Minimal spectral radius of graphs with given matching number

By Theorem 3.1, T_2^*\in \mathcal{G}_{n,2} is a tree. Then we can get a dominating set X of T_2^* with |X|=2 which satisfies conditions (i) and (ii) by Theorem 2.3. Moreover, the quasi-adjacency graph of X are block graphs by Corollary 2.1. In fact, the quasi-adjacency graph of X is P_2. There are only two types of graphs, as illustrated in Fig. ? (top). Let’s us denominate by T_1(a,b) the tree on the left, having two central nodes and by T_2(a,b) the tree on the right. Our goal is to show that, for a given n\geq 4, T_2(a,b) is the one minimizing the spectral radius. Suppose that the conclusion does not hold. Then T_2^* is isomorphic to T_1(a,b), where a,b \ge 1. Without loss of generality, we may assume that both a,b \ge 2; otherwise, T_1(a,b) can be viewed as T_2(a,b-1) or T_2(a-1,b). The graph T_2(a-1,b) can be obtained from T_1(a,b) by first subdividing the unique internal edge and then deleting one pendant vertex. By Lemmas 3.1 and 3.2, we have

which contradicts the minimality of the spectral radius of T_2^*. Therefore, T_2^* is isomorphic to T_2(a,b). Now, we invoke Lemma 3.3 to show that among all T_2(a,b) the minimizer it the balanced. T_2^* \cong T_2\!\left(\lfloor \tfrac{n-3}{2} \rfloor,\, \lceil \tfrac{n-3}{2} \rceil\right).

By Theorem 3.1, T_3^* \in \mathcal{G}_{n,3} is a tree. For trees with matching number 3, the quasi-adjacency graph can only be P_3 or C_3 by Corollary 2.1. According to the mapping in Fig. ?, there are four types of graphs in \mathcal{T}_{n,3}, as shown in Fig. ?.

By lemma 3.4, T_3^* \cong T_3(a,b,c), where a+b+c=n-5. By symmetry, we may assume that a \geq c. By Lemma 3.3, c+1 \geq a \geq c.

Note that n \ge 11. Then s = \left\lfloor \frac{n-5}{3} \right\rfloor \ge 2. Next we will distinguish three cases to discuss.

Case (i) If n - 5 = 3s. Since T_3(a,b,c) has proper subgraphs K_{1,a+1}, K_{1,b+2}, and K_{1,c+1}, we have s+1 + \sqrt{3} = \rho^{2}(T_3(s+1,\, s-2,\, s+1)\ge \rho^{2}(T_3(a,b,c))> \max\{a+1,\; b+2,\; c+1\}, which induces that a,c \le s + 1 and b \le s. By Claim ? and a \geq c, we have T_3^* \cong T_3(s+1,\, s-2,\, s+1) or T_3(s+1,\, s-1,\, s). Since \varphi(T_3(s+1,\, s-1,\, s),x)=s+1 + \sqrt{3})=\sqrt{3}-2 < 0, \rho(T_3(s+1,\, s-1,\, s)) \geq s+1 + \sqrt{3} = \rho^{2}(T_3(s+1,\, s-2,\, s+1). Thus, we have T_3^* \cong T_3(s+1,\, s-2,\, s+1).

Case (ii) If n - 5 = 3s+1. Since T_3(a,b,c) has proper subgraphs K_{1,a+1}, K_{1,b+2}, and K_{1,c+1}, we have s+3 = \rho^{2}(T_3(s+1,\, s-1,\, s+1)\ge \rho^{2}(T_3(a,b,c))> \max\{a+1,\; b+2,\; c+1\}, which induces that a,c \le s + 1 and b \le s. By Claim ? and a \geq c, we have T_3^* \cong T_3(s+1,\, s-1,\, s+1).

Case (iii) If n - 5 = 3s+2. Since T_3(a,b,c) has proper subgraphs K_{1,a+1}, K_{1,b+2}, and K_{1,c+1}, we have s+2+ \sqrt{2} = \rho^{2}(T_3(s+1,\, s,\, s+1)\ge \rho^{2}(T_3(a,b,c))> \max\{a+1,\; b+2,\; c+1\}, which induces that a,c \le s + 2 and b \le s+1. By Claim ? and a \geq c, we have T_3^* \cong T_3(s+1,\, s,\, s+1) , T_3(s+2,\, s-2,\, s+2) or T_3(s+2,\, s-1,\, s+1). After calculation, \rho^{2}(T_3(s+2,\, s-2,\, s+2)=\frac{2s + 3 \pm \sqrt{17}}{2}=s+3.5616 \geq \rho^{2}(T_3(s+2,\, s-1,\, s+1)=s+3.5321 \geq \rho^{2}(T_3(s+1,\, s,\, s+1)=s+2+ \sqrt{2}. Therefore, T_3^* \cong T_3(s+1,\, s,\, s+1).

This completes the proof.

By Proposition 4.1, we conclude that

By Lemma 4.13, the conclusion of Theorem 1.3 follows immediately.

Let G be a simple connected graph in \mathcal{G}_{n,\beta}, and let M = \{e_1, \dots, e_\beta\} be a maximum matching in G. We construct a spanning tree T of G that preserves \beta as its matching number.

Case 1: G is a tree. Now, T = G itself is the required spanning tree, as M remains a maximum matching in T.

Case 2: G contains cycles. Let E_M = E(G) \setminus M. We claim that every cycle C in G contains at least one edge e \in E_M. Indeed, if C \subseteq M, then C would consist of disjoint edges, which is impossible for a cycle. Thus, C \cap E_M \neq \emptyset.

In order to construct T, we iteratively remove edges from cycles as follows: Select a cycle C in the current graph G. Remove an arbitrary edge e \in C \cap E_M, yielding G' = G - e. Since e \notin M, M remains a matching in G' with a size of \beta. The graph G' stays connected because e belongs to a cycle. Repeat until no cycles remain.

The resulting graph T is a spanning tree of G because it is connected, acyclic, and contains all vertices of G. Moreover, since M is preserved throughout the process, we have \beta(T) = \beta.

Let T \in \mathcal{T}_{n,\beta} and fix an arbitrary maximal matching M = \{e_1 = u_1v_1, \dots, e_\beta = u_\beta v_\beta\} of T. Let U = V(M). If T[U] is connected, then it must be a tree since T is acyclic. If T[U] is disconnected, we iteratively modify M while preserving maximality until T[U] becomes connected.

Step 1: Initialize l := 1. While T[U] is disconnected, perform the following steps:

Step 2: Let T_l be the largest connected component of T[U].

Step 3: Select an edge e_{t_l} = u_{t_l}v_{t_l} \in M \setminus E(T_l) that minimizes the distance \bar{d}_T(E(T_l), e_{t_l}). Let u'v' \in E(T_l) be an edge realizing this minimum distance, i.e.,

Without loss of generality, let d(v',v_{t_l})=d(u'v',e_{t_l}). Then we claim that d(v',v_{t_l}) = 2. Otherwise, if d = 1, it contradicts that T_1 is the largest connected component; if d \geq 3, we can find a path u'v'w_1 \cdots w_kv_{t_l}u_{t_l} (k \geq 2, w_i \notin U) connecting u'v' and e_{t_l}, and thus find a matching M + w_1w_2 larger than M since w_i \notin U, which is a contradiction.

Therefore, Let P = u'v'w_1v_{t_l}u_{t_l} be the unique path in T connecting u'v' to e_{t_l}, where w_1 \notin U. Replace e_{t_l} in M with w_1v_{t_l}. This replacement preserves the matching size \beta while reducing the number of connected components in T[U].

Step 4: Update U to reflect the modified matching and increment l.

The procedure terminates when T[U] becomes connected because: (i) Each iteration strictly decreases the number of connected components. (ii) The total number of components is finite (at most \beta). (iii) The matching remains maximal at every step.

Upon termination, T[U] is a connected induced subgraph of the tree T and thus must itself be a tree containing all edges of M. This completes the proof.

By Theorem 2.2, there exists a maximal matching M = \{e_1 = u_1v_1, \ldots, e_\beta = u_\beta v_\beta\} of T such that the induced subgraph T[U] is a tree, where U = \{u_1, v_1, \ldots, u_\beta, v_\beta\}. Let W = V(T) \setminus U. First, we present the structural properties of W as follows.

Our goal is to construct a dominating set X satisfying conditions (i) and (ii). By Claim ?, we may assume N_W(u_i) = \emptyset for all i \in \{1, \ldots, \beta\} and define X = \{v_1, \ldots, v_\beta\}. Since M covers U, every u \in U is either in X or matched to a vertex in X. By Claim Proof and the assumption N_W(u_i) = \emptyset, every w \in W is a leaf adjacent to some v_i \in X. Thus, X is a dominating set of T with |X| = \beta.

By the choice of X and Claim ?, we have Q \subseteq X \subset U. Furthermore, we assert that every vertex in X has at least one quasi-adjacent vertex. To verify this, note that since T is a tree, for any v_i \in X and any other v_j \in X (v_j \neq v_i), there is a unique path connecting v_i and v_j in T. Along this path, select the vertex v_k \in X (distinct from v_i) closest to v_i; by definition, v_k is a quasi-adjacent vertex of v_i, confirming the assertion.

Next, we adjust X so that it satisfies condition (ii) while ensuring that condition (i) remains fulfilled. Since n \geq 2\beta + 1, at least one of v_1, v_2, \ldots, v_\beta is adjacent to vertices in W; without loss of generality, let v_1 be such a vertex. We traverse all quasi-adjacent vertices of v_1 in X \setminus A, and in each iteration select the closest one, denoted by v_{t_1}. By Claim ?, d_T(v_1, v_{t_1}) \le 3. Initialize A = \{v_1\}, and let T_0 be the subgraph induced by \{u_1, v_1, u_{t_1}, v_{t_1}\} and their neighbors in W. We then process these quasi-adjacent vertices in the order of Cases 1–4 (each time denoted by v_{t_1}) and perform the corresponding operations:

Case 1. d_T(v_1, v_{t_1}) = 1 and d_W(v_{t_1}) = 0. If u_{t_1} is not a leaf, swap the labels of u_{t_1} and v_{t_1}, and then the structure of T_0 becomes (2) in Fig.?. If u_{t_1} is a leaf, no operation is performed, and then T_0 has structure (1) in Fig.?. Notably, label swaps preserve the condition (i).

Case 2. d_T(v_1, v_{t_1}) = 1 and d_W(v_{t_1}) \neq 0. no operation is performed, and then T_0 has structure (3) in Fig.?.

Case 3. d_T(v_1, v_{t_1}) = 2. No operation is performed, and then T_0 has structure (2), (4), or (5) in Fig.?.

Case 4. d_T(v_1, v_{t_1}) = 3. In fact, if there exists a quasi-adjacent vertex of v_1 at distance three, then there also exists a (possibly different) vertex v_t lying on a path v_1u_iu_tv_t; thus, we select such a vertex v_t as v_{t_1}. We assert d_W(v_{t_1}) = 0; otherwise, since d_W(v_1) \ge 1, a larger matching than M would exist, contradicting its maximality. Hence, we swap the labels of u_{t_1} and v_{t_1}. It is worth noting that the label swaps also preserve condition (i).

After each iteration, update A := A \cup \{v_{t_1}\} and repeat the process until X \setminus A contains no quasi-adjacent vertices of v_1.

Let X_1 = A \setminus \{v_1\}, which is the set of quasi-adjacent vertices of v_1. After the above swap operations, we have d_T(v_1, v_{t_1}) \le 2 for all v_{t_1} \in X_1.

We first traverse the quasi-adjacent vertices of v_1 to obtain X_1. Then, for each vertex in X_1, we successively find all of its quasi-adjacent vertices that are closest to it, whose union forms X_2. Similarly, for each vertex in X_2, we obtain all its quasi-adjacent vertices, whose union forms X_3, and so on (see Fig.?). This process continues iteratively until all vertices in X have been traversed. At each iteration, the corresponding operations are performed according to the four cases (Case 1–Case 4). As a result, we obtain d_T(v_{t_i}, v_{t_{i+1}}) \leq 2(i=1,2,\dots,k) for v_{t_i} \in X_i and v_{t_{i+1}} \in X_{i+1} with v_{t_i} \dot{\sim} v_{t_{i+1}}.

Next, we only need to consider the distances between two quasi-adjacent vertices in X_i. In fact, for every X_i, if v_{t_i}, v_{t_i}' \in X_i and v_{t_i} \dot{\sim} v_{t_i}', then there must exist quasi-adjacent vertices, say v_{i-1}, v_{i-1}', in X_{i-1}. If v_{i-1} \neq v_{i-1}', then T will create a cycle, a contradiction. If v_{i-1} = v_{i-1}', then P_{v_{i-1}v_{t_i}} = v_{i-1}u_jv_{t_i}, P_{v_{i-1}v_{t_i}'} = v_{i-1}u_jv_{t_i}', where u_j \in X_j, otherwise T will create a cycle. Therefore, we have d(v_{t_i}, v_{t_i}') = 2.

Consequently, X satisfies condition (ii), that is, for any two quasi-adjacent vertices v_i, v_j \in X, d_T(v_i, v_j) \le 2. This completes the proof.

Suppose there is a cycle C = v_1v_2\cdots v_kv_1 of length k in the quasi-adjacent graph T_X of the tree T, where X is a control set chosen in accordance with Theorem 2.3 of T. We aim to show that any two vertices v_i and v_j on C (with i < j) are quasi-adjacent.

If j = i+1 or (i=1 \text{ and } j=k), the conclusion follows immediately from the definition of a cycle (since consecutive vertices in a cycle are connected by an edge, hence quasi-adjacent by construction of the quasi-adjacent graph). For the remaining case where v_i and v_j are not consecutive in C, suppose, by contradiction, that v_i and v_j are not quasi-adjacent. Their positions in this scenario are illustrated in Figure ?.

We now establish the distances in T between consecutive vertices in the cycle of T_X.

By Claim ?, we may form a cycle C' in T_X of length k-1 by substituting the path v_i v_{i+1} v_{i+2} in C with the quasi-adjacent edge v_i \dot{\sim} v_{i+2}, yielding C' = v_1 v_2 \cdots v_i v_{i+2} \cdots v_k v_1.

We proceed by induction on the distance between vertices along C. For the base case, Claim ? establishes quasi-adjacency for vertices with two intermediate vertices (i.e., distance 2 in C). For the inductive step, consider vertices v_i and v_{i+3}: since v_i \dot{\sim} v_{i+2} (by Claim ?) and v_{i+2} \dot{\sim} v_{i+3} (as consecutive vertices in C), an argument analogous to the proof of Claim ? (applying the cycle C' in place of C) shows v_i \dot{\sim} v_{i+3}.

Iterating this reasoning, we conclude v_i \dot{\sim} v_m in T_X for all m satisfying i < m \leq k. Hence, all pairs v_i, v_j \in C with i < j are quasi-adjacent, completing the proof.

Let G^* be a graph in \mathcal{G}_{n,\beta} with minimal spectral radius \rho(G^*). Suppose, for contradiction, that G^* is not a tree. By Theorem 2.1, there exists a spanning tree T of G^* that belongs to \mathcal{G}_{n,\beta} and has the same matching number \beta. Since T is a proper subgraph of G^* (as G^* contains cycles while T is acyclic), it follows from Lemma 3.2 that \rho(T) < \rho(G^*). This contradicts the minimality of \rho(G^*) in \mathcal{G}_{n,\beta}. Therefore, G^* must be a tree.

By contradiction, assume that T_4^* \cong K_1(a,b,c,d) with a, b, c, d \geq 1. Construct a tree T_1 by subdividing one internal edge of K_1(a,b,c,d) and deleting one pendant edge incident to v^* from K_1(a,b,c,d). Clearly, T_1 \in \mathcal{G}_{n,4}. By Lemmas 3.1 and 3.2, we have

which contradicts the minimality of \rho(T_4^*). Hence, the lemma follows.

Suppose, by contradiction, that T_4^* \cong K_2(a,b,c,d) with a \geq 1, b \geq 0, c \geq 1, d \geq 1. We consider the following cases:

If a = 1, b = 0 and d = 1, then let T_2 be the tree obtained from K_2(a,b,c,d) by deleting one pendant edge incident to v_3 and attaching a new pendant edge to the only leaf adjacent to v_1. Then T_2 \in \mathcal{T}_{n,4}. By Lemma 4.1, \rho(T_2) < \rho(K_2(a,b,c,d)), which also contradicts the minimality of \rho(T_4^*).

If a \geq 2 or b \geq 1 or d \geq 2, then T^* we notice that it always has an internal edge. Let T_3 be the tree obtained from K_2(a,b,c,d) by subdividing one internal edge (v_2v_3 or v_3v_4) and deleting one pendant edge incident to v^*. Clearly, T_3 \in \mathcal{T}_{n,4}. By Lemmas 3.1 and 3.2, we have \rho(T_3) < \rho(K_2(a,b,c,d)), contradicting the minimality of \rho(T_4^*).

Assume that T_4^* \cong K_3(a,b,c,d) with a, b, c, d \geq 1.

If a \geq 2 or d\geq 2, let T_4 be the tree obtained from K_3(a,b,c,d) by subdividing v_1v_2 (if a \geq 2) or v_3v_4 (if d \geq 2) and deleting one pendant edge incident to v^*. Clearly, T_4 \in \mathcal{T}_{n,4}. By Lemmas 3.1 and 3.2, we have \rho(T_4) < \rho(K_3(a,b,c,d)), which contradicts the minimality of \rho(T_4^*). Otherwise, when a=1 and d=1, let T_4 be the tree obtained from K_3(a,b,c,d) by subdividing one edge of P_{v_2v_3} and deleting one pendant edge incident to v_1. Clearly, T_4 \in \mathcal{T}_{n,4}. By Lemmas 3.1 and 3.2, we also have \rho(T_4) < \rho(K_3(a,b,c,d)), which contradicts the minimality of \rho(T_4^*).

Assume that T_4^* \cong K_4(a,b,c,d) with a \ge 1, b \ge 0, c \ge 0, and d \ge 1.

For d \ge 2 and (a,b,c) \neq (1,0,0), let T_5 be the tree obtained from K_4(a,b,c,d) by subdividing the edge v_3v_4 and deleting one pendant edge incident to v^*. Clearly, T_5 \in \mathcal{G}_{n,4}. By Lemmas 3.1 and 3.2, \rho(T_5) < \rho(K_4(a,b,c,d)), contradicting the minimality of \rho(T_4^*).

If d \ge 2 and a = 1, b = c = 0, then let T_6 be the tree obtained from K_4(a,b,c,d) by deleting d-1 pendant edges incident to v_4 and attaching these d-1 pendant edges to the unique leaf adjacent to v_4. Clearly, T_6 \in \mathcal{G}_{n,4}. By Lemma 4.1, \rho(T_6) < \rho(K_4(a,b,c,d)), contradicting the minimality of \rho(T_4^*).

If d = 1 and exactly one of v_1, v_2, v_3 has degree greater than 3. Let v' be that vertex. Let T_7 be the tree obtained from K_4(a,b,c,d) by deleting all pendant edges incident to v' and then attaching the same number of pendant edges to the leaf adjacent to v_4. Clearly, T_7 \in \mathcal{G}_{n,4}. By Lemma 4.1, \rho(T_7) < \rho(K_4(a,b,c,d)), contradicting the minimality of \rho(T_4^*).

If d = 1 and at least two of v_1, v_2, v_3 have degree greater than 3. Let v'' and v''' be two such vertices. Let T_8 be the tree obtained from K_4(a,b,c,d) by subdividing an edge on the path P_{v''v'''} and then deleting one pendant edge incident to v_4. Clearly, T_8 \in \mathcal{G}_{n,4}. By Lemmas 3.1 and 3.2, \rho(T_8) < \rho(K_4(a,b,c,d)), contradicting the minimality of \rho(T_4^*).

Assume for contradiction that T_4^* \cong K_5(a,b,c,d) with a \ge 1, b \ge 0, c \ge 0, and d \ge 1. Observe that when both b and c of K_{5}(a,b,c,d) are zero, it can be regarded as K_{4}(a,b,c,d). By Lemma 4.6, the conclusion is immediate. Therefore, it suffices to consider K_{5}(a,b,c,d) with b \ge 1, without loss of generality.

Let \mathbf{x}=(x_1, x_2, \dots, x_n) be the Perron vector of K_{6}(a,b-1,c,d). Let u_1 denote the unique common neighbor of v_1 and v_2, and let u_2 denote the unique common neighbor of v_2 and v_3, as shown in Fig ?. Let W_i denote the set of all leaves adjacent to v_i, for i=1,2,3,4.

If x_{v_2} \ge x_{u_2}, by deleting the edge u_2v_3 from K_{6}(a,b-1,c,d) and adding the edge v_2v_3, we obtain K_{5}(a,b,c,d). By Lemma 4.2, we have \rho(K_{6}(a,b-1,c,d)) < \rho(K_{5}(a,b,c,d)), contradicting the minimality of \rho(T_4^*). If x_{v_2} < x_{u_2}, then K_{5}(a,b,c,d)=K_{6}(a,b-1,c,d)-\{v_2w:w \in W_2\}-u_1v_2+\{u_2w:w \in W_2\}+u_1u_2. By Lemma 4.2, we again have \rho(K_{6}(a,b-1,c,d)) < \rho(K_{5}(a,b,c,d)), which also contradicts the minimality of \rho(T_4^*).

Assume for contradiction that T_4^* \cong K_7(a,b,c,d) with a,b,c,d \ge 1.

If At least one of a,c,d is at least 2. Without loss of generality, suppose a \ge 2. Let T_{9} be the tree obtained from K_7(a,b,c,d) by subdividing the edge v_1v_2, then deleting one pendant edge incident to v^*. Clearly, T_{9} \in \mathcal{G}_{n,4}. By Lemmas 3.1 and 3.2, we have \rho(T_{9}) < \rho(K_7(a,b,c,d)), contradicting the minimality of \rho(T_4^*).

If a = c = d = 1. Let T_{10} be the tree obtained from K_7(a,b,c,d) by deleting the edge v_2v_4 and connecting v_4 to the unique leaf adjacent to v_3. Clearly, T_{10} \in \mathcal{G}_{n,4}. By Lemma 4.1, \rho(T_{10}) < \rho(K_7(a,b,c,d)), contradicting the minimality of \rho(T_4^*).

If a\geq 2 or c\geq 2 or d\geq 2, we notice that K_7(a,b,c,d) has an internal edge that when subdivided keeps the tree in \mathcal{G}_{n,4} and by the same reasoning it can’t be minimal. It remains to study the case a=c=d=1 and b\geq 1, when there is no internal path. As n\geq 10 already, we may assume that b\geq 2. In this case we take one pendant edge in v_2 and make it pendant in the pendant edge of v_3. The resulting tree is still in \mathcal{G}_{n,4} and has smaller spectral radius by Lemma 4.1, contradicting the minimality.

Assume for contradiction that T_4^* \cong K_8(a,b,c,d) where a,b,c,d \ge 1.

If c \ge 2 or d \ge 2. We may assume c\geq 2. Let T_{11} be the tree constructed from K_8(a,b,c,d) by subdividing the edge v_2v_3, and deleting one pendant edge incident to v^*. Clearly, T_{11} \in \mathcal{G}_{n,4}. By Lemmas 3.1 and 3.2, we have \rho(T_{11}) < \rho(K_8(a,b,c,d)), which contradicts the minimality of \rho(T_4^*).

If c = d = 1. Let T_{12} be the tree obtained from K_8(a,b,c,d) by deleting the edge v_2v_4 and connecting v_4 to the unique leaf adjacent to v_3. Clearly, T_{12} \in \mathcal{G}_{n,4}. By Lemma 4.1, \rho(T_{12}) < \rho(K_8(a,b,c,d)), contradicting the minimality of \rho(T_4^*).

Assume for contradiction that T_4^* \cong K_9(a,b,c,d) where a \ge 1, b \ge 0, c \ge 1, and d \ge 1. We analyze the following cases:

If d \ge 2. Let T_{13} be the tree obtained from K_9(a,b,c,d) by first subdividing the edge v_2v_4, then deleting one pendant edge incident to v^*. Clearly, T_{13} \in \mathcal{T}_{n,4}. By Lemmas 3.1 and 3.2, we have \rho(T_{13}) < \rho(K_9(a,b,c,d)), which contradicts the minimality of \rho(T_4^*).

If d = 1 and either a \ge 2 or c \ge 2. Without loss of generality, suppose a \ge 2. Let T_{14} be the tree obtained from K_9(a,b,c,d) by first subdividing an edge on the path P_{v_1v_2}, then deleting one pendant edge incident to v_4. Clearly, T_{14} \in \mathcal{T}_{n,4}. By Lemmas 3.1 and 3.2, \rho(T_{14}) < \rho(K_9(a,b,c,d)), which contradicts the minimality of \rho(T_4^*).

If d = 1, a = 1, and c = 1. Let T_{15} be the tree obtained from K_9(a,b,c,d) by deleting the edge v_2v_4 and connecting v_4 to the unique leaf adjacent to v_3. Clearly, T_{15} \in \mathcal{T}_{n,4}. By Lemma 4.1, \rho(T_{15}) < \rho(K_9(a,b,c,d)), which contradicts the minimality of \rho(T_4^*).

Assume, by way of contradiction, that T_4^* \cong K_{11}(a,b,c,d) with a,b,c,d \ge 1. Observe that when d = 1 in K_{11}(a,b,c,d), it can be regarded as K_{12}(a,b,c,d-1). By Lemma 4.12, the result follows directly. Hence, we only need to consider K_{11}(a,b,c,d) with d \ge 2.

Let \mathbf{y}=(y_1, y_2, \dots, y_n) be the Perron vector of K_{12}(a,b,c,d-1). Let u_2 denote the unique common neighbor of v_3 and v_4, as shown in Fig ?. Let W_i denote the set of all leaves adjacent to v_i, for i=1,\ldots,4.

If y_{v_4} \ge y_{u_2}, by deleting the edge v_3u_2 from K_{12}(a,b,c,d-1) and adding the edge v_3v_4, we obtain K_{11}(a,b,c,d). By Lemma 4.2, we have \rho(K_{12}(a,b-1,c,d)) < \rho(K_{11}(a,b,c,d)), contradicting the minimality of \rho(T_4^*). Otherwise, if y_{v_4} < y_{u_2}, then K_{11}(a,b,c,d)=K_{12}(a,b,c,d-1)-\{v_4w:w \in W_4\}+\{u_2w:w \in W_4\}. By Lemma 4.2, we again have \rho(K_{12}(a,b,c,d-1)) < \rho(K_{11}(a,b,c,d)), which also contradicts the minimality of \rho(T_4^*).

Observe that a and b of K_{12}(a,b,c,d) cannot be zero simultaneously. Without loss of generality, assume that b \ge 1. The characteristic polynomial of a star K_{1,n} is given by \varphi(K_{1,n}, x) = x^{n-1}(x^2 - n). Both K_{12}(a,b,c,d) and K_6(a,b-1,c,d) are composed of several star graphs. By Lemma 3.5, we compute the characteristic polynomials as follows: \varphi(K_{12}(a,b,c,d), x) = x^{a+b+c+d-2}\big[(x^4 - (a+b+2)x^2 + ab + a + b)(x^4 - (c+d+2)x^2 + cd + c + d) - (x^2 - a)(x^2 - b)(x^2 - d - 1)\big], and \varphi(K_6(a,b-1,c,d), x) = x^{a+b+c+d-2}\big[ (x^4 - (a+b+1)x^2 + a(b-1) + a + b - 1) \cdot (x^4 - (c+d+3)x^2 + (c+1)d + c + d + 1) - (x^4 - (c+d+2)x^2 + cd + c + d)(x^2 - (a+1))\big].

For x > \max\left\{\sqrt{d+1}, \sqrt{b-1}\right\}, we obtain: \varphi(K_6(a,b-1,c,d), x) - \varphi(K_{12}(b,a,c,d), x)\\ = x^{a+b+c+d-2}(x^2 - (d+1))(x^2 - (b-1)) > 0.

Notice that \rho(K_{1,d+1})=\sqrt{d+1} and \rho(K_{1,b-1})= \sqrt{b-1}. Since both K_6(a,b-1,c,d) and K_{12}(b,a,c,d) contain K_{1,d+1} and K_{1,b-1} as induced proper subgraphs, it follows from Lemma 3.2 that: \rho(K_6(a,b-1,c,d)), \rho(K_{12}(b,a,c,d) > \max\big\{ \sqrt{d+1}, \sqrt{b-1}\big\}. Hence, we conclude that: \rho(K_6(a,b-1,c,d)) < \rho(K_{12}(b,a,c,d)), which implies that T_4^*\ncong K_{12}(a,b,c,d).

Observe that among a, b, c, and d of K_{13}(a,b,c,d), at most one can be zero. Without loss of generality, assume that c \ge 1. Next, we compare the characteristic polynomials of K_{12}(a,b,c-1,d) and K_{13}(a,b,c,d). By Lemma 3.5, they are given by \varphi(K_{13}(a,b,c,d), x) \\= x^{a+b+c+d-3} (x^2 - a)\big[ (x^2 - b)(x^4 - (c+d+2)x^2 + cd + c + d) - (x^2 - c)(x^2 - d) - (x^2 - b)(x^2 - c)(x^2 - d) \big] , and \varphi(K_{12}(a,b,c-1,d), x) = x^{a+b+c+d-3} \big[(x^4 - (a+b+2)x^2 + ab + a + b)\cdot (x^4 - (c+d+1)x^2 + (c-1)d + c + d - 1) - (x^2 - a)(x^2 - b)(x^2 - d - 1) \big].

For x > \max\big\{\sqrt{c-1}), \sqrt{a}, \sqrt{b}\big\}, we have:

Since both graphs contain K_{1,c-1}, K_{1,a}, and K_{1,b} as induced proper subgraphs, from Lemma 3.2 we have \rho(K_{12}(a,b,c-1,d)),\rho(K_{13}(a,b,c,d))>\max\big\{\sqrt{c-1}), \sqrt{a}, \sqrt{b}\big\}. Hence, \rho(K_{13}(a,b,c,d)) > \rho(K_{12}(a,b,c-1,d)), which implies T_4^*\ncong K_{13}(a,b,c,d).