Linear bounds for rich-neighbor edge-colorings

Let \varphi be the coloring, with color set \mathcal C. By properness and (?),

If G is simple, then \mu_G(u,v)=1. If, in addition, G is nonempty and \delta(G)\ge2, then every edge is non-isolated, so every rich-neighbor edge-coloring contains a rich edge. Applying the bound to such an edge gives \chi'_{\mathrm{rn}}(G)\ge2\delta(G)-1. \Box

If F is total edge-dominating, then every edge of G has an endpoint in V(F), so V(G)\setminus V(F) is independent. Conversely, suppose that V(G)\setminus V(F) is independent. Every edge in E(G)\setminus E(F) has an endpoint in V(F) and hence is adjacent to an edge of F. Every edge of F also has a neighbor in F, since its component contains at least two edges. \Box

A forest in the support graph of G can be realized as an isomorphic simple forest in G by choosing one edge from each corresponding parallel class. Since G and its support graph have the same independent sets, we may assume that G is simple. We induct on |V(G)|. If |V(G)|=3, take a spanning two-edge star.

For |V(G)|\ge4, Lemma 2.3 gives a large star S\cong K_{1,t} such that G-V(S) has at most one nontrivial component. Let c be its center and x_1,\ldots,x_t its leaves. Let I be the set of isolated vertices of G-V(S), and let H be its nontrivial component, if one exists. If H does not exist, take F=S. If |V(H)|\ge3, let F_H be given by induction and take F=S\cup F_H. In both cases, F is an almost-star forest and its uncovered vertex set is independent, since the vertices of I are isolated in G-V(S).

It remains to consider H\cong K_2, with edge uv. Since G is connected, we may assume that u has a neighbor in S. If uc\in E(G), add u to S as a leaf; the uncovered set I\cup\{v\} is independent. Otherwise, we may label the leaves so that ux_1\in E(G). If t\ge3, take F to consist of the two vertex-disjoint stars with edge sets \{uv,ux_1\} and \{cx_j\colon\, 2\le j\le t\}. Both stars are large, and the uncovered set is I.

Finally, suppose that t=2, and put y=x_2. If y has a neighbor w\in I, take the two vertex-disjoint stars with edge sets \{uv,ux_1\} and \{yc,yw\}. The uncovered set is I\setminus\{w\}. Otherwise, take F to be the path vux_1c. The uncovered set I\cup\{y\} is independent because y has no neighbor in I. The final assertion follows from Lemma 2.2. \Box

Fix i and write B=B_i. Suppose that B has a nonempty subgraph H with \delta(H)\ge3. Put n=|V(H)| and M=\{e\in M_i\colon\, \widehat e\in V(H)\}. Since M is a matching, its edges have 2n distinct endpoints, and each edge of F witnesses at most one edge of H.

Let P be the P_4-component of F, or the empty graph if no such component exists. Let E_P be the set of edges of H having a witness in P, and choose one such witness for each edge of E_P. Put E^*=E(H)\setminus E_P, p=|E_P|, and m=|E^*|. Let U=V(P)\cap V_G(M) and t=|U|. The chosen witnesses for E_P are distinct edges of P[U]. Since P[U] is a forest on t vertices,

For each edge of E^*, choose a witness in a star component of F. Let \mathcal S be the set of star components containing a chosen witness, and put s=|\mathcal S|.

Suppose that an edge e\in M meets neither the center of a star in \mathcal S nor an internal vertex of P. Each endpoint of e then lies on at most one chosen witness: a star leaf or an endpoint of P lies on only one edge of F, whereas a vertex outside V(F) or the center of a star outside \mathcal S lies on no chosen witness. Thus d_H(\widehat e)\le2, a contradiction. Every edge of M therefore meets a center of a star in \mathcal S or an internal vertex of P. Since M is a matching and P has at most two internal vertices,

If t=0, no edge of M meets P, so the same argument gives

The m chosen star witnesses are distinct and have distinct leaf endpoints, since distinct edges of a simple star have distinct leaves and the components of F are vertex-disjoint. These m leaves, the s star centers, and the t vertices of U are pairwise distinct endpoints of edges in M. Consequently,

If t=0, equations (?) and (?) give |E(H)|=m\le n. If t\ge1, equations (?), (?), and (?) give

However, H is simple and \delta(H)\ge3, so n\ge4 and |E(H)|\ge3n/2>n+1, a contradiction. Thus B is 2-degenerate. \Box

The lower bound follows from Lemma 2.1, since \delta(K_n)=n-1.

For the upper bounds, fix v\in V(K_n). Assign distinct colors to the n-1 edges incident with v, and properly edge-color K_n-v using a disjoint palette of \chi'(K_{n-1}) colors. Every edge incident with v is rich. Since n\ge3, each of these edges is adjacent to another rich edge. Every remaining edge is also adjacent to a rich edge incident with v. Thus we obtain a rich-neighbor edge-coloring with n-1+\chi'(K_{n-1}) colors.

The stated bounds follow from \chi'(K_{n-1})=n-2 when n is odd and \chi'(K_{n-1})=n-1 when n is even. \Box

For every rich edge uv, equation (?) and |\Phi(u)|=|\Phi(v)|=n-1 give

Suppose that uv and vw are distinct edges of R. Put \alpha=\varphi(uv), \beta=\varphi(vw), and M=\mathcal C\setminus\Phi(v). By (Proof), \Phi(u)=M\cup\{\alpha\}, \Phi(w)=M\cup\{\beta\}. Since \alpha\ne\beta, we have \Phi(u)\cap\Phi(w)=M, whose size is n-2\ge2. Thus uw is not rich, so R contains no triangle. Suppose now that wz\in E(R) for some z\notin\{u,v,w\}, and put \gamma=\varphi(wz). Properness gives \gamma\in\Phi(w)\setminus\{\beta\}=M, and (Proof) yields \Phi(z)=(\Phi(v)\setminus\{\beta\})\cup\{\gamma\}. Hence \Phi(u)\cap\Phi(z)=\{\alpha,\gamma\}. The color of uz must belong to this intersection, but \alpha is already used on uv and \gamma on wz, contradicting properness. Thus R contains neither a triangle nor a path of length three, so every nontrivial component of R is a star.

If \varphi is a rich-neighbor edge-coloring, no component of R consists of a single edge, since that edge would have no rich neighbor. Nor can R have two isolated vertices, since the edge joining them in K_n would have no rich neighbor. \Box

Let \mathcal C be the color set, let \Phi(v) be the set of colors on edges incident with v, and let R be the rich-edge graph. By Lemma 3.2, every nontrivial component of R is a star with at least two edges, and R has at most one isolated vertex. Also, R contains an edge by the rich-neighbor condition. By (Proof), if uv is rich, then every edge incident with v other than uv has its color outside \Phi(u).

Suppose that R has exactly one nontrivial component, with center a. All but at most one vertex of K_n-a are leaves of this star. Hence every edge of K_n-a has its color in \mathcal C\setminus\Phi(a), which has size n-2. Since n-1 is odd, each color class in K_n-a has at most (n-2)/2 edges. Thus these colors cover at most (n-2)^2/2<\binom{n-1}{2} edges, a contradiction. Therefore, R has at least two nontrivial components.

Now suppose that n\equiv2\pmod4 and that R has exactly two nontrivial components. Write n=2r, where r\ge3 is odd. Let a,b be the centers of the two stars, and let A,B be their leaf sets, respectively. Put p=|A| and q=|B|, so p,q\ge2. Let s\in\{0,1\} be the number of isolated vertices of R. Then p+q=2r-2-s. Let t=|\Phi(a)\cap\Phi(b)| and put T=\mathcal C\setminus(\Phi(a)\cup\Phi(b)). Since |\Phi(a)|=|\Phi(b)|=2r-1, we have |T|=t-1. Every edge joining A to B has its color in T. The edges from a to B and from b to A have their colors in \Phi(a)\setminus\Phi(b) and \Phi(b)\setminus\Phi(a), respectively. Properness therefore gives t-1\ge\max\{p,q\} and 2r-1-t\ge\max\{p,q\}. Adding these inequalities yields \max\{p,q\}\le r-1. Together with p+q=2r-2-s, this implies, after interchanging the stars if necessary, that q=r-1, p=r-1-s and t=r. Put Q=\Phi(b)\setminus\Phi(a). Then |Q|=|T|=r-1 and \mathcal C\setminus\Phi(a)=T\cup Q.

At each vertex of A, the r-1 edges to B use all colors in T. Every other incident edge, except the edge to a, must therefore have its color in Q. If s=0, let J=K_n[A\cup\{b\}]. If s=1, let x be the isolated vertex of R and let J=K_n[A\cup\{b,x\}]-bx. In both cases, J has r vertices and \binom r2-s edges, all colored from Q.

Since r is odd, each color class in J has at most (r-1)/2 edges. Thus the r-1 colors in Q cover at most (r-1)^2/2 edges. However,

since p\ge2 and s\le1. This contradiction proves the second assertion. \Box

By Lemma 3.1, it suffices to exclude a rich-neighbor edge-coloring of K_4 with five colors and of K_6 with nine colors.

By Lemmas 3.2 and 3.3, the rich-edge graph of the first coloring would have at least two nontrivial star components, each containing at least three vertices. This is impossible on four vertices. For the second coloring, the same lemmas require at least three such components, which is impossible on six vertices. \Box

By Lemma 3.1, it suffices to construct a rich-neighbor edge-coloring of K_n with 2n-3 colors. Write n=2r, where r\ge4 is even. Choose distinct vertices a,b, and partition the remaining vertices into A=\{a_1,\ldots,a_{r-1}\} and B=\{b_1,\ldots,b_{r-1}\}. We construct a coloring in which all edges aa_i and bb_i are rich.

The induced subgraphs K_n[\{a\}\cup B] and K_n[\{b\}\cup A] are copies of K_r. Since r is even, each decomposes into r-1 perfect matchings. The edges between A and B also decompose into r-1 perfect matchings, since they form K_{r-1,r-1}. Give each of these 3(r-1) matchings a different color. Next, give the r-1 matchings \{aa_i,bb_i\}, where 1\le i\le r-1, distinct new colors, and give ab one further new color. Every edge is now colored, and each color class is a matching, so the coloring is proper.

For each i, the edges incident with a use only colors assigned to K_n[\{a\}\cup B] and the new colors. The edges incident with a_i other than aa_i lie in K_n[\{b\}\cup A] or join A to B. Their colors are therefore different from all colors used at a, so aa_i is rich. Similarly, every bb_i is rich. These rich edges form two stars that together cover all vertices, each with at least two edges. Hence every edge has a distinct adjacent rich edge. The coloring is therefore rich-neighbor and uses 3(r-1)+(r-1)+1=2n-3 colors. \Box

A one-factorization of K_k, together with an empty color class, gives a proper edge-coloring from [k]. Among all proper edge-colorings \varphi:E(K_k)\to[k], choose one minimizing \sum_{i=1}^k |E_i|^2, where E_i=\varphi^{-1}(i). If |E_i|\ge|E_j|+2, then the subgraph with edge set E_i\cup E_j has an alternating path component with one more edge of E_i than of E_j. Interchanging the two colors on this path preserves properness and decreases the sum of squares, a contradiction. Thus the class sizes differ by at most one. Their average is (k-1)/2=h-\tfrac12, so exactly h classes have size h and the other h have size h-1.

Each vertex is incident with k-1 distinct colors and hence is missed by exactly one color class. The pairs of vertices missed by the classes of size h-1 therefore partition V(K_k). Construct the required partition by placing one vertex of each pair in X_0 and the other in X_1. \Box

Take a one-factorization of K_{A\cup\{u\},\,B\cup\{v\}}. Each factor not containing uv contains edges va and ub for some a\in A and b\in B. These pairs define a bijection f:A\to B. Since |A_1|=|B_1|, applying a suitable permutation of B to the entire factorization allows us to assume that f(A_1)=B_1. For each a\in A_1, replace va and uf(a) in their common factor by ua and vf(a). Each factor remains a perfect matching, and the resulting r+1 factors partition E(H). \Box

Use the cyclic one-factorization above with q=2h-1, w=x, and v_{2h-2}=y. The matching M=\{v_{2j}v_{2j+1}\colon\,0\le j\le h-2\} misses exactly x and y. Its edge colors are 4j+1 modulo 2h-1, which are pairwise distinct since \gcd(4,2h-1)=1. Thus M is rainbow. \Box

For t\in A, the first identity gives \Phi(a)\cap\Phi(t)=\{\varphi(at)\}, so at is rich by (?). The same argument applies to the other two stars. Each star has at least two edges, and their union covers all vertices except x. Thus their union is total edge-dominating by Lemma 2.2, and \varphi is a rich-neighbor edge-coloring. \Box

Write k=2h. Use the vertex partition in Lemma 3.9 with |A|=|B|=k-1 and |C|=k. Choose disjoint color sets with the sizes listed for Construction I in Table ?, and put \mathcal O=\{\alpha,\beta\}. These sets contain 6k+1 colors in total.

The color sets \mathcal P_{AC}, \mathcal P_{BC}, and \mathcal S_C. By Lemma 3.6, color K_C with \mathcal S_C and choose a partition C=C_0\cup C_1, where |C_0|=|C_1|=h, such that h color classes are perfect matchings and each of the other h classes has h-1 edges and misses one vertex in each part. The missed pairs partition C. Choose A'\subseteq A and B'\subseteq B with |A'|=|B'|=h.

Use \mathcal P_{AC} for a one-factorization of K_{C,A\cup\{b\}}. For each v\in C_0, replace bv in its color class by xv and bw, choosing the vertices w\in B' bijectively. The replaced edges lie in distinct classes, and each replacement preserves the matching property. The resulting classes cover every vertex of A\cup C\cup\{b\} and partition E(A,C)\cup E(b,C_1)\cup E(x,C_0)\cup E(b,B'). Similarly, use \mathcal P_{BC} for a one-factorization of K_{C,B\cup\{a\}}. For each v\in C_1, replace av by xv and a distinct edge of E(a,A'). The resulting classes cover every vertex of B\cup C\cup\{a\} and partition

To each \mathcal S_C-class missing v_0\in C_0 and v_1\in C_1, add bv_0 and av_1. Add ab to one of the original perfect classes. To each of the other h-1 perfect classes, add one edge from E(a,A\setminus A') and one from E(b,B\setminus B'), using every such edge once. This is possible since |A\setminus A'|=|B\setminus B'|=h-1. Each class remains a matching and now covers every vertex of C\cup\{a,b\}.

The color sets \mathcal S_A and \mathcal S_B. Choose a_*\in A and b_*\in B. By Lemma 3.8, choose one-factorizations of K_{A\cup\{b\}} and K_{B\cup\{a\}} with palettes \mathcal S_A and \mathcal S_B, respectively, admitting rainbow matchings M_A and M_B of size h-1 that miss exactly \{b,a_*\} and \{a,b_*\}. Choose partitions A=U_A\cup V_A\cup\{a_*\}, B=U_B\cup V_B\cup\{b_*\}, where |U_A|=|V_A|=|U_B|=|V_B|=h-1, such that each edge of M_A joins U_A to V_A and each edge of M_B joins U_B to V_B.

For each uv\in M_A, with u\in U_A and v\in V_A, replace uv in its \mathcal S_A-class by cu and xv. Since M_A is rainbow, each class is modified at most once and remains a matching. There are (k-1)-(h-1)=h unmodified classes. Add the edges of E(c,C_0) to these classes, one edge per class. Every \mathcal S_A-class now covers every vertex of A\cup\{b,c\}. Apply the analogous replacements to M_B, and add the edges of E(c,C_1) to the h unmodified \mathcal S_B-classes, one edge per class. Every \mathcal S_B-class then covers every vertex of B\cup\{a,c\}. Reserve the removed edges of M_A and M_B for \mathcal P_{AB}.

The remaining color classes. Apply Lemma 3.7 with (u,v)=(c,x), A_1=V_A, and B_1=U_B. Use k colors of \mathcal P_{AB} for the resulting one-factorization of the graph with edge set

Assign the remaining color of \mathcal P_{AB} to M_A\cup M_B\cup\{ca_*,xb_*\}. This is a perfect matching on A\cup B\cup\{c,x\} and contains no edge of the preceding graph. Thus every vertex of A\cup B\cup\{c,x\} sees every color in \mathcal P_{AB}. Finally, set \varphi(ac)=\varphi(bx)=\alpha, \varphi(bc)=\varphi(ax)=\beta.

Verification. Every color class is a matching. The edges of M_A and M_B have been removed from their original classes and assigned to \mathcal P_{AB}. The assignments at c and x partition their edges to A as follows:

where the first set in each row uses \mathcal P_{AB} and the second uses \mathcal S_A. The corresponding partitions for B use \mathcal P_{AB} and \mathcal S_B. The edges of E(c,C_0) and E(c,C_1) use \mathcal S_A and \mathcal S_B, respectively. Among the six edges on \{a,b,c,x\}, the edge ab uses \mathcal S_C, the edge cx uses \mathcal P_{AB}, and the other four use \mathcal O. Together with the first step, these assignments partition E(K_{3k+2}).

Each center has the palette in Table ?. Every vertex of A sees all colors in \mathcal P_{AB}\cup\mathcal P_{AC}\cup\mathcal S_A on edges other than its star edge. These 3k colors form its entire non-star palette. The same argument applies to B and C, so all palette equalities in Table ? hold. Lemma 3.9 completes the proof. \Box

Write k=2h, where h\ge3 is odd. Use the vertex partition in Lemma 3.9 with |A|=|B|=k-2 and |C|=k. Choose disjoint color sets with the sizes listed for Construction II in Table ?, and put \mathcal O=\{\alpha,\beta\}. These sets contain 6k-3 colors in total. Partition A=A_0\cup A_1 and B=B_0\cup B_1, where each part has size h-1. Write C=\{v_0,\ldots,v_{k-1}\}, and put C_0=\{v_{2j}:0\le j<h\} and C_1=\{v_{2j+1}:0\le j<h\}, with subscripts on the vertices v_i taken modulo k.

The color sets \mathcal P_{AB}, \mathcal S_A, and \mathcal S_B. Apply Lemma 3.7 with (u,v)=(c,x) and the subsets A_1,B_1. Use \mathcal P_{AB} for the resulting one-factorization of the graph with edge set

Take a one-factorization of K_{A\cup\{b,w\}}, where w is an auxiliary vertex, and assign its k-1 factors the colors in \mathcal S_A. In each factor, replace the edge wu by cu if u\in A_0, by xu if u\in A_1, and by bx if u=b. Exactly h-1 classes now cover c. Add the h edges of E(c,C_0) to the remaining classes, one edge per class. Each class remains a matching and covers A\cup\{b,c\}. The resulting classes partition E(K_A)\cup E(b,A)\cup E(c,A_0) \cup E(x,A_1)\cup\{bx\}\cup E(c,C_0).

Similarly, use \mathcal S_B for a one-factorization of K_{B\cup\{a,w\}}. Replace wu by cu if u\in B_1, by xu if u\in B_0, and by ax if u=a. Add the edges of E(c,C_1) to the h classes missing c, one edge per class. Each resulting class is a matching covering B\cup\{a,c\}, and these classes partition E(K_B)\cup E(a,B)\cup E(c,B_1) \cup E(x,B_0)\cup\{ax\}\cup E(c,C_1).

The color sets \mathcal P_{AC}, \mathcal P_{BC}, and \mathcal O. Choose u_A\in A and u_B\in B, and set \varphi(ac)=\varphi(bu_B)=\alpha and \varphi(bc)=\varphi(au_A)=\beta. Both color classes are matchings. Order A\cup\{b,x\} as \ell_0=b,\ell_1=x,\ell_2,\ldots,\ell_{k-1}. The matchings M_t=\{\ell_i v_{t+i}:0\le i<k\}, where 0\le t<k, form a one-factorization of K_{A\cup\{b,x\},C}. Assign them the colors in \mathcal P_{AC}. For each even t, replace bv_t,xv_{t+1} by bu_t^B,v_tv_{t+1}, where the u_t^B are distinct vertices of B\setminus\{u_B\}. These choices are possible since k-3=2h-3\ge h. Each class remains a matching covering A\cup C\cup\{b\}. Their edge sets partition E(A,C)\cup E(b,C_1)\cup E(x,C_0) \cup\{bu_t^B:0\le t<k,\ t\text{ even}\}\cup D_0, where D_0=\{v_{2i}v_{2i+1}:0\le i<h\}.

For \mathcal P_{BC}, use the corresponding cyclic one-factorization of K_{B\cup\{a,x\},C}, with a,x as the first two vertices. For each odd t, replace av_t,xv_{t+1} by au_t^A,v_tv_{t+1}, using distinct vertices u_t^A\in A\setminus\{u_A\}. Each class remains a matching covering B\cup C\cup\{a\}. Their edge sets partition E(B,C)\cup E(a,C_0)\cup E(x,C_1) \cup\{au_t^A:0\le t<k,\ t\text{ odd}\}\cup D_1, where D_1=\{v_{2i+1}v_{2i+2}:0\le i<h\}.

The color set \mathcal S_C. Since h is odd, for each j\in\{0,1\}, decompose K_{C_j} into near-perfect matchings N_0^j,\ldots,N_{h-1}^j, where N_i^j misses v_{2i+j}. Use h colors of \mathcal S_C for the matchings N_i^0\cup N_i^1\cup\{bv_{2i},av_{2i+1}\}, where 0\le i<h. Each covers C\cup\{a,b\}.

The remaining uncolored edges within C form K_{C_0,C_1}-D_0-D_1. They decompose into the perfect matchings F_d=\{v_{2i}v_{2(i+d)+1}:0\le i<h\}, where 1\le d\le h-2. Indeed, D_0 and D_1 correspond to cyclic shifts 0 and -1 modulo h, respectively, so these are precisely the remaining shifts. Add ab to F_1. To each of the other h-3 factors, add one unused edge of E(a,A) and one unused edge of E(b,B), using every such edge once. Each of these stars has exactly (k-2)-1-h=h-3 unused edges, so these additions are possible. Assign the remaining h-2 colors of \mathcal S_C to the augmented factors. When h=3, only F_1 occurs and no star edges remain. Every \mathcal S_C-class is a matching covering C\cup\{a,b\}.

Verification. The listed edge sets partition E(K_{3k}). The edges inside C are partitioned by D_0,D_1 and the restrictions of the \mathcal S_C-classes to C. The edges from c,x to A,B are partitioned by (?) and the two auxiliary-vertex constructions, while E(c,C) is split between \mathcal S_A and \mathcal S_B. The edges of E(a,A) are divided among \mathcal O,\mathcal P_{BC},\mathcal S_C in numbers 1,h,h-3, respectively. The same counts hold for E(b,B), with \mathcal P_{AC} in place of \mathcal P_{BC}. On \{a,b,c,x\}, the edge ab uses \mathcal S_C, the edges ac,bc use \mathcal O, the edges ax,bx use \mathcal S_B,\mathcal S_A, respectively, and cx uses \mathcal P_{AB}. Every color class is a matching.

The center palettes are those in Table ?. Every leaf receives all colors in its prescribed non-star palette on non-star edges. Each such palette has 3k-2 colors, equal to the number of incident non-star edges, so the palette equalities in the table hold. Lemma 3.9 completes the proof. \Box

Write k=2h. Use the vertex partition in Lemma 3.9 with |A|=|B|=|C|=k. Choose disjoint color sets with the sizes listed for Construction III in Table ?, and put \mathcal O=\{\alpha,\beta\}. These sets contain 6k+5 colors in total.

Apply Lemma 3.6 to K_A,K_B,K_C, using \mathcal S_A,\mathcal S_B,\mathcal S_C, respectively. Choose the resulting partitions A=A_0\cup A_1, B=B_0\cup B_1, and C=C_0\cup C_1, where each part has size h. Each coloring has h perfect classes and h classes of size h-1, each missing one vertex in each part. The missed pairs partition the corresponding leaf set.

The color sets \mathcal P_{AB}, \mathcal P_{BC}, and \mathcal P_{AC}. Apply Lemma 3.7 to the three rows of Table ?, with ordered parts (A,B), (B,C), and (C,A), respectively. Each resulting graph has a one-factorization with k+1 factors. Assign its factors the colors in the indicated color set.

The color sets \mathcal S_A, \mathcal S_B, and \mathcal S_C. Consider the chosen \mathcal S_A-coloring of K_A. To each class of size h-1 missing u\in A_0 and v\in A_1, add cu and bv. To each of the h perfect classes, add one edge of E(b,B_0) and one edge of E(c,C_1), using every such edge once. Each class remains a matching covering A\cup\{b,c\}. Extend the \mathcal S_B- and \mathcal S_C-classes cyclically as specified in Table ?. They cover B\cup\{a,c\} and C\cup\{a,b\}, respectively, in every color.

Together with the original colorings inside A,B,C, the two tables partition all edges except ab,bc,ca. At this stage, the center palettes are those in Table ?, with \mathcal O omitted.

The center triangle. Choose a color \gamma\in\mathcal S_C whose original class in K_C is a perfect matching. Its extended class contains au and bv for some u\in A_0 and v\in B_1. Replace these two edges in the \gamma-class by ab, and set \varphi(au)=\varphi(bc)=\beta and \varphi(bv)=\varphi(ac)=\alpha. The modified \gamma-class and the new \alpha- and \beta-classes are matchings. Both a and b retain color \gamma, and every center receives both \alpha and \beta. Thus the center palettes are exactly those in Table ?.

Verification. The construction partitions E(K_{3k+4}) into matchings and hence defines a proper edge-coloring. Before the final recoloring, every vertex of A receives all colors in \mathcal P_{AB}\cup\mathcal P_{AC}\cup\mathcal S_A on its non-star edges. These are 3k+2 colors, equal to the number of incident non-star edges. The same argument applies to B and C. Since only two prescribed star edges were recolored, all non-star palettes remain unchanged. The palette equalities in Table ? therefore hold. As each leaf set has size k\ge2, Lemma 3.9 completes the proof. \Box