A note on dilation coefficient, plane-width, and resolution coefficient of graphs Martin Milaniˇc, University of Primorska—FAMNIT, Glagoljaˇska 8, 6000 Koper, Slovenia, [email protected] Tomaˇz Pisanski, University of Ljubljana, IMFM, Jadranska 19, 1111 Ljubljana, Slovenia, and 0 University of Primorska—PINT, Muzejski trg 2, 6000 Koper, Slovenia, 1 0 [email protected] 2 Arjana Zˇitnik n University of Ljubljana, IMFM, Jadranska 19, 1111 Ljubljana, Slovenia, a [email protected] J 2 Keywords: dilation coefficient; plane-width; resolution coefficient; graph representation; ] graph drawing; chromatic number; circular chromatic number; bandwidth. O C Math. Subj. Class. (2000) 05C62 . h Abstract t a m In this note we study and compare three graph invariants related to the ‘compactness’ of graph drawing in the plane: the dilation coefficient, defined as the smallest possible [ quotient between the longest and the shortest edge length; the plane-width, which is the 1 smallest possible quotient between the largest distance between any two points and the v shortest length of an edge; and the resolution coefficient, the smallest possible quotient 0 between the longest edge length and the smallest distance between any two points. These 3 3 three invariants coincide for complete graphs. 0 We show that graphs with large dilation coefficient or plane-width have a vertex with 1. large valence but thereexist cubic graphs with arbitrarily large resolution coefficient. Sur- 0 prisinglyenough,theone-dimensionalanaloguesofthesethreeinvariantsallowustorevisit 0 the three well known graph parameters: the circular chromatic number, the chromatic 1 number, andthebandwidth. Wealsoexaminetheconnectionbetweenboundedresolution : v coefficient and minor-closed graph classes. i X r 1 Introduction a Givenasimple, undirected, finitegraphG = (V,E), arepresentationofGinthed-dimensional Euclidean space Rd is a function ρ assigning to each vertex of G a point in Rd. An R2- representation is called planar, an R3-representation is called spatial. A representation of a graphinRd canbeviewedasadrawingofthegraphwithstraightedges. Thelengthofanedge uv E is the Euclidean distance between ρ(u) and ρ(v), i.e., ρ(u) ρ(v) . A representation 2 ∈ k − k ρ is non-vertex-degenerate (NVD) if the function ρ is one-to-one. A representation ρ is non- edge-degenerate (NED) if for all edges uv E(G), it holds that ρ(u) = ρ(v); in other words, ∈ 6 if the minimal edge length is positive. Clearly, a non-vertex-degenerate representation is also non-edge-degenerate. 1 Inthispaperwestudythreegraphinvariantsthatarerelevantforrepresentationsofgraphs. We restrict ourselves to planar representations. To avoid trivialities, we only consider graphs with at least one edge. Following [12], we define the dilation coefficient of a non-edge-degenerate representation of a graph as the ratio between the longest and the shortest edge length in the representation. The minimum of the set of dilation coefficients of all non-edge-degenerate representations of a graph G is called the dilation coefficient of G, and is denoted by dc(G). The plane-width of a graph G = (V,E), introduced in [10] and denoted by pwd(G), is the minimum diameter of the image of the graph’s vertex set, over all representations ρ with the property that d(ρ(u),ρ(v)) 1 for each edge uv E, where d is the Euclidean distance. ≥ ∈ Equivalently, the plane-width of a graph can be defined as the minimum, over all non-edge- degenerate representations, of the ratio between the largest distance between two points and the shortest length of an edge. We define the resolution coefficient of a non-vertex-degenerate representation of a graph as the ratio between the longest length of an edge and the smallest distance between the images of any two distinct vertices. The minimum of the set of resolution coefficients of all representations of a graph G is called the resolution coefficient of G, and is denoted by re(G). In formulae, denoting by V (G) the set of all 2-element subsets of V(G), we have 2 max d(ρ(u),ρ(v)) uv E(G) dc(G) = min ∈ : ρ is an NED representation of G , (1) min d(ρ(u),ρ(v)) (cid:26) uv E(G) (cid:27) ∈ max d(ρ(u),ρ(v)) pwd(G) = min uv∈V2(G) : ρ is an NED representation of G , (2) min d(ρ(u),ρ(v)) (cid:26) uv E(G) (cid:27) ∈ max d(ρ(u),ρ(v)) uv E(G) re(G) = min ∈ : ρ is an NVD representation of G . (3) min d(ρ(u),ρ(v)) (cid:26) uv V2(G) (cid:27) ∈ Note that the above parameters are well-defined, as we only consider graphs with at least one edge, and each of the three parameters can be expressed as the minimum value of a continuous real-valued function over a compact subset of R2V(G). | | Noticealsothatconsideringthesmallestratiobetweenthelargestandthesmallestdistance between any two distinct points in a non-vertex-degenerate representation does not lead to a meaningful graph parameter – indeed, this quantity depends only on the number of vertices of the graph and not on the graph itself. The problem of determining the minimum value of this parameter for a given number of points n has previously appeared in the literature in different contexts such as finding the minimum diameter of a set of n points in the plane such that each pair of points is at distance at least one [3], or packing non-overlapping unit discs in the plane so as to minimize the maximum distance between any two disc centers [13]. For a given n, we denote the value of this parameter by h(n): max d(ρ(u),ρ(v)) h(n) = min uv∈V2(Kn) : ρ is an NVD representation of Kn . min d(ρ(u),ρ(v)) (cid:26) uv V2(Kn) (cid:27) ∈ 2 n 2 3 4 5 6 7 8 h(n) 1 1 √2 1+√5 2sin72 2 (2sin(π/14)) 1 2 ◦ − approx. 1 1 1.414 1.618 1.902 2 2.246 Table 1: Known values of h(n) = dc(K ) = pwd(K ) = re(K ). n n n The exact values for h(n) are known only up to n = 8 [1, 3], see Table 1. It follows directly from the definitions that if G is a complete graph, the three parameters dc, pwd and re coincide. Proposition 1. For every complete graph K , n dc(K ) = pwd(K ) = re(K ) = h(n). n n n The parameter h(n) was recently studied by Horvat, Pisanski and Zˇitnik [8]. In this paper we give some relationships between the three parameters. After giving some basic properties and examples in Section 3, we show in Section 4 that the dilation coefficient andtheplane-widthareequivalentgraphparametersinthesensethattheyareboundedonthe same sets of graphs. These two parameters are also equivalent to the chromatic number, and are therefore bounded from above by a function of the maximum degree. On the other hand, there exist cubic graphs with arbitrarily large resolution coefficient. In particular, while the plane-widthandthedilationcoefficientareboundedfromabovebyafunctionoftheresolution coefficient, the converse is not true. In Section 5 we examine the natural one-dimensional analogues of these three parameters (denoted by dc (G), pwd (G) and re (G)), and show that they coincide or almost coincide 1 1 1 withthreewellstudiedgraphparameters: thecircularchromaticnumberχ (G), thechromatic c number χ(G), and the bandwidth bw(G). As a corollary of some relations among dc (G), 1 pwd (G)andre (G),weobtainindependentproofsofsomeofthewell-knownrelationsbetween 1 1 χ (G), χ(G) and bw(G). c In Section 6 we show that G is planar whenever re(G) < √2, while there exist graphs with re(G) = √2 that contain arbitrarily large clique minors. 2 Preliminaries In this section we review some definitions and basic results. For terms left undefined, we refer the reader to [6]. As usual, let χ(G) denote the chromatic number, i.e., the least number of colors needed for a proper vertex coloring of G, and ω(G) the clique number, i.e., the maximum size of a complete subgraph of G. Clearly, ω(G) χ(G) for any graph G. Graphs ≤ G such that for all induced subgraphs H of G the equality ω(H) = χ(H) holds are called perfect. Recently perfect graphs have been characterized by Chudnovsky, Robertson, Seymour and Thomas [4], who proved the famous Strong Perfect Graph Theorem conjectured by Berge in 1961 [2]. Some useful connections between the plane-width and the chromatic number of a graph, for small values of these two parameters, are given in [10]: 3 Theorem 2. For all graphs G, (a) pwd(G) = 1 if and only if χ(G) 3, ≤ (b) pwd(G) (2/√3,√2] if and only if χ(G) = 4, ∈ (c) pwd(K ) = √2. 4 The following property of the plane-width will also be used in some of our proofs. Proposition 3 ([10]). Let G be a graph such that χ(G) = ω(G). Then, pwd(G) = h(χ(G)). Obviously, the statement of the above proposition holds for perfect graphs. 3 Basic properties and examples The dilation coefficient, the plane-width and the resolution coefficient of a graph are related by the following inequalities. Proposition 4. For every graph G, 1 dc(G) min pwd(G),re(G) . ≤ ≤ { } Proof. The inequality dc(G) 1 is clear. ≥ Let ρ be a non-edge-degenerate representation of G. Then, since E(G) V (G), 2 ⊆ max d(ρ(u),ρ(v)) max d(ρ(u),ρ(v)) uv∈E(G) uv∈V2(G) . min d(ρ(u),ρ(v)) ≤ min d(ρ(u),ρ(v)) uv E(G) uv E(G) ∈ ∈ Taking the minimum over all such representations, it follows that dc(G) pwd(G). ≤ To see that dc(G) re(G), let ρ be a non-vertex-degenerate representation of G that ≤ achieves the minimum in the definition of re(G). Then, since ρ is non-edge-degenerate and E(G) V (G), we have 2 ⊆ max d(ρ(u),ρ(v)) max d(ρ(u),ρ(v)) uv E(G) uv E(G) dc(G) ∈ ∈ = re(G). ≤ min d(ρ(u),ρ(v)) ≤ min d(ρ(u),ρ(v)) uv E(G) uv V2(G) ∈ ∈ Clearly, the inequalities from Proposition 4 are tight. For instance, dc(G) = pwd(G) = re(G) = 1 for every graph with at most three vertices and at least one edge. Moreover, the dilation coefficient of a graph G is equal to 1 if and only if G admits a representation with all edges of the same length. Such graphs are known under the name of unit distance graphs. There is a strong chemical motivation for exploring unit distance graphs and related concepts, since relevant chemical graphs tend to have all bond lengths of almost the same size. By Theorem 2, graphs of plane-width 1 are precisely the 3-colorable graphs. To the best of our knowledge, graphs of unit resolution coefficient have not previously appeared in the literature. Thefollowingtwoexamples,basedonunitdistancegraphs,showthatdc(G)canbestrictly smaller than min pwd(G),re(G) , and that re(G) and pwd(G) are incomparable. { } 4 Example 1. Let G be the Moser spindle, see Fig. 1. This is a 4-chromatic unit distance graph on 7 vertices. Since G is unit distance, we have dc(G) = 1. On the other hand, since G is 4-chromatic, pwd(G) > 2/√3 by Theorem 2. Furthermore, it is easy to see that G cannot be drawn in the plane with all edges of unit length and all the other pairs of points at least unit distance apart. Therefore, re(G) > 1. Figure 1: A unit-distance representation of the Moser spindle Example 2. Let G be the graph depicted on Fig. 2. Figure 2: A graph with re(G) < pwd(G) (all edge lengths are the same) Observe that G is 4-chromatic, thus pwd(G) > 2/√3 by Theorem 2. The drawing from Fig. 2 gives a representation ρ of G such that all edges are of unit length, and all the other pairs of points are at least unit distance apart. Therefore, max d(ρ(u),ρ(v)) = uv E(G) ∈ min d(ρ(u),ρ(v)), which implies that re(G) = 1. This shows that in general, pwd(G) uv V2(G) ∈ is not bounded from above by re(G). To see that also re(G) is not bounded from above by pwd(G), let G be the 4-wheel, that is, the graph obtained from a 4-cycle by adding to it a dominating vertex. Then, since G is 3-colorable, pwd(G) = 1 by Theorem 2. However, it is easy to see that G cannot be drawn in the plane with all edges of the same length, which implies that re(G) > 1. In Section 4 we will determine which of these three parameters is bounded from above by a function of another one. A homomorphism of a graph G to a graph H is an adjacency-preserving mapping, that is a mapping φ : V(G) V(H) such that φ(u)φ(v) E(H) whenever uv E(G). We say → ∈ ∈ that a graph G is homomorphic to a graph H if there exists a homomorphism of G to H. A graph invariant f is homomorphism monotone if f(G) f(H) whenever G is homomorphic ≤ 5 to H. In [10], it was shown that the plane-width is homomorphism monotone. We now show that the same property holds for the dilation coefficient, and that the resolution coefficient is monotone with respect to the subgraph relation. Proposition 5. Let G and H be graphs with at least one edge. (i) If G is homomorphic to H then dc(G) dc(H). ≤ (ii) If G is a subgraph of H then re(G) re(H). ≤ Proof. It follows from definitions that dc(G) is the minimum value of p such that G is homo- morphic to some graph H whose vertex set is a subset of R2 and such that every edge of H connects two points at Euclidean distance at least 1 and at most p. This observation together with the transitivity of the homomorphism relation immediately implies (i). Tosee(ii),letρbeanon-vertex-degeneraterepresentationofH thatachievestheminimum in the definition of re(H). Then, since the restriction of ρ to V(G) is a non-vertex-degenerate representation of G, E(G) E(H) and V (G) V (H), we have 2 2 ⊆ ⊆ max d(ρ(u),ρ(v)) max d(ρ(u),ρ(v)) uv E(G) uv E(H) re(G) ∈ ∈ = re(H). ≤ min d(ρ(u),ρ(v)) ≤ min d(ρ(u),ρ(v)) uv V2(G) uv V2(H) ∈ ∈ 4 More on relationship between dc, pwd, and re In this section, we examine more closely the relations between these three parameters. First, we show in Theorem 7 that the plane-width of a graph is bounded from above by a function of its dilation coefficient. Our proof will make use of the following result from [10] (combining Lemmas 2.2 and 3.7 therein). Lemma 6. ([10]) There exists a constant C > 0 such that for every graph G, 2√3 pwd(G) χ(G)+C. ≤ s π Theorem 7. There exist positive constants K,C > 0 such that for every graph G, pwd(G) K dc(G)+C. ≤ · Proof. Let ρ be a representation of G achieving the minimum in the definition of the di- lation coefficient (cf. Equation (1)). We may assume, without loss of generality, that min d(ρ(u),ρ(v)) = 1(otherwise,wescaletherepresentationaccordingly). Hencedc(G) uv E(G) ∈ equals the maximum length of an edge of G w.r.t. the representation ρ. Let us cover the set ρ(v) : v V(G) with pairwise disjoint translates of the half-open { ∈ } square S = [0,t/√2) [0,t/√2), arranged in a grid-like way, where t = √2dc(G) + 1. × ⌈ ⌉ Furthermore, we partition each copy S of S into t2 pairwise disjoint translates A (S ) of ′ ij ′ the set [0,1/√2) [0,1/√2), for each i,j 1,...,t . Here, the pair (i,j) denotes the × ∈ { } 6 “coordinates” of the square A (S ) within S . We do the assignment of the coordinate pairs ij ′ ′ to the small squares in the same way for all copies S of S. ′ Let c : V(G) 1,...,t 2 be a coloring of the vertices of G that assigns to each v V(G) → { } ∈ theunique(i,j) 1,...,t 2 suchthatthereexistsatranslateS ofS suchthatρ(v) A (S ). ′ ij ′ ∈ { } ∈ By construction, c is a proper coloring of G. Therefore, χ(G) t2. By Lemma 6, pwd(G) ≤ ≤ 2√3χ(G)+C 2√3t+C. Combiningthisinequalitywiththeinequalityt √2dc(G)+2, π ≤ π ≤ qthe proposition folqlows: we may take K = 4√3 1.4850.1 π ≈ q We say that two graph parameters f and g are equivalent if they are bounded on precisely the same sets of graphs, that is, if for every set of graphs , we have G sup f(G) : G < { ∈ G} ∞ if and only if sup g(G) : G < . { ∈ G} ∞ It follows from Proposition 4 and Theorem 7 that the dilation coefficient and the plane-width are equivalent graph parameters, which, according to the following result from [10], are also equivalent to the chromatic number χ: Theorem 8 ([10]). For every ǫ > 0 there exists an integer k such that for all graphs G of chromatic number at least k, √3 2√3 ǫ χ(G) < pwd(G) < +ǫ χ(G). 2 − s π  ! p p   On the other hand, it turns out that the resolution coefficient is not equivalent to the dilation coefficient or to the plane-width. Theorem9. Foreverypositivefunctionf thereexistsagraphGsuchthatre(G) > f(pwd(G)). There exists a function f such that for every graph G, pwd(G) f(re(G)). ≤ Proof. TakeG = K . Sincethesegraphsarebipartite, pwd(K ) = 1foralln. However, the 1,n 1,n valuesofre(K )tendtoinfinitywithincreasingn. Thisfollowsbyobservingthatthereexists 1,n a constant C > 0 such that for every N > 0, in any non-vertex-degenerate representation ρ of K with min d(ρ(u),ρ(v)) = 1, less than CN2 vertices can be mapped to distance at 1,n uv V2(G) ∈ most N from the image of the center of the star. Thus, re(K ) = Ω(√n). This shows the 1,n first part of the theorem. GivenagraphGwithatleastoneedge, let∆(G)denoteitsmaximumvertexdegree. Since K is a subgraph of G, Proposition 5 implies that re(G) re(K ) = Ω( ∆(G)). 1,∆(G) 1,∆(G) ≥ Therefore,themaximumdegreeofagraphisboundedfromabovebyafunctionofitsresolution p coefficient. In particular, since χ(G) ∆(G)+1, the chromatic number of G is also bounded ≤ from above by a function of re(G), and the second part of the theorem follows by Theorem 2. 1AbetterconstantK =q83√π3+ǫ≤1.2126canbeobtainedbycoveringtheplanewithhexagonsinsteadof the squares (similarly as was done in Lemma 3.6 in [10]). 7 In the above proof, large degree caused the resolution coefficient to be large. The next example shows that large maximum degree is only a sufficient but not a necessary condition for large resolution coefficient. Therefore, while the chromatic number of a graph is bounded from above by a function of its maximum vertex degree, and then the same is true for the dilation coefficient and the plane-width, this is not the case for the resolution coefficient. Theorem 10. For each R > 0 there exists a graph G of maximum degree at most 3 such R that re(G ) > R. R Proof. Let H denote a full cubic tree with k layers, rooted at vertex v . Let V denote the k 0 i number of vertices of layer i. Then V = 1, V = 3, V = 3 2, ..., V = 3 2k 1 and 0 1 2 k − · · V(H ) = V +V + +V = 3 2k 2. Let ρ be a representation of H , where re(H ) is k 0 1 k k k | | ··· · − achieved. Let r = min d(ρ(u),ρ(v)) . There exists at least one vertex x of V such uv∈V2(Hk){ } k that d(ρ(v ),ρ(x)) > C r√3 2k 2 =: R. Since the graph distance d(V ,x) = k there exists 0 ′ 0 · · − an edge pq on the path from v to x such that 0 C r√3 2k 2 d(ρ(p),ρ(q)) R′/k = · · − . ≥ k Therefore d(ρ(p),ρ(q)) C √3 2k 2 ρ(H ) · · − . k ≥ r ≥ k For a given R define k to be the smallest k for which C (3 2k 2) R and define 0 · · − ≥ G = H . Clearly, G has maximum degree at most 3 and re(G ) > R. R k0 R R 5 One-dimensional restrictions It is interesting to consider the restrictions of these three parameters to a line instead of the plane. That is, every vertex gets mapped to a point on the real line instead of being mapped to a point in R2 and the definitions are analogous to those given by Equations (1)-(3). We denote the corresponding parameters by dc (G), pwd (G) and re (G). 1 1 1 It turns out that the one-dimensional variant of the dilation coefficient is strongly related to the circular chromatic number of the graph, the “line-width” is strongly related to the chromatic number of the graph, while the one-dimensional variant of the resolution coefficient coincides with the graph’s bandwidth, defined as bw(G) := min max π(u) π(v) , π:V 1,...,n ,bij.uv E| − | →{ } ∈ where n = V(G) . | | Given a graph G = (V,E) and c 1, a circular c-coloring is a mapping f : V [0,c) such ≥ → that for every edge uv E, it holds that 1 f(u) f(v) c 1. The circular chromatic ∈ ≤ | − | ≤ − number χ (G) of G is defined as c χ (G) = inf c : G admits a circular c-coloring . c { } It is known that the infimum in the definition above is always attained (see, e.g., [15]). 8 Theorem 11. For every graph G, the following holds: (i) dc (G) = χ (G) 1 1 c − (ii) pwd (G) = χ(G) 1. 1 − (iii) re (G) = bw(G). 1 Proof. Let G = (V,E) be a graph. (i) Let c = χ (G) and let f be a circular c-coloring of G. Then, f defines a non-edge- c degenerate one-dimensional representation of G such that min f(u) f(v) 1 and uv E χma(xGu)v∈E1|.f(u)−f(v)| ≤ c−1. It follows that mmainxuuvv∈∈EE||ff((uu))−−ff((vv))|| ≤∈c−| 1, im−plying| d≥c1(G) ≤ c − Conversely, let ρ : V R be a non-edge-degenerate one-dimensional representation of G ∗ → that achieves the minimum in the definition of dc (G). Without loss of generality, we may 1 assume that min ρ (u) ρ (v) = 1 and that min ρ (v) = 0. Let c = dc (G)+1, and uv E ∗ ∗ v V ∗ 1 define f : V R a∈s fo|llows:− | ∈ → For all v V, ∈ ρ (v) ∗ f(v) = ρ (v) c . ∗ − · c (cid:22) (cid:23) To show that χ (G) dc (G)+1 = c, it suffices to verify that f is a circular c-coloring of G. c 1 ≤ Since 0 ρ∗(v) ρ∗(v) < 1, f maps vertices of G to the interval [0,c). It remains to show ≤ c − c that for every edgejuv kE, it holds that 1 f(u) f(v) c 1. Notice that by the choice ∈ ≤ | − | ≤ − of ρ , we have 1 ρ (u) ρ (v) c 1 for every edge uv E. ∗ ∗ ∗ ≤ | − | ≤ − ∈ Let uv E. We may assume that ρ (u) ρ (v). Let k = ρ∗(u) and ℓ = ρ∗(v) . The ∈ ∗ ≤ ∗ c c inequality ρ∗(u) ρ∗(v) implies that k ℓ. Moreover, the inejqualityk ρ∗(v) ρj∗(u)+k c 1 ≤ ≤ ≤ − implies ℓ k+1. ≤ If ℓ = k, then f(u) f(v) = ρ (u) ρ (v) and the inequalities 1 f(u) f(v) c 1 ∗ ∗ − − ≤ | − | ≤ − follow. Suppose now that ℓ = k+1. Then f(u) f(v) = ρ (u) ρ (v)+c. Therefore, it follows ∗ ∗ − − from ρ (v) ρ (u) c 1 that f(u) f(v) 1, implying f(u) f(v) = f(u) f(v) 1. ∗ ∗ − ≤ − − ≥ − | − | ≥ Similarly, it follows from ρ (v) ρ (u) 1 that f(u) f(v) = f(u) f(v) c 1. ∗ ∗ − ≥ − | − | ≤ − This shows that f is a circular c-coloring of G, which implies that χ (G) dc (G)+1. c 1 ≤ (ii)Anyk-coloringofGwithcolorsintheset 1,...,k Rdefinesanon-edge-degenerate { } ⊆ one-dimensional representation ρ of G such that maxuv∈V2(G)|ρ(u)−ρ(v)| k 1. Therefore, minuv E ρ(u) ρ(v) ≤ − pwd (G) χ(G) 1. ∈ | − | 1 ≤ − Conversely, let ρ : V R be a non-edge-degenerate one-dimensional representation of ∗ → G that achieves the minimum in the definition of pwd (G). Without loss of generality, we 1 may assume that min ρ (u) ρ (v) = 1 and that min ρ (v) = 1. Define f : V R uv E ∗ ∗ v V ∗ ∈ | − | ∈ → as follows: For all v V, let f(v) = ρ (v) . Then, f is a proper k-coloring of G, where ∗ ∈ ⌊ ⌋ k = pwd (G) +1. Therefore, χ(G) pwd (G) +1 pwd (G)+1. ⌊ 1 ⌋ ≤ ⌊ 1 ⌋ ≤ 1 (iii) Any bijective mapping π : V 1,...,n R defines a non-vertex-degenerate → { } ⊆ one-dimensional representation of G such that max π(u) π(v) uv E ∈ | − | = max π(u) π(v) . minuv∈V2(G)|π(u)−π(v)| uv∈E| − | 9 Therefore, re (G) bw(G). 1 ≤ Conversely, let ρ : V R be a non-vertex-degenerate one-dimensional representation of ∗ → G that achieves the minimum in the definition of re (G). Without loss of generality, we may 1 assume that min ρ (u) ρ (v) = 1. Then uv∈V2(G)| ∗ − ∗ | max ρ (u) ρ (v) uv E ∗ ∗ re1(G) = ∈ | − | = max ρ∗(u) ρ∗(v) . minuv∈V2(G)|ρ∗(u)−ρ∗(v)| uv∈E| − | Therefore, to show that bw(G) re (G), it suffices to show that bw(G) max ρ (u) 1 uv E ∗ ≤ ≤ ∈ | − ρ (v) . ∗ | Order the vertices of V = v ,...,v according to the increasing values of their images: 1 n { } ρ (v ) < ρ (v ) < < ρ (v ), ∗ 1 ∗ 2 ∗ n ··· and let π(v ) = i for all i 1,...,n . This defines a bijective mapping π : V 1,...,n . i ∈ { } → { } Let uv E, and assume that i := π(u) < j := π(v). By the definition of π and since ∈ min ρ (x) ρ (y) = 1,weinferthat ρ (u) ρ (v) j i. Ontheotherhand, π(u) xy∈V2(G)| ∗ − ∗ | | ∗ − ∗ | ≥ − | − π(v) = j i, which implies that π(u) π(v) ρ (u) ρ (v) . Therefore, max π(u) ∗ ∗ uv E | − | − | ≤ | − | ∈ | − π(v) max ρ (u) ρ (v) , implying bw(G) max π(u) π(v) max ρ (u) uv E ∗ ∗ uv E uv E ∗ | ≤ ∈ | − | ≤ ∈ | − | ≤ ∈ | − ρ (v) = re (G). ∗ 1 | Proposition 12. For every graph G, dc (G) = pwd (G) re (G). ⌈ 1 ⌉ 1 ≤ 1 Proof. In the same way as the inequality dc(G) pwd(G) from Proposition 4, one can prove ≤ the inequality dc (G) pwd (G). Since pwd (G) = χ(G) 1 by Theorem 11, pwd (G) is 1 ≤ 1 1 − 1 always integral, therefore dc (G) pwd (G). ⌈ 1 ⌉ ≤ 1 To see that pwd (G) dc (G) , consider a non-edge-degenerate one-dimensional repre- 1 ≤ ⌈ 1 ⌉ sentation ρ : V R of G that achieves the minimum in the definition of dc (G). Without ∗ 1 → loss of generality, we may assume that min ρ (u) ρ (v) = 1 and that min ρ (v) = 0. uv E ∗ ∗ v V ∗ Let k := dc (G) +1, and define ρ : V ∈0,1|,...,k− 1 | R as follows: For ∈all v V, let 1 ′ ⌈ ⌉ → { − } ⊆ ∈ ρ(v) = ρ (v) (mod k). ′ ∗ ⌊ ⌋ For every uv E, we have ρ(u) ρ(v) 1. Indeed: if ρ(u) = ρ(v) then ′ ′ ′ ′ ∈ | − | ≥ ρ (u) ρ (v) k > dc (G), contrary to the choice of ρ and the definition of k. Therefore, ∗ ∗ 1 ∗ | − | ≥ ρ is a non-edge-degenerate one-dimensional representation of G, and since the image of ρ is ′ ′ contained in the set 0,1,...,k 1 , we have max ρ(u) ρ(v) k 1 = dc (G) , { − } uv∈V2(G)| ′ − ′ | ≤ − ⌈ 1 ⌉ implying that pwd (G) dc (G) . 1 ≤ ⌈ 1 ⌉ In the same way as the inequality dc(G) re(G) from Proposition 4, one can prove the ≤ inequality dc (G) re (G). Since re (G) = bw(G) by Theorem 11, re (G) is always integral, 1 1 1 1 ≤ and consequently dc (G) re (G). 1 1 ⌈ ⌉ ≤ Theorem 11 and Proposition 12 provide an alternative proof of the following well known relations: The equality χ(G) = χ (G) showing that the circular chromatic number is the refine- c • ⌈ ⌉ ment of the chromatic number [14]. 10