11.4 Edge-Chromatic number

Definition 11.12.

A (proper) k-edge colouring of a graph G is a function c:E[k] such that c9e)c(e) for ee. Define the edge-chromatic number or chromatic index as χ(G):=min{k:G has a k-edge colouring}.

Recall the definition of line graph L(G) of G. It follows from definition that χ(G)=χ(L(G)). Since L(G) has a clique of order k if there exists a vertex of degree k, from the trivial bound for chromatic number we have that Δ(G)χ(G). One can check that Δ(L(G))2Δ(G)2 and so

Δ(G)χ(G)2Δ(G)2.

The above bound was considerably tightened leading to very rigid behaviour for edge-chromatic numbers.

Theorem 11.13 (Vinzing, 1964).

Δ(G)χ(G)Δ(G)+1.

It is non-trivial to determine whether it is Δ(G) or Δ(G)+1. See [Vinzing Theorem] for more on this theorem, variants and applications. We first state a lemma and prove the theorem. We defer the proof of the Lemma to later.

Lemma 11.14.

Let v be a vertex such that v and its neighbours have at most degree k with exactly one neighbour having degree k. Then if Gv is k-edge colourable so is G.

Proof of Theorem 11.13.

We prove the upper bound by induction. The theorem holds trivially for G=K2. Without loss of generality assume that G is connected and G has strictly more than 2 vertices. Assume that the theorem holds for any graph with fewer vertices or edges.

Let u be the vertex with maximum degree i.e., deg(u)=Δ(G). Let vu such that deg(v)2. If no such v exists then, G is a star graph and χ(G)=Δ(G) proving the theorem. Add an additional vertex u and connect it to u. Call this new graph G. Observe that in G, v satisfies the conditions of Lemma 11.14 with k=Δ(G)+1. Furthermore Gv has fewer edges than G and so χ(Gv)Δ(Gv)+1=Δ(G)+1. Thus by Lemma 11.14, we have that G is (Δ(G)+1)-edge colourable and so is G. ∎

An alternate proof of the theorem using edge-deletions but nevertheless using some of the ideas in proof of the Lemma 11.14 can be found in [Vinzing Theorem].

Proof of Lemma 11.14.

We shall give a sketch of proof. For full details, see [Sudakov 2019, Chapter 8].

The proof is by induction on k. Let k=1. Then v has exactly one neighbour u and (u,v) is an isolated edge. If Gv is 1-edge colourable, then easily we can colour the isolated edge (u,v) by the same colour and get a 1-edge colouring of G.

Let k>1. By adding additional vertices if necessary, we can assume without loss of generality that deg(v)=k and there exists wv such that deg(w)=k and deg(u)=k1 for all uv,uw.

Let c be a k-edge colouring of Gv. Let Xi be the number of neighbours of v missed by colour i. One neighbour misses only one colour and rest of the neighbours miss 2 colours. So by double counting, we have that

i=1k|Xi|=2(k1)+1.

Among all k-edge colourings choose one that minimizes i=1k|Xi|2. Suppose it holds for such a colouring that for all i,j

||Xi||Xj||2. (11.1)

Then since i=1k|Xi| is odd, there exists |Xi|=1. Without loss of generality, we can assume that |Xk|=1. Set Xk={u}. Let G be the graph obtained from G by deleting (u,v) and all edges of colour k. So Gv has a (k1)-edge colouring. By induction hypothesis, we can verify that G has a (k1)-edge colouring. Now putting back the edges of colour k as well as colouring (u,v) with colour k, we get a k-colouring of v. Note that other neighbours of v have other missing colours that can be used.

We are left to prove (11.1). WLOG, let i=1,j=2 and assume |X1|>|X2|+2. Define H to be the sub-graph formed by edges of colour 1 or 2. Observe that H consists of path or even cycle components. Since internal vertices in a path component and all vertices in a cycle component are missed by both the colours, the difference between |X1| and |X2| arises from a path component, say P. P starts and ends with a vertex in X1. Now interchanging the edge colours along P, we get a new colouring c with |X1|=|X1|2, |X2|=|X2|+2 and the rest of Xi’s remain unchanged. So trivially,

i|Xi|2|Xi|2=(|X1|2)2+(|X2|2)2|X1|2|X2|2<0,

which contradicts the minimality of i=1k|Xi|2. Thus |X1||X2|+2 and thus (11.1) holds. ∎