Decomposing multigraphs into two Class I subgraphs with prescribed maximum degrees

Note that E_r\ne\varnothing since \chi'(G)=r.

For (i), suppose that \alpha\in \bar{\theta}(u)\cap \bar{\theta}(v). Since \theta(uv)=r, we have \alpha\ne r. Recolor uv with \alpha. If |E_r(\theta)|\ge2, the resulting proper r-edge-coloring has a color-r class of size |E_r(\theta)|-1, contradicting the minimality of |E_r|. If |E_r(\theta)|=1, then we obtain a proper (r-1)-edge-coloring, contradicting r=\chi'(G). Hence \bar{\theta}(u)\cap \bar{\theta}(v)=\varnothing.

For (ii), r\notin \bar{\theta}(u)\cup\bar{\theta}(v). By (i), |\bar{\theta}(u)|+|\bar{\theta}(v)|\le r-1. Since |\bar{\theta}(u)|=r-d_G(u) and |\bar{\theta}(v)|=r-d_G(v), we obtain d_G(u)+d_G(v)\ge r+1.

For (iii), suppose that u and v are not (\alpha,\beta)-linked for some \alpha\in \bar{\theta}(u) and \beta\in \bar{\theta}(v). Interchange \alpha and \beta on the (\alpha,\beta)-component containing u. Let \theta' be the resulting proper r-edge-coloring. Since v lies in a different component, it is unchanged and still misses \beta, while u now also misses \beta. Since r\notin\{\alpha, \beta \}, E_r(\theta')=E_r(\theta). Moreover, \beta\in \bar{\theta'}(u)\cap\bar{\theta'}(v), contrary to (i).

Let X=\bar{\psi}(x) and Y=\bar{\psi}(y). Since X\cap Y=\varnothing, Y\subseteq[r]\setminus X. Moreover, |Y|=r-d_G(y)\le r-q=p and |[r]\setminus X|=d_G(x)\ge p. Hence there exists S\subseteq[r] such that Y\subseteq S\subseteq[r]\setminus X and |S|=p. Let T=[r]\setminus S. Then |T|=q.

Let H_p be the spanning subgraph formed by the color classes in S, and let H_q be the spanning subgraph formed by the color classes in T. Then

Since H_p is the union of p color classes, \chi'(H_p)\le p and \Delta(H_p)\le p. Because S\subseteq[r]\setminus X, every color in S appears at x; and so d_{H_p}(x)=p. Thus

Similarly, Y\subseteq S gives T\cap Y=\varnothing, so every color in T appears at y. Hence d_{H_q}(y)=q, and therefore

Since the theorem is assumed false, Lemma 2.2 cannot apply to the endpoints of uv. By Lemma 2.1(i), \bar{\varphi}(u)\cap \bar{\varphi}(v)=\varnothing. Hence it is impossible that both

Therefore d_G(u)\le p-1 or d_G(v)\le q-1.

By Lemma 2.1(ii), d_G(u)+d_G(v)\ge r+1=p+q+1. Consequently, exactly one of the following two alternatives holds:

1. d_G(u)\le p-1, \qquad d_G(v)\ge q+2;
2. p+2\le d_G(u)\le d_G(v)\le q-1.

It remains to exclude (i).

Suppose, to the contrary, that d_G(u)\le p-1 and d_G(v)\ge q+2. Let A=\bar{\varphi}(u) and B=\bar{\varphi}(v). The set B is nonempty because d_G(v)\le\Delta<r. Choose \beta\in B. Since A\cap B=\varnothing, u sees \beta. Let ux_\beta be the unique edge incident with u that is colored \beta. Since v misses \beta, x_\beta\ne v.

Fix \alpha\in A. By Lemma 2.1(iii), the vertices u,v are (\alpha,\beta)-linked. Their common (\alpha,\beta)-component is a path joining u and v and contains ux_\beta. Since x_\beta\ne v, the vertex x_\beta is internal on that path and hence is incident with an edge colored \alpha. This holds for every \alpha\in A. Therefore

We claim \bar{\varphi}(x_\beta)\cap B=\varnothing. Suppose, to the contrary, that there exists \gamma\in\bar{\varphi}(x_\beta)\cap B. Since x_\beta is incident with an edge colored \beta, \gamma\ne\beta. Choose \alpha\in A. By Lemma 2.1(iii), the vertices u,v lie in one (\alpha,\gamma)-component, which is a path P joining u and v. The vertex x_\beta is incident with an edge colored \alpha and misses \gamma, so it has degree one in G_\varphi(\alpha,\gamma). Since x_\beta\notin\{u,v\}, it cannot be internal on P. Thus x_\beta\notin V(P). Interchange \alpha,\gamma on P. Then \gamma is missing at u, while x_\beta is unchanged and still misses \gamma. Recolor ux_\beta by \gamma. Now \beta is missing at u, while v still misses \beta. Recolor uv by \beta. The result is a proper r-edge-coloring with one fewer edge of color r, contradicting the minimality of |E_r|. Therefore \bar{\varphi}(x_\beta)\cap B=\varnothing.

Now

and

Lemma 2.2 applies to x_\beta,v, a contradiction. Hence (i) is impossible, and (ii) follows.

Since d_G(z)=\Delta, |\bar{\theta}(z)|=r-\Delta=k. Moreover, uv\in E_r(\theta), and hence r\notin A\cup B. By Lemma 2.1(i), A\cap B=\varnothing. Since r\in\bar{\theta}(z), it follows that

On the other hand, Lemma 2.3 gives d_G(v)\le q-1. Therefore |B|=r-d_G(v)\ge r-(q-1)=p+1\ge k+1, where the last inequality follows from p\ge k. Consequently,

Call a proper r-edge-coloring \theta admissible if E_r(\theta)=E_r(\varphi), \bar{\theta}(u)=A, \bar{\theta}(v)=B and r\in\bar{\theta}(z). By the definitions of A and B and by the observation above, the coloring \varphi is admissible.

Among all admissible colorings, choose \theta such that |A\cap\bar\theta(z)| is minimum. Suppose, to the contrary, that A\cap\bar{\theta}(z)\ne\varnothing. Choose a\in A\cap\bar{\theta}(z). By Lemma 2.4, there exists b\in B\setminus\bar{\theta}(z).

Since \bar{\theta}(u)=A, \bar{\theta}(v)=B and A\cap B=\varnothing, the color a is missing at u and present at v, while the color b is missing at v and present at u. Moreover, E_r(\theta)=E_r(\varphi) and uv\in E_r(\theta). By Lemma 2.1(iii), the vertices u and v are (a,b)-linked under \theta. Hence the (a,b)-component containing u and v is a path with endvertices u and v.

At z, the color a is missing and the color b is present. Thus z has degree one in G_\theta(a, b). Moreover, z\notin\{u,v\}, because d_G(z)=\Delta, whereas Lemma 2.3 gives d_G(u),d_G(v)\le q-1\le\Delta-1. Hence z cannot lie in the (a,b)-component containing u and v: every non-endvertex of that path has degree two in G_\theta(a, b), while z has degree one. Thus P_z(a, b, \theta) is distinct from the (a,b)-component containing u and v.

Interchange a and b on P_z(a, b, \theta), obtaining a proper r-edge-coloring \theta'. Since u,v\notin V(P_z(a, b, \theta)), \bar{\theta'}(u)=A and \bar{\theta'}(v)=B. Also a,b\ne r, because color r is present at both u and v. Therefore

Moreover,

Since a\in A and b\in B with A\cap B=\varnothing,

Finally, a, b\ne r, so r remains missing at z. Hence \theta' is again admissible.

This contradicts the choice of \theta. Therefore A\cap\bar\theta(z)=\varnothing. Taking \psi=\theta completes the proof.

If k=0, then \chi'(G)=\Delta. Let \varphi be a proper \Delta-edge-coloring of G, and choose a vertex z with d_G(z)=\Delta. Partition the color set [\Delta] into two sets S and T with |S|=p and |T|=q. Let H_p and H_q be the spanning subgraphs formed by the color classes in S and T, respectively. Since every color in [\Delta] appears at z, we have

Hence

Thus the result holds when k=0. Henceforth assume k\ge1.

Suppose, to the contrary, that the theorem fails. Choose \varphi and the minimum color class E_r(\varphi) as above, and fix an edge uv\in E_r(\varphi). Relabel u and v, if necessary, so that d_G(u)\le d_G(v). By Lemma 2.3, p+2\le d_G(u)\le d_G(v)\le q-1.

Let A=\bar{\varphi}(u), B=\bar{\varphi}(v), and choose a vertex z with d_G(z)=\Delta.

By Lemma 2.5, there exists a proper r-edge-coloring \psi such that

and

Hence

Moreover,

and

Therefore Lemma 2.2, applied to u and z under the coloring \psi, yields an edge decomposition

into two spanning Class I subgraphs satisfying

This contradicts the assumption that G is a counterexample.