose now that G is enbedded in the projective plane P. Let π : SS → P a universal 2-cover, and let G0 be embedded on SS such that π(G0 ) = G. Let f 0 , f 00 be the two face of G0 in the preimage of G0 . Then for any closed walk W in G, we have W ∈ W(f ) if and only if the preimage π(W ) is a circuit of unit homotopy class in SS − int(f 0 ) − int(f 00 ). Since G0 can be constructed in polynomial time, we are done by the previous paragraph. Remark 3. 1. It is interesting to ask whether Lemma 3.3 can be extended to graphs G embedded on other surfaces X. If X is not the plane or projective plane, then there exists an infinite graph G0 embedded on C and a universal cover π : C → X which maps G0 to G. One is tempted to compute a finite surface subgraph G00 ⊆ G0 consisting of all faces, edges and vertices of G0 that are within some distance d of a fixed face f 0 in φ−1 (f ). The idea is to compute a shortest closed walk in WG00 (f 0 ) as in the above proof. The problem is to polynomially bound the size of G00 . If X is the torus or Klein bottle, then |E| grows with the square of d, and such a scheme can be made to work, although we omit details here. When X has negative characteristic, then the size of G0 grows exponentically with d. Thus we ask the following: Given a (combinatorially described) graph G embedded on some 2-manifold, and a face f ∈ F (G), can a shortest closed walk surrounding f be computed in polynomial time? DRAFT: NOT FOR DISTRIBUTION 11 Lemma 3.4. For any graph G embedded on a surface X we have minmaxfl(G) = max min f ∈F (G) W ∈W(f ) length(W ). (5) Proof. Let k denote the right-hand side of (5). Let H0 be a surface subgraph of G which is optimal for (1). For f ∈ F (G), let Wf ∈ W(f ) be a closed walk which is optimal for the right hand side of (5). Every f ∈ F (G) is surrounded by some closed walk W f which is the boundary walk of some face of H0 . We have length(Wf ) ≤ length(W f ) ≤ minmaxfl(G). This inequality holds for every f ∈ F (G), so k ≤ minmaxfl(G). For each f ∈ F (G) we select Wf ∈ W(G) as in the previous paragraph, but, subject to this, Wf is chosen to surround as few faces of G as possible. By Proposition 3.1, {int(Wf ) | f ∈ F (G)} is a laminar family of open discs in X. We construct a surface subgraph H0 ⊆ G by deleting (or bcontracting) any edge of G whose interior is contained in int(Wf ) for some f ∈ F (G). Let W0 be the boundary walk of a longest face of H0 . By Proposition 3.1 we have W0 = Wf0 for some f0 ∈ F (G), and the face f0 achieves the maximum on the right-hand side of (5). We have k = length(Wf0 ) = max length(f ) ≥ minmaxfl(G). f ∈F (H0 ) We have proven equation (5). For some faces f ∈ F (G), the closed walks W ∈ W(f ) appearing on the right hand side of (5) might not be simple circuits in G. However, we may further restrict the set W(f ) in case |G| is the projective plane. Let C(f ) denote the set of walks in W(f ) which are simple circuits. It is possible that C(f ) is the empty set. We define the point-girth of an embedded graph G as follows. pointgirth(G) := max min length(C) f ∈F (G) C∈C(f ) We define pointgirth(G) = ∞ in case C(f ) = ∅ for some f ∈ F (G). Lemma 3.5. For any graph G embedded on the projective plane, we have minmaxfl(G) = min { 2 edgewidth(G), pointgirth(G) } (6) Proof. Since |G| is the projective plane, any noncontractible closed walk of minimum length is a circuit which induces a surface subgraph of 12 DRAFT: NOT FOR DISTRIBUTION G having exactly one face of length 2g, where g = edgewidth(G). Thus minmaxfl(G) ≤ 2g. Observing that C(f ) ⊆ W(f ) and equation (5), we have that minmaxfl(G) is at most the right hand side of (6). Suppose that f0 ∈ F (G) achieves the maximum in (5). Let W0 ∈ W(f0 ) have shortest possible length. Suppose that W0 is not a circuit. We aim to show that length(W0 ) ≥ 2g. Let x be a vertex occurring at least twice on W0 . Let then W0 is the concatenation of two nonempty closed subwalks W1 , W2 which begin and end at x. By Proposition ??, int(W0 ) is the boundary walk of a unique open disc in X. Therefore no proper subwalk of W0 is contractible on X. [[This may be vague]] In particular, W1 and W2 each have length at least g. It follows that length(W0 ) ≥ 2g. This completes the proof. We will use a formula for χc (G) which is based on orientations. An orientation of a graph G is a directed graph H obtained by assigning one of two possible directions to each edge of G. A walk in a directed graph H is a walk in the underlying undirected graph G. A circuit in H is a simple closed walk in H. Suppose W has a forward edges and b backward edges, where W has length a + b. Note that an edge appearing more than once in W may contribute more than once to a and b. We define the imbalance of W in H to be a+b imbalH (W ) ≥ min(a, b) We allow the value imbalH (W ) = ∞ if one of a, b equals zero. It was observed by Minty [20] that χ(G) can be expressed in terms of circuit imbalances. The following analogous result for χc (G) appears in [11]. Lemma 3.6. For any graph G we have χc (G) = min max imbalH (C), H C where H ranges over the set of orientations of G, and C ranges over the set of circuits in H. It is not difficult to see that this formula remains valid if we extend the set of closed walks over which C ranges. That is, for any graph G we have χc (G) = min max imbalH (W ), H W (7) where H ranges over the set of orientations of G, and W ranges over the set of closed walks in H. The following is a special case of the main result of [6]. We provide here a short self-contained proof of this special case. DRAFT: NOT FOR DISTRIBUTION 13 Lemma 3.7. Suppose that G is a nonbipartite even-faced graph embedded on the projective plane. Then χc (G) ≥ 2 + 2 r−1 Where the 2r = maxfl(G). Proof. An orientation of G is face-balanced if every face boundary walk in G has exactly half its edges oriented in each direction. We claim that G does not have a face-balanced orientation. Suppose H is a face-balanced orientation of G. Let e be an edge incident with distinct faces of H. Then H\e is a face-balanced orientation of G\e. A similar statement holds for any b-contraction H/e of H. Thus, every surface subgraph of H is facebalanced. Let C0 be an odd circuit in H. Then imbalH (C0 ) > 2. Furthermore C0 is a noncontractible circuit in H which induces a surface subgraph H0 of H. A boundary walk W0 of the unique face of H0 consists of two traversals of C0 , so imbalH (W0 ) = imbalH (C0 ) > 2. This contradiction proves the claim. Therefore for any orientation H of G, there exists a face f such that its boundary walk Wf satisfies imbalH (Wf ) ≥ 2s/(s − 1), where 2s = length(f ). The result follows from (7) and the fact s ≤ k. 4. PROOF OF MAIN RESULT Let G be an even-faced projective plane graph. We may assume G loopless and nonbipartite. Deleting an edge of G preserves the even-face property, so minmaxfl(G) is an even integer. A circuit in G has odd length if and only if it is noncontractible. Thus edgewidth(G) is odd and greater than two, and pointgirth(G) is even and positive, if not infinite. Let r, s, t be positive integers such that minmaxfl(G) = 2r, pointgirth(G) = 2s, edgewidth(G) = 2t + 1. We permit the value s = ∞. If s = ∞, then let f0 be any face of G. Otherwise, let f0 ∈ F (G) be such that every closed walk surrounding G has length at least 2s. Since |G| is the projective plane, the face f0 is the first and last face in some one-sided simple closed dual walk ω = f0 , e1 , f1 . . . , ek , f0 in G. (For example, we may define ω to correspond to a shortest noncontractible walk through f0 in G∗ .) We label the ends of each ei ∈ E(ω) with xi , yi so that the labels occur in the order x1 , x2 , . . . , xk , y1 , y2 , . . . , yk along the boundary of the 14 DRAFT: NOT FOR DISTRIBUTION Möbius band associated with ω. We point out that it is possible for one vertex to receive several labels in this scheme. Since G is loopless, the surface minor G0 = G/{e1 , . . . , ek } exists. The odd-length circuits in G are exactly the noncontractible circuits in G, and such circuits contain an odd number of the edges in ω. It follows that G0 is bipartite. By Lemma 3.6, there exists an edge-reorientation H 0 of G0 such that every circuit in H 0 has imbalance 2. We can extend H 0 to an edge-reorientation H of G by directing each ei ∈ E(ω) from xi to yi . It is not difficult to verify that every circuit C satisfies ±2 if C surrounds f0 a − b = ±1 if C is not contractible 0 otherwise. where C contains a forward edges and b backward edges of H. Therefore 2m m−1 if C surrounds f0 , and has length 2m 2`+1 imbalH (C) = if C is not contractible, and has length 2` + 1 ` 2 otherwise. By choice of f0 and the hypothesis, we have 2m ≥ 2s and 2` + 1 ≥ 2t + 1. Therefore every circuit C has imbalance at most 2 2 2s 2t + 1 =2+ , =2+ . max s−1 t min{s − 1, 2t} r−1 The final equation comes from Lemma 3.5. (If s = ∞, then no circuit surrounds f0 , and the above equation is still valid provided we interpret 2s s−1 = 2.) It follows from Lemma 3.6 that χc (G) ≤ 2 + 2/(r − 1). It remains to prove the opposite inequality. Let H be a surface subgraph of G such that maxfl(H) = minmaxfl(G) = 2r. Then χc (G) ≥ χc (H). However, we have χc (H) ≥ 2/(r − 1) by Lemma 3.7 This completes the proof. REFERENCES 1. D. Archdeacon, J. Hutchinson, A. Nakamoto, S. Negami, and K. Ota, Chromatic numbers of quadrangulations on closed surfaces, J. Graph Theory 37 (2001) 100– 114. 2. A. Bouchet, Nowhere-zero integral flows on a bidirected graph, J. Combin. Theory Ser. B 34 (1983) 279–292. 3. R. Brunet, B. Mohar, R. B. Richter, Separating and nonseparating disjoint homotopic cycles in graph embeddings, J. Combin. Theory Ser. B 66 (1996) 201–231. 4. G. Chang, L. Huang, X. Zhu, Circular chromatic numbers of Mycielski’s graphs, Discrete Math. 205 (1999) 23–37. DRAFT: NOT FOR DISTRIBUTION 15 5. M. DeVos, Flows on bidirected graph, preprint. 6. M. DeVos, L. Goddyn, B. Mohar, D. Vertigan, X. Zhu, Coloring-flow duality of embedded graphs, Trans. Amer. Math. Soc., submitted March 2003. 7. S. Fisk, The nonexistence of colorings, J. Combin. Theory Ser. B 24 (1978) 247–248. 8. S. Fisk, B. Mohar, Coloring graphs without short non-bounding cycles, J. Combin. Theory Ser. B 60 (1994) 268–276. 9. P. J. Giblin, Graphs, surfaces and homology. An introduction to algebraic topology, Second edition, Chapman and Hall, London-New York, 1981. 10. J. Gimbel, C. Thomassen, Coloring graphs with fixed genus and girth, Trans. Amer. Math. Soc. 349 (1997) 4555–4564. 11. L. A. Goddyn, M. Tarsi, C.-Q. Zhang, On (k, d)-colorings and fractional nowhere-zero flows, J. Graph Theory 28 (1998) 155–161. 12. L. A. Goddyn, D. Vertigan, X. Zhu, Minty type formulas, preprint. 13. M. de Graaf, A. Schrijver, Grid minors of graphs on the torus, J. Combin. Theory Ser. B 61 (1994) 57–62. 14. B. Grünbaum, Conjecture 6, in “Recent Progress in Combinatorics”, Ed. W. T. Tutte, Academic Press, New York, 1969, p. 343. 15. J. P. Hutchinson, Three-coloring graphs embedded on surfaces with all faces evensided, J. Combin. Theory Ser. B 65 (1995) 139–155. 16. J. Hutchinson, R. B. Richter, P. Seymour, Colouring Eulerian triangulations, J. Combin. Theory Ser. B 84 (2002) 225–239. 17. F. Jaeger, Flows and generalized coloring theorems in graphs. J. Combin. Theory Ser. B 26 (1979) 205–216. 18. F. Jaeger, Nowhere-zero Flow Problems, in, Selected Topics in Graph Theory, volume 3, pages 71–96, Academic Press, 1988. 19. B. Mohar, Graph minors and graphs on surfaces, in “Surveys in Combinatorics 2001,” Ed. J. W. P. Hirschfeld, London Mathematical Society Lecture Note Series 288, pp. 145–163. 20. G. J. Minty, A theorem on n-coloring the points of a linear graph, Amer. Math. Monthly 69 (1962) 623-624. 21. B. Mohar, P. D. Seymour, Coloring locally bipartite graphs on surfaces, J. Combin. Theory Ser. B 84 (2002) 301–310. 22. B. Mohar, C. Thomassen, Graphs on Surfaces, Johns Hopkins University Press, Baltimore, 2001. 23. J. Mycielski, Sur le coloriage des graphs, Colloq. Math. 3 (1955). 161–162. 24. A. Nakamoto, S. Negami, K. Ota, Chromatic numbers and cycle partitions of quadrangulations on nonorientable surfaces, preprint. 25. N. Robertson, P. D. Seymour, Graph minors VII. Disjoint paths on a surface, J. Combin. Theory Ser. B 45 (1988) 212–254. 26. G. Ringel, “Map color theorem”, Die Grundlehren der mathematischen Wissenschaften, Band 209. Springer-Verlag, New York-Heidelberg (1974). 27. A. Schrijver, Decomposition of graphs on surfaces and a homotopic circulation theorem, J. Combin. Theory Ser. B 51 (1991), 161–210. 28. P. D. Seymour, Nowhere-zero 6-flows, J. Combin. Theory Ser. B 30 (1981) 130–135. 16 DRAFT: NOT FOR DISTRIBUTION 29. C. Tardif, Fractional chromatic numbers of cones over graphs, J. Graph Theory 38 (2001) 87–94. 30. C. Thomassen, Five-coloring maps on surfaces, J. Combin. Theory Ser. B 59 (1993) 89–105. 31. C. Thomassen, The chromatic number of a graph of girth 5 on a fixed surface. J. Combin. Theory Ser. B, to appear. 32. C. Thomassen, Color-critical graphs on a fixed surface, J. Combin. Theory Ser. B 70 (1997) 67–100. 33. W. T. Tutte, A contribution to the theory of chromatic polynomials, Canad. J. Math. 6 (1954) 80–91. 34. W. T. Tutte, A class of Abelian groups, Canad. J. Math. 8 (1956) 13–28. 35. D. A. Youngs, 4-chromatic projective graphs, J. Graph Theory 21 (1996) 219–227. 36. X. Zhu, Planar graphs with circular chromatic numbers between 3 and 4, J. Combin. Theory Ser. B 76 (1999) 170–200. 37. X. Zhu, Circular chromatic number: a survey, Discrete Math. 229 (2001) 371–410. 38. X. Zhu, Construction of graphs with given circular flow number, preprint.
© Copyright 2026 Paperzz