Definition. Let G = (V, E) be a graph. A function c : E → N is a proper edge coloring of G, if for every xy, xz ∈ E we have c(xy) 6= c(xz), that is, if two edges have a common endpoint, then their colors must be different. Notation:
χ0(G) = min{k :∃c:E →[k] proper edge coloring},
so this is the smallest number of colors necessary for properly coloring the edges of G.We call χ0(G) the chromatic index or edge-chromatic number of G.
It is easy to see that ∆(G)≤χ0(G),ifG is a simple graph. There are graphs for which the chromatic index is larger than the maximum degree. For example, let G be any cycle with odd length. Then ∆(G) = 2,but, as is easily seen, χ0(G) = 3.
Similarly to vertex coloring, ifGhas a loop, then it does not allow a proper edge coloring. Having parallel edges, on the other hand, makes sense, unlike in vertex coloring. Soon we will talk more about this.
Lemma 7.10. We have that χ0(G)≤2∆(G)−1.
Proof. First construct the line graphL(G) of G.Recall, that the vertex set ofL(G) isE(G),and two vertices,e1 ande2 are adjacent inL(G),if the edges e1 ande2 have at least one common endpoint inG.Clearly, ∆(L(G))≤2(∆(G)−1),and applying the greedy vertex coloring we end up using at most ∆(L(G)) + 1 = 2∆(G)− 1 colors.
In case G is bipartite, we can say more.
Theorem 7.11 (D´enes K¨onig (1916)). If G = (A, B, E) is a bipartite graph, then χ0(G) = ∆(G).
Proof. The theorem follows from the fact that if G is an r-regular bipartite graph for some positive integer r, then it has a perfect matching. This in turn is an easy consequence of the K¨onig-Hall theorem for the existence of perfect matchings in bipartite graphs, we leave it as an exercise.
Now if G was r-regular, find a perfect matching M1 in it, and then delete its edges from G. The resulting graph G−M1 is (r−1)-regular, hence, if r−1 ≥ 1, then G−M1 has a perfect matching M2. We may continue this way, and stop only when there is no edge left. At this point we have the perfect matchingsM1, . . . , Mr. Clearly, if e, e0 ∈ Mi for some 1 ≤ i ≤ r, then they may get the same edge color without creating a conflict. Hence, if G is an r-regular bipartite graph, then its edges can be properly colored byr colors. This must be an optimal coloring, as less colors are not sufficient.
Finally assume that G is not r-regular for r = ∆(G). If G is not balanced, say,
|A|>|B|,then we add|A| − |B|new vertices toB to make the graph balanced. We are going to add edges to the graph to make it regular. Assume that both x ∈ A and y∈ B have degrees less than r. Then we add the edge xy to the graph – note, that this way we might create multiple edges betweenxand y.We repeat the above procedure, and when we finish, every vertex will have degree r, so the K¨onig-Hall theorem can be applied for finding a perfect matching.
Let us remark, that if originallyGwas a simple graph, then we may just create a simpler-regular graph from it as well. This goes as follows. Ifx, ybelong to different vertex classes, are non-adjacent, and both have degrees less thanr, then we add the edge xy toG. This is repeated until it is just possible. When we stop and find two vertices from different vertex classes that both have small degrees, then they must
be adjacent. For such an x, y pair of vertices we add a gadget to the graph. This gadget is aKr,r.We delete one of the edges of the gadget having endpoints uandv.
Say,u and x belong to the same vertex class. Then we add the edges uy and vx to the graph. With this we increased the degrees of x and y, and every vertex in the gadget still have degreer.It is easy to see that repeating this procedure results in a possibly much larger graphG0, but what is important for us,G0 isr-regular, hence, its edges can be properly colored byr = ∆(G0) = ∆(G) colors.
In general it is possible that χ0(G) > ∆(G), but surprisingly, the difference of the two is always small. The result below was discovered first by Vizing, then inde-pendently, by Gupta. It is usually referred to as Vizing’s theorem in the literature.
Theorem 7.12 (Vizing (1964), Gupta (1966)). Let G be a simple graph. Then
∆(G)≤χ0(G)≤∆(G) + 1.
We do not prove Vizing’s theorem, but mention an interesting fact: while χ0(G) may take on only two different values, it is NP-hard to decide, if ∆(G) or ∆(G) + 1 is the edge-chromatic number of G.
The following is a generalization of Vizing’s theorem for graphs having parallel edges, it was proved by the same authors. We introduce a new notion. We letµ(x, y) denote the multiplicity of thex, y pair (this is the number of edges connectingxand y), and let µ(G) denote the maximum of the multiplicities of edges of G.
Theorem 7.13 (Vizing (1964), Gupta (1966)). Let G be a graph. Then χ0(G) ≤
∆(G) +µ(G).
Sinceµ(G) = 1 ifGis a simple graph, the above generalizes Theorem 7.12. This also shows that the theorem is sharp.
There is another bound for the chromatic index of graphs by Shannon.
Theorem 7.14 (Shannon (1949)). If G is a graph, then χ0(G)≤ 32∆(G).
We will prove a slightly weaker result, also by Shannon, which uses a theorem of Petersen. We need a definition before stating this it: a 2-factor of a graph is a spanning subgraph of it in which every vertex has degree 2. It is easy to see that a 2-factors is a collection of disjoint cycles.
Theorem 7.15(Petersen (1890)). Assume thatG is a 2k-regular graph, wherek is a positive integer. Then G can be decomposed into k edge-disjoint 2-factors.
Proof. Since every degree inGis even, we have an Eulerian circuit in it. Traversing this Eulerian circuit gives a natural orientation to every edge. Since we leave and enter every vertexktimes, the in-degree and out-degree of every vertex will bek.Let us construct an auxiliary bipartite graph H= (V1, V2, E(H)). Wheneveru∈V(G), we have two copies of u in H, u1 ∈V1 and u2 ∈ V2. If an edge vw of G is oriented fromv towards w, then we have the edge v1w2 in E(H). It is easy to see that H is ak-regular bipartite graph.
Hence, by Theorem 7.11 we have a decomposition ofE(H) intokperfect match-ings,M1, . . . , Mk.In everyMi,every vertex ofGappears exactly twice, once in both copies of V. Therefore, for every i the edges of Mi induce a spanning subgraph in which every vertex ofG has degree exactly 2.
The weaker version of Theorem 7.14 is as follows.
Theorem 7.16(Shannon (1949)). Let Gbe a graph. If∆(G) is even, thenχ0(G)≤
3
2∆(G). If ∆(G) is odd, then χ0(G)≤ 32(∆(G) + 1).
Proof. We begin with a preprocessing ofG in case ∆(G) = 2k−1 for some positive integer k.Similarly to the proof of 7.11 we first turn Ginto a 2k−1-regular graph by adding edges and vertices toGbetween points that have degree less than 2k−1.
Next we add an arbitrary perfect matching to the graph. After this preparation we obtain a 2k-regular graph G0.
Now we are ready to use Theorem 7.15. We can find thek edge-disjoint 2-factors F1, . . . , Fk. Clearly, the edges of a 2-factor can be colored by at most 3 colors – if every cycle has even length, 2 colors are sufficient, otherwise we need 3. In total we need at most 3·k colors. This proves what was desired.
The theorem of Shannon is tight, as the following example shows. Let u, v and w be the vertices of a graph, and assume that between any two of the vertices we have 3 parallel edges. So this graph is a “multitriangle”.
We remark, that depending on the graph, either the boundχ0(G)≤∆(G)+µ(G) by Vizing and Gupta or the one χ0(G)≤ 32∆(G) by Shannon is sharper.
Chapter 8
Planar drawings
8.1 Planar multigraphs
Definition. A drawing of a multigraph G is a pair (ρ, γ) where ρ: V(G) → R2 is an injective function that maps each vertex v ∈V(G) to a point ρ(v) in the plane, andγ is a function that maps each edge e=uv ∈E(G) to a continuous plane curve γe between ρ(u) and ρ(v), such that γe does not contain ρ(w) as an interior point for any w ∈ V(G). We refer to the points ρ(v) as the points of the drawing of G, and the curves γe are referred as edge curves.
An edge crossing in a drawing is a point on the plane which is contained in two (or more) different edge curvesγe, γf as interior points. A multigraphGisplanar, if it has a drawing without edge crossings. Such a drawing is called a planar drawing (or planar embedding) ofG.
Example. For example, the complete graphK4 is planar, as justified by the second and third drawings in Figure 8.1.
Figure 8.1: A non-planar and two planar drawings of K4
As an other example, we note that all trees are planar. This follows from the structure theorem of trees (Theorem 1.10) and the observation that a pendant edge can be always added to a planar drawing without introducing edge crossings.
Polyhedral graphs are also planar. (A polyhedral graph is a graph formed from the vertices and edges of a 3-dimensional convex polyhedron.) For example, a planar drawing of the cube is shown in Figure 8.2; and K4, our first example of planar graphs, is also a polyhedral graph (of a tetrahedron).
Figure 8.2: The cube graph