Preprint
Article

This version is not peer-reviewed.

Degree-Based Weighted Adjacency Spectra of Thorny Graphs with Positive Multiplicities

Submitted:

12 October 2026

Posted:

13 October 2026

You are already at the latest version

Abstract
Let G be a finite simple graph, let \( m = (m_v)_{v∈V(G)} \) with \( m_v ≥ 1 \), and form \( G^m \) by attaching \( m_v \) pendant vertices at each \( v \). For the weighted adjacency matrix \( A_φ(G^m) \) induced by a real symmetric degree-dependent edge weight φ, a generalized Fiedler-type decomposition gives M − n forced zero eigenvalues and \( χ_{A_φ(G^m)}(λ) = λ^{M−n} det(λ²I_n − λB_φ − R_φ) \).Here \( n = |V(G)| \), \( M = ∑_v m_v, B_φ \) is the core block formed from the core-vertex degrees in the thorny graph, and \( R_φ \) is diagonal with squared pendant-fiber couplings. The decomposition gives an arbitrary-core nullity bound, exact nullity and inertia for nonzero pendant couplings, Cauchy interlacing with \( B_φ \), and a 2n-dimensional compression independent of M. For fixed r ≥ 1, m, and φ, the nonzero weighted spectrum of a uniform attachment to an r-regular core determines the core adjacency spectrum whenever the induced core-edge weight is nonzero. On connected walk-regular cores, let exceptional vertices receive m + q rather than m pendant vertices. For independent exceptional sets S, T with |S| = |T| ≥ 2, all moments below twice the smaller minimum separation agree for every real symmetric φ. If the separations differ, then at the first possible mixed order the moment of the smaller-separation placement minus that of the larger-separation placement is an explicit nonnegative multiple of a sum of squared geodesic counts. If the ordinary–ordinary core-edge weight is nonzero and its square differs from that of the exceptional–ordinary core-edge weight, the spectrum determines the minimum separation. On distance-regular cores, it also determines the closest-pair count. Both the ordinary Sombor and Randić weights satisfy this condition.
Keywords: 
;  ;  ;  ;  ;  ;  

MSC:  Primary 05C50; Secondary 05C22; 05C76; 05C92

1. Introduction

The Sombor index was introduced by Gutman [1] as a degree-based topological index in which an edge u v contributes d ( u ) 2 + d ( v ) 2 . It has since been studied on chemical graphs and extremal graph classes, and in relation to other degree-based descriptors [2,3,4]. The literature on degree-based graph invariants also includes work on the modified symmetric division deg index, vertex-edge Sombor, Nirmala, and misbalance indices, Sombor entropy, reverse Laplacian energy, reverse hyper-Zagreb indices, and the atom-bond sum-connectivity index [5,6,7,8,9,10]. These papers concern scalar graph indices, entropy measures, or graph energies rather than the full spectra of degree-based weighted adjacency matrices. The matrix analogue of the Sombor index is the Sombor matrix. Its characteristic polynomial, spectral radius, energy, Estrada index, and related spectral quantities have been studied in [11,12,13,14,15,16]. Reverse Sombor energy is another recent degree-based spectral construction [17]. More generally, Das et al. [18] defined a general extended adjacency matrix whose u v -entry on an edge is F ( d ( u ) , d ( v ) ) for a symmetric function F. Further spectral properties of degree-based weighted adjacency matrices have been studied in [19,20,21,22]. More recently, Gao et al. [23] studied Li–Feng transformations for degree-based weighted adjacency matrices. Their results concern largest-eigenvalue behavior under specific transformations rather than the full-spectrum placement questions considered here. Spectra and integrality for multipartite and crown multipartite families have also been studied in [24]. Characteristic-polynomial formulas for the Sombor matrices of several standard transformations of regular graphs were obtained in [25]. Accordingly, the weighted thorny-graph matrices studied here belong to this established class of degree-based weighted adjacency matrices.
Thorny graphs are obtained by attaching pendant vertices to a core graph. For connected cores, this construction goes back at least to the work of Gutman [26] on distances in thorny graphs. The uniform case is the corona G ∘ K m ¯ [27]. The unweighted construction, including non-uniform multiplicities, is already a specialization of the rooted product of Godsil and McKay [28]. The ith rooted factor is K 1 , m i rooted at its center. Corona spectra were later treated explicitly by Barik et al. [29]. More generally, Fiuj Laali et al. [30] determined characteristic, Laplacian, and signless Laplacian polynomials for generalized coronas in which the attached graph may depend on the core vertex. Thus, for φ ≡ 1 , the same non-uniform construction is also the generalized corona with H i = K m i ¯ . Weighted corona products have also been studied with externally assigned weight factors [31]. More directly, Rojo and Medina [32, Theorems 4.2 and 4.6] characterized the adjacency spectrum of a connected positive-weight core with a possibly different positive-weight generalized Bethe tree attached at each core vertex. Taking each attached tree to be a weighted star gives the connected positive-weight, non-uniform case of Theorem 1. Disconnected positive-weight cores follow componentwise. At a more general matrix level, Cardoso et al. [33] developed a multi-block generalization of Fiedler’s lemma. Cardoso et al. [34] then proved the arbitrary pairwise rank-one symmetric block-coupling theorem. The 2 n -dimensional decomposition in Theorem 1 is a direct specialization of that theorem. The same decomposition also has a standard equitable-partition interpretation. For A φ ( G m ) , the core singletons together with the pendant fibers form an equitable partition, and in the normalized cell basis the corresponding quotient is the 2 n -dimensional compression constructed in Theorem 1. For repeated copies of a single rooted weighted graph, Cardoso et al. [35] obtained a complementary Kronecker-product spectral reduction. In its stated connected positive-weight setting, the adjacency case contains the uniform weighted-star attachment considered in Section 5. Theorem 1 also gives the same uniform formula without sign restrictions on the induced weights. Thus the general block-coupling reduction and its positive-weight star-attachment specializations, both uniform and non-uniform, are not presented here as new matrix principles.
For cycle cores, closely related underlying graphs also occur under the names sun, broken sun, and generalized sun. Boulet [36] studied spectral characterizations of sun and broken-sun graphs. Braga et al. [37] obtained eigenvalue-location results for symmetric matrices whose underlying graph is a generalized sun. Wu and Ma [38, Theorems 4–6] determined the adjacency, Laplacian, and Randić spectra of the non-uniform sun family. In particular, non-uniform cycle attachments with a degree-based Randić weighting have already been studied. Liu et al. [19] used kth spectral moments for p-Sombor matrices. Walk-regular graphs were introduced by Godsil and McKay [39]. Dalfó et al. [40, Proposition 3.1, Lemma 4.1, Proposition 4.5, and Theorems 4.2 and 5.3] characterized h-punctually walk-regular graphs through the cospectrality of their perturbed graphs, including pendant-edge additions, and obtained related characterizations of distance-regular graphs. In a specific sun-like family, Koval and Kwan [41, Claims 4.7–4.8] recovered the minimum hub separation and the number of closest hub pairs from spectral moments. They also noted that adjacency moments give the direct closed-walk formulation [41, Remark 2.4]. The scalar-constancy calculation in Section 6 and the two-exception cycle formula in Proposition 5 are special cases of the general two-level scalar placement formula on regular cores proved in a preprint by the author [42, Theorem 3].
Section 4 develops the structural consequences of the reduction, including an arbitrary-core nullity bound, exact nullity and inertia when the pendant couplings are nonzero, direct interlacing with the core block, and compression to dimension 2 n independent of the total number of pendant vertices. Section 5 treats uniform thorns on regular cores. For fixed r ≥ 1 , m, and φ , the nonzero weighted-thorn spectrum determines the adjacency spectrum of the core whenever the induced core-edge weight is nonzero. The main placement result appears in Section 6. For a connected walk-regular core G and fixed m , q , two independent exceptional sets S , T with | S | = | T | ≥ 2 have equal spectral moments at every order below twice the smaller minimum separation for every real symmetric φ . When their minimum separations differ, at the first possible mixed order the moment of the smaller-separation placement minus that of the larger-separation placement is an explicit nonnegative multiple of a sum of squared geodesic counts. A nondegeneracy condition on the two core-edge weights makes this difference strict for both the ordinary Sombor and Randić weights. Under the same condition, on distance-regular cores the critical moment also determines the number of exceptional pairs at minimum distance. An explicit cycle example shows that higher moments can distinguish placements even when the complete pairwise-distance multisets agree. The associated degree-based scalar index remains constant across independent placements of the same cardinality.

2. Preliminaries

Standard graph-theoretic notation is used throughout. See, for example, Harary’s monograph [43]. All graphs are finite, simple, undirected, and nonempty. For a graph H, let V ( H ) , E ( H ) , and A ( H ) denote its vertex set, edge set, and adjacency matrix. The degree of x ∈ V ( H ) is written as d H ( x ) , or d ( x ) when no confusion is possible. For a connected graph G, d G ( x , y ) denotes graph distance. For a square matrix T, let Spec ( T ) denote its eigenvalue multiset, counted with algebraic multiplicity, and write χ T ( λ ) = det ( λ I − T ) for its characteristic polynomial. For each integer ℓ ≥ 0 , call tr ( T ℓ ) its ℓth spectral moment. For ℓ ≥ 1 , tr ( T ℓ ) = ∑ λ ∈ Spec ( T ) λ ℓ . The adjacency spectrum of a graph G is Spec ( A ( G ) ) .
Following the general degree-based matrix, energy, and index construction of Das et al. [18], let φ be a real-valued symmetric function on positive integer pairs,
φ ( x , y ) = φ ( y , x ) .
For a graph H, define its φ -weighted adjacency matrix A φ ( H ) by
( A φ ( H ) ) x y = φ ( d H ( x ) , d H ( y ) ) , x y ∈ E ( H ) , 0 , x y ∉ E ( H ) .
The choice
φ SO ( x , y ) = x 2 + y 2
gives the ordinary Sombor matrix introduced by Gutman [11]. The φ -weighted spectrum of H is Spec ( A φ ( H ) ) . The φ -energy of H is
E φ ( H ) = ∑ λ ∈ Spec ( A φ ( H ) ) | λ | .
The associated scalar index is
I φ ( H ) = ∑ x y ∈ E ( H ) φ ( d H ( x ) , d H ( y ) ) .
Let G be a graph with vertex set { v 1 , … , v n } . A positive thorn vector on G is a vector
m = ( m 1 , … , m n ) ∈ Z ≥ 1 n .
The corresponding thorny graph G m is obtained from G by adding, over each v i , a pendant fiber of m i new vertices adjacent only to v i . For connected G, this is the positive-multiplicity case of the thorn-graph construction studied by Gutman [26]. The same attachment definition is used here without a connectedness assumption. The term “positive” means m i ≥ 1 for every i, so each pendant fiber is nonempty. Put
D i = d G ( v i ) + m i , M = ∑ i = 1 n m i .
Thus D i = d G m ( v i ) and
| V ( G m ) | = n + M .

3. Scalar and Spectral Thorn Invariants

The following identity is the scalar counterpart of the matrix reduction developed in Section 4. For the ordinary Sombor weight, it specializes to the scalar thorny-graph formula studied in [44].
Proposition 1.
Let G be a graph on v 1 , … , v n and let m be a positive thorn vector. Then
I φ ( G m ) = ∑ v i v j ∈ E ( G ) φ ( D i , D j ) + ∑ i = 1 n m i φ ( D i , 1 ) .
Proof. 
The edges of G m fall into two types. An original edge v i v j contributes φ ( D i , D j ) . Each of the m i pendant edges over v i contributes φ ( D i , 1 ) . Summing these contributions gives the formula. □
Proposition 2.
Let G be a graph on v 1 , … , v n , let m = ( m 1 , … , m n ) be a positive thorn vector, and let σ ∈ Aut ( G ) . Write σ ( v i ) = v σ ( i ) , and define m σ by m σ ( i ) σ = m i . Then G m and G m σ are isomorphic. In particular, their φ-weighted adjacency matrices are permutation similar and have the same characteristic polynomial.
Proof. 
Send each core vertex v i to v σ ( i ) , and send the pendant vertices over v i bijectively onto the pendant vertices over v σ ( i ) . Since σ preserves adjacency in the core and the corresponding pendant fibers have equal sizes, this map is a graph isomorphism. It therefore preserves degrees, so the weighted matrices are permutation similar. □

4. A Determinant Reduction

Theorem 1.
Let G be a graph on n vertices, let m = ( m 1 , … , m n ) be a positive thorn vector, and let M = ∑ i m i . Define the n × n matrix B φ by
( B φ ) i j = φ ( D i , D j ) , v i v j ∈ E ( G ) , 0 , v i v j ∉ E ( G ) , .
Define Q φ = diag ( m i φ ( D i , 1 ) ) i = 1 n . Set R φ = Q φ 2 = diag ( m i φ ( D i , 1 ) 2 ) i = 1 n . The matrix B φ uses the degrees D i in G m and therefore need not equal A φ ( G ) , whose weights use the degrees in G. Then
χ A φ ( G m ) ( λ ) = λ M − n det ( λ 2 I n − λ B φ − R φ ) .
Proof. 
Order the vertices of G m by listing the n core vertices first and then the pendant vertices fiber by fiber. In this order,
A φ ( G m ) = B φ P P T 0 .
Here P is an n × M matrix, the zero block is M × M , and the ith row of P equals φ ( D i , 1 ) in exactly the m i columns belonging to the pendant vertices over v i and is zero in all other columns. The pendant fibers are disjoint, so
P P T = diag m 1 φ ( D 1 , 1 ) 2 , … , m n φ ( D n , 1 ) 2 = R φ .
For each i, the subspace of vectors supported on the pendant fiber over v i and having coordinate sum zero has dimension m i − 1 and lies in the zero eigenspace. Since the fiber supports are disjoint, these subspaces form an ( M − n ) -dimensional forced zero subspace.
The orthogonal complement of this forced zero subspace is spanned by the n core vectors and n normalized fiber vectors. Since A φ ( G m ) is symmetric and the forced zero subspace is invariant, its orthogonal complement is invariant as well. The ith normalized fiber vector equals m i − 1 / 2 on the pendant vertices over v i and zero elsewhere. In the resulting orthonormal basis, the restriction to this 2 n -dimensional complement is
C φ = B φ Q φ Q φ 0 .
Together with a basis of the forced zero subspace, these vectors give an orthonormal basis of R n + M adapted to the two invariant subspaces. Hence
χ A φ ( G m ) ( λ ) = λ M − n det λ I n − B φ − Q φ − Q φ λ I n .
For λ ≠ 0 , the Schur complement gives
det λ I n − B φ − Q φ − Q φ λ I n = det ( λ 2 I n − λ B φ − Q φ 2 ) .
Since Q φ 2 = R φ and both sides are polynomials in λ , the identity holds for all λ . □
  • Relation with a generalized Fiedler reduction.
The orthogonal decomposition above is a direct specialization of the symmetric block-coupling theorem proved by Cardoso et al. [34, Theorem 3], which builds on the earlier multi-block generalization of Fiedler’s lemma in [33]. Take 2 n blocks consisting of n singleton blocks ( 0 ) for the core vertices and, for each i, a zero block 0 m i for the ith pendant fiber. Choose the unit vector 1 in each singleton block and m i − 1 / 2 ( 1 , … , 1 ) T in the corresponding zero block. Set the coupling between core singletons i and j equal to φ ( D i , D j ) when v i v j ∈ E ( G ) and to 0 otherwise, and set the coupling between core singleton i and its fiber block equal to m i φ ( D i , 1 ) , with all other couplings zero. The reduced coupling matrix in that theorem is exactly C φ , while the eigenvectors in the zero fiber blocks orthogonal to the chosen fiber vectors contribute the M − n zero eigenvalues. The direct proof also makes P , Q φ , R φ explicit and gives the quadratic determinant used later.
  • Unweighted specializations.
If φ ≡ 1 , then B φ = A ( G ) and R φ = diag ( m 1 , … , m n ) , so Theorem 1 becomes
χ A ( G m ) ( λ ) = λ M − n det λ 2 I n − λ A ( G ) − diag ( m 1 , … , m n ) .
This specializes the rooted-product formula of Godsil and McKay [28, Theorem 2.1] by taking the ith rooted factor to be K 1 , m i rooted at its center. It is also the H i = K m i ¯ specialization of the generalized-corona characteristic-polynomial formula in [30, Theorem 3.1]. If in addition m 1 = ⋯ = m n = m , write G ( m ) for G m . If μ 1 , … , μ n are the adjacency eigenvalues of G, then
χ A ( G ( m ) ) ( λ ) = λ n ( m − 1 ) ∏ i = 1 n ( λ 2 − μ i λ − m ) .
This is the usual uniform-thorn/corona adjacency formula already contained in [28,29].
Corollary 1.
Let G be a graph on n vertices, let m = ( m 1 , … , m n ) be a positive thorn vector, and put M = ∑ i m i . Then the nullity of A φ ( G m ) is at least M − n . After accounting for this ( M − n ) -dimensional forced zero subspace, the remaining 2 n eigenvalues, counted with multiplicity, are exactly the eigenvalues of
C φ = B φ Q φ Q φ 0 .
Assume that all pendant couplings are nonzero:
φ ( D i , 1 ) ≠ 0 ( 1 ≤ i ≤ n ) .
The multiplicity of 0 is then exactly M − n and
In A φ ( G m ) = ( n , n , M − n ) .
Here In ( T ) = ( n + ( T ) , n − ( T ) , n 0 ( T ) ) denotes the inertia of a real symmetric matrix.
Proof. 
The construction in Theorem 1 gives the orthogonal decomposition
A φ ( G m ) ∼ 0 M − n ⊕ C φ .
This proves the first assertion. If all φ ( D i , 1 ) are nonzero, then Q φ is invertible. Set
S = I n 0 − 1 2 Q φ − 1 B φ I n .
Since B φ and Q φ are symmetric,
S T C φ S = 0 Q φ Q φ 0 .
For each diagonal entry q i ≠ 0 of Q φ , the corresponding 2 × 2 block 0 q i q i 0 has one positive and one negative eigenvalue. Sylvester’s law of inertia therefore gives In ( C φ ) = ( n , n , 0 ) , and the result follows. □
  • Related cycle-core inertia.
When the core is a cycle and all core-edge and pendant couplings are nonzero, the underlying graph of A φ ( G m ) is a generalized sun. In that special setting, the general symmetric-matrix result of Braga et al. [37, Theorem 3.3], with pendant diagonal value 0, gives the same inertia pattern. Corollary 1 is stated for an arbitrary core and requires only the pendant couplings φ ( D i , 1 ) to be nonzero. The core-edge weights themselves may vanish.
Proposition 3.
Under the hypotheses and notation of Theorem 1, list the eigenvalues of A φ ( G m ) in nonincreasing order as
λ 1 ≥ ⋯ ≥ λ n + M .
Likewise, list the eigenvalues of B φ in nonincreasing order as
θ 1 ≥ ⋯ ≥ θ n .
Then
λ i ≥ θ i ≥ λ i + M , 1 ≤ i ≤ n .
If γ 1 ≥ ⋯ ≥ γ 2 n are the eigenvalues of the compression C φ in Corollary 1, then also
γ i ≥ θ i ≥ γ i + n , 1 ≤ i ≤ n .
Proof. 
The matrix B φ is the principal submatrix of A φ ( G m ) indexed by the core vertices, and it is also the leading principal submatrix of C φ . Both inequalities are therefore direct applications of Cauchy’s interlacing theorem. □
  • Spectral compression.
The decomposition above separates M − n zero eigenvalues and replaces the ( n + M ) × ( n + M ) eigenvalue problem by the 2 n × 2 n symmetric matrix C φ , or equivalently by the n × n quadratic matrix polynomial in Theorem 1. For dense eigensolvers, this reduces the matrix dimension from n + M to 2 n whenever M > n , with a corresponding reduction in the leading cubic cost. The dimension reduction is most pronounced when the number of pendant vertices is much larger than the size of the core. This is a structural reduction, not a claim of superiority over specialized sparse eigensolvers.

5. Uniform Thorns on Regular Cores

Let G be an r-regular graph of order n, and let m ≥ 1 be an integer. Attach exactly m pendant vertices to each core vertex, and denote the resulting graph by G ( m ) . This is the corona G ∘ K m ¯ of G with the edgeless graph on m vertices [27]. Put
D = r + m , α = φ ( D , D ) , β = φ ( D , 1 ) .
Then B φ = α A ( G ) and R φ = m β 2 I n . The adjacency spectrum of graph coronas is described in [29]. For the degree-sum weight φ ( x , y ) = x + y , the corresponding uniform-thorn characteristic polynomial over a regular core was obtained by Shinde et al. [45, Theorem 4.5]. When G is connected and α , β > 0 , the attachment theorem of Cardoso et al. [35, Theorem 2.1] for rooted weighted graphs applies by assigning weight α to each core edge and attaching at each core vertex a star with m leaves whose edges have weight β . Thus their adjacency result contains this positive-weight connected case of Theorem 2. The spectral, energy, and spectral-radius formulas below remain valid without sign restrictions on the induced values of α and β . For inverse recovery, a nonzero core-edge weight suffices whenever r ≥ 1 . When r ≤ 1 , a nonzero pendant weight suffices for recovery by order detection.
Theorem 2.
Let G be an r-regular graph with adjacency eigenvalues μ 1 , … , μ n . For G ( m ) one has
Spec ( A φ ( G ( m ) ) ) = { 0 n ( m − 1 ) } ∪ α μ i + α 2 μ i 2 + 4 m β 2 2 , α μ i − α 2 μ i 2 + 4 m β 2 2 : 1 ≤ i ≤ n .
The union is taken with multiplicity.
Proof. 
Theorem 1 gives
χ A φ ( G ( m ) ) ( λ ) = λ n ( m − 1 ) det ( λ 2 − m β 2 ) I n − λ α A ( G ) .
If x is an eigenvector of A ( G ) with eigenvalue μ i , then the determinant factor acts on x as multiplication by
λ 2 − α μ i λ − m β 2 .
The two roots of this quadratic are the two displayed eigenvalues. The factor λ n ( m − 1 ) accounts for n ( m − 1 ) additional zero eigenvalues. □
Corollary 2.
For the induced values α and β defined above, one has
E φ ( G ( m ) ) = ∑ i = 1 n α 2 μ i 2 + 4 m β 2 .
With ρ ( T ) denoting the spectral radius of T, one also has
ρ ( A φ ( G ( m ) ) ) = | α | r + α 2 r 2 + 4 m β 2 2 .
Proof. 
For a real number t, let
x ± ( t ) = α t ± α 2 t 2 + 4 m β 2 2 .
Their product is − m β 2 ≤ 0 . Hence
| x + ( t ) | + | x − ( t ) | = α 2 t 2 + 4 m β 2 .
This identity also holds when β = 0 . Summing over t = μ 1 , … , μ n proves the energy formula. Also,
max { | x + ( t ) | , | x − ( t ) | } = | α t | + α 2 t 2 + 4 m β 2 2 .
Every adjacency eigenvalue t of an r-regular graph satisfies | t | ≤ r , while r itself is an adjacency eigenvalue. The displayed expression is nondecreasing in | t | , so its maximum is attained at | t | = r , giving the spectral-radius formula. The n ( m − 1 ) forced zero eigenvalues in Theorem 2 contribute nothing to the energy and do not affect the spectral radius.
□
  • Ordinary Sombor specialization.
For the ordinary Sombor matrix, φ = φ SO and hence
α = 2 ( r + m ) , β = ( r + m ) 2 + 1 .
Thus
E φ SO ( G ( m ) ) = ∑ i = 1 n 2 ( r + m ) 2 μ i 2 + 4 m ( r + m ) 2 + 1 .
Proposition 4.
Fix integers r ≥ 0 and m ≥ 1 and a weight φ. Suppose either r ≥ 1 and φ ( r + m , r + m ) ≠ 0 , or r ≤ 1 and φ ( r + m , 1 ) ≠ 0 . Then, for every r-regular graph G, the nonzero spectrum of A φ ( G ( m ) ) determines the adjacency spectrum of G.
Proof. 
By Theorem 2, if α β ≠ 0 , the two roots contributed by a core eigenvalue μ have product − m β 2 and sum α μ . Since − m β 2 < 0 , each quadratic contributes exactly one positive and one negative root. Hence the positive part of the nonzero thorn spectrum has exactly n eigenvalues, counted with multiplicity. Every positive root λ is paired with − m β 2 / λ , and the corresponding core eigenvalue is μ = ( λ − m β 2 / λ ) / α . Thus the full adjacency spectrum is recovered.
Suppose next that r ≥ 1 , α ≠ 0 , and β = 0 . The nonzero thorn eigenvalues are exactly α μ i for the nonzero adjacency eigenvalues μ i of G. Dividing by α recovers these μ i , while
∑ i = 1 n μ i 2 = tr ( A ( G ) 2 ) = n r .
This identity determines n and hence the multiplicity of 0 in the adjacency spectrum.
The only remaining case allowed by the hypotheses has r ≤ 1 , α = 0 , and β ≠ 0 . Then the nonzero thorn spectrum consists of ± m | β | , each with multiplicity n, so it recovers the order. A 0-regular core has adjacency spectrum { 0 n } , while a 1-regular core has spectrum { 1 n / 2 , ( − 1 ) n / 2 } . Thus the core spectrum is determined in every case. □
Corollary 3.
Let r ≥ 0 and m ≥ 1 be integers, and let φ be a weight. If r ≤ 1 or φ ( r + m , r + m ) ≠ 0 , then two r-regular cores have the same φ-weighted uniform-thorn spectrum if and only if they are adjacency cospectral.
Proof. 
One implication follows from Theorem 2. Conversely, if r ≤ 1 , equality of the full thorn spectra forces the cores to have the same order, since each thorn spectrum has n ( m + 1 ) eigenvalues. The adjacency spectrum of a 0- or 1-regular graph is determined by its order. If r ≥ 2 and α ≠ 0 , equality of the full thorn spectra gives equality of their nonzero parts, and Proposition 4 recovers the core adjacency spectra. □
For r ≥ 2 , the hypothesis α ≠ 0 cannot be dropped in general. If α = 0 , Theorem 2 gives the same weighted-thorn spectrum to all r-regular cores of the same order. For example, the 2-regular cores C 6 and C 3 ∪ C 3 have different adjacency spectra but identical weighted-thorn spectra in this case.
Example 1.
Let G 1 = K 3 □ K 2 be the triangular prism and let G 2 = K 3 , 3 . Both graphs are connected cubic graphs of order 6. For fixed φ and m, Proposition 1 gives I φ ( G 1 ( m ) ) = I φ ( G 2 ( m ) ) . Their adjacency spectra are
Spec ( A ( G 1 ) ) = { 3 , 1 , 0 , 0 , − 2 , − 2 } , Spec ( A ( G 2 ) ) = { 3 , − 3 , 0 , 0 , 0 , 0 } .
They are therefore not adjacency cospectral. For the ordinary Sombor weight, the induced core-edge weight is nonzero, so Corollary 3 shows that G 1 ( m ) and G 2 ( m ) have different Sombor spectra for every m ≥ 1 . Thus scalar indices of the form I φ do not distinguish this pair, whereas their Sombor spectra do.

6. Separated Thorn Placements on Walk-Regular Cores

Placement effects are not confined to cycle cores. Dalfó et al. [40, Proposition 3.1, Lemma 4.1, Proposition 4.5, and Theorems 4.2 and 5.3] characterized h-punctually walk-regular graphs through the cospectrality of their perturbed graphs, including pendant-edge additions. They also proved a preservation result for removal-cospectral vertex sets and characterized distance-regular graphs through perturb-cospectrality of isometric subsets. The present perturbation differs in that changing a thorn multiplicity both attaches additional pendant vertices and changes the core-vertex degree used to define the weight of every pre-existing edge incident with that core vertex. The theorem below gives an exact critical-moment comparison in this setting.
Recall that a graph G is walk-regular if ( A ( G ) ℓ ) v v is independent of v for every integer ℓ ≥ 0 [39]. Walk-regular graphs are regular, as the case ℓ = 2 shows, and every vertex-transitive graph is walk-regular. Let G be a connected walk-regular graph on n vertices, and write its degree as r. Fix integers m , q ≥ 1 . For S ⊆ V ( G ) , let H S be the thorny graph obtained by attaching m + q pendant vertices to every vertex of S and m pendant vertices to every vertex outside S. Put
D = r + m , D ′ = r + m + q , a = φ ( D , D ) , b = φ ( D ′ , D ) .
When S is independent and | S | ≥ 2 , define its minimum separation by
δ ( S ) = min { d G ( x , y ) : x , y ∈ S , x ≠ y } .
For vertices x , y with d G ( x , y ) = h , let g h ( x , y ) denote the number of x–y geodesics in G. For U ⊆ V ( G ) and h ≥ 2 , put
Γ h ( U ) = ∑ { x , y } ⊆ U d G ( x , y ) = h g h ( x , y ) 2 .
For k = | S | , let e G ( S ) be the number of core edges spanned by S. Regularity gives e G ( S ) , r k − 2 e G ( S ) , and n r / 2 − r k + e G ( S ) core edges of types ( D ′ , D ′ ) , ( D ′ , D ) , and ( D , D ) , respectively. Proposition 1 therefore shows that I φ ( H S ) depends only on k and e G ( S ) . Hence independent sets of the same cardinality have the same scalar index [42, Theorem 3]. The weighted spectrum can nevertheless distinguish such sets.
On a vertex-transitive core, single-exception cospectrality follows from automorphisms. The following lemma establishes the same conclusion under walk-regularity.
Lemma 1.
Let G be a connected walk-regular graph, let m , q ≥ 1 , and form H { v } by making one core vertex exceptional as above. Then the characteristic polynomial of A φ ( H { v } ) is independent of the choice of v ∈ V ( G ) .
Proof. 
Write the degree of G as r and put D = r + m and D ′ = r + m + q . Let A = A ( G ) , let e = e v be the standard basis vector at v, and put w = A e . Set
a = φ ( D , D ) , b = φ ( D ′ , D ) , c = φ ( D , 1 ) , c ′ = φ ( D ′ , 1 ) , κ = ( m + q ) c ′ 2 − m c 2 .
For the single-exception graph, Theorem 1 gives the quadratic matrix polynomial
K v ( λ ) = K 0 ( λ ) − λ ( b − a ) ( e w T + w e T ) − κ e e T , K 0 ( λ ) = ( λ 2 − m c 2 ) I − λ a A .
Write X v = ( e w ) and
C λ = − κ − λ ( b − a ) − λ ( b − a ) 0 ,
so that K v ( λ ) = K 0 ( λ ) + X v C λ X v T . Whenever K 0 ( λ ) is invertible, the matrix determinant lemma gives
det K v ( λ ) = det K 0 ( λ ) det I 2 + C λ X v T K 0 ( λ ) − 1 X v .
The entries of X v T K 0 ( λ ) − 1 X v are determined by the three quantities
e T K 0 ( λ ) − 1 e , e T K 0 ( λ ) − 1 A e , e T A K 0 ( λ ) − 1 A e .
For each such fixed λ , the Cayley–Hamilton theorem expresses K 0 ( λ ) − 1 as a polynomial in K 0 ( λ ) , hence as a polynomial in A. The three quantities are therefore diagonal entries at v of polynomials in A, so walk-regularity makes them independent of v. Hence det K v ( λ ) is independent of v whenever K 0 ( λ ) is invertible. Since det K 0 ( λ ) has leading term λ 2 n , K 0 ( λ ) is invertible for all but finitely many λ , and polynomiality extends the identity to every λ . The fixed factor λ n m + q − n in Theorem 1 now gives the result. □
Theorem 3.
Let G be a connected walk-regular graph of degree r, let m , q ≥ 1 , and let φ be a real symmetric weight. Put
D = r + m , D ′ = r + m + q , a = φ ( D , D ) , b = φ ( D ′ , D ) .
Let S , T ⊆ V ( G ) be independent sets with | S | = | T | ≥ 2 , and put
h = min { δ ( S ) , δ ( T ) } .
Then
tr A φ ( H S ) ℓ = tr A φ ( H T ) ℓ ( 0 ≤ ℓ < 2 h ) .
If δ ( S ) < δ ( T ) , so that h = δ ( S ) , then
tr A φ ( H S ) 2 h − tr A φ ( H T ) 2 h = 2 h a 2 h − 4 ( b 2 − a 2 ) 2 Γ h ( S ) .
When h = 2 , the factor a 2 h − 4 is a 0 = 1 . In the displayed unequal-separation comparison, the critical-moment difference is nonnegative, and it is positive exactly when
b 2 ≠ a 2 and h = 2 or a ≠ 0 .
Consequently, if a ≠ 0 and b 2 ≠ a 2 , then for fixed G , m , q , φ and exceptional-set cardinality, the φ-weighted spectrum determines the minimum separation of an independent exceptional set.
Proof. 
The case ℓ = 0 follows because H S and H T have the same number of vertices. Fix ℓ ≥ 1 and write k = | S | = | T | . Start from the uniform thorn graph G ( m ) and adjoin k disjoint sets L 1 , … , L k , each consisting of q isolated vertices. Let
A 0 = A φ ( G ( m ) ) ⊕ 0 k q .
For a core vertex v and a label j ∈ { 1 , … , k } , let E v ( j ) denote the matrix perturbation produced by making v exceptional and using L j for its q additional leaves. Every nonzero entry of E v ( j ) corresponds to an edge incident with v. If an independent k-set U is enumerated as U = { u 1 , … , u k } , then no core edge has two endpoints in U, so the core-edge changes from distinct labels do not overlap. The original pendant fibers and the sets L j are disjoint, so
A φ ( H U ) = A 0 + ∑ j = 1 k E u j ( j ) .
The terms in the trace expansion that use no perturbation label are independent of U. For a fixed label j, the total contribution of all words using that label at least once and no other perturbation label is
tr ( A 0 + E u j ( j ) ) ℓ − tr ( A 0 ℓ ) .
The matrix A 0 + E u j ( j ) is permutation similar to A φ ( H { u j } ) ⊕ 0 ( k − 1 ) q , so Lemma 1 makes this contribution independent of u j . Hence the combined contribution of all words using at most one perturbation label depends only on G , m , q , φ , k , and ℓ.
Suppose that a word involving at least two distinct perturbation labels has nonzero trace. Some closed walk contributing to its trace visits the center of every perturbation label used by the word. Choose two such centers x and y and one occurrence of each. The two cyclic subwalks between these occurrences each have length at least d G ( x , y ) . Excursions through pendant vertices cannot shorten either route between the two core vertices. Hence the word length is at least 2 d G ( x , y ) . It follows that no word involving at least two distinct perturbation labels has nonzero trace when ℓ < 2 δ ( U ) . Since h = min { δ ( S ) , δ ( T ) } , this proves the equality of the traces below 2 h .
Assume now that δ ( S ) < δ ( T ) , so that h = δ ( S ) . It remains to compute the mixed contribution at order 2 h for U = S and U = T . No word of length 2 h involving at least three distinct perturbation labels can have nonzero trace. If it did, some closed walk contributing to its trace would visit at least three distinct exceptional centers. Choose one occurrence of any three such centers and list the three occurrences in cyclic order. The intervening subwalks each have length at least h, giving total length at least 3 h > 2 h . The preceding two-label bound also shows that, at order 2 h , only pairs of centers at distance h can contribute. Fix a pair { x , y } ⊆ U with d G ( x , y ) = h . The total contribution of the words using both corresponding labels is the inclusion–exclusion quantity
tr ( A x y 2 h ) − tr ( A x 2 h ) − tr ( A y 2 h ) + tr ( A 0 2 h ) .
Here A x , A y , and A x y denote the matrices obtained from A 0 by applying the indicated one or two exceptional perturbations.
A closed walk contributing to this quantity must visit both x and y. Choose one occurrence of each center. The two cyclic subwalks between these occurrences each have length at least h. Since their lengths sum to the minimum possible value 2 h , both have length h. The walk therefore stays in the core and the two subwalks are x–y geodesics. No other exceptional center lies on either geodesic because U has minimum separation at least h. Along such a walk, the four traversals incident with x or y have weight b in A x y , two have weight b and two have weight a in each of A x and A y , and all four have weight a in A 0 . The remaining 2 h − 4 traversals have weight a in all four matrices. Thus each based walk has inclusion–exclusion weight
a 2 h − 4 b 4 − 2 a 2 b 2 + a 4 = a 2 h − 4 ( b 2 − a 2 ) 2 .
There are g h ( x , y ) 2 ordered choices of the two geodesics and 2 h choices of the starting position on the resulting closed walk. This parametrization is bijective. Viewed cyclically, every such minimum-length walk visits x and y exactly once, so the based walk determines the starting position and the ordered pair of x–y geodesics. Hence the mixed contribution associated with the unordered pair { x , y } is
2 h a 2 h − 4 ( b 2 − a 2 ) 2 g h ( x , y ) 2 .
Summing over the pairs at distance h gives 2 h a 2 h − 4 ( b 2 − a 2 ) 2 Γ h ( S ) . For U = T there is no pair at distance h, since δ ( T ) > h , so its mixed contribution at order 2 h is zero. The positivity criterion follows because at least one pair of S is at distance h and every such pair has at least one geodesic. □
For minimum separations greater than 2, the condition a ≠ 0 cannot in general be omitted from the minimum-separation recovery statement. Write C n = v 0 v 1 … v n − 1 v 0 . On C 12 with m = q = 1 , take φ ( x , y ) = | x − 3 | + | y − 3 | . Then a = 0 and b = 1 . For two exceptional vertices at distance at least 3, the core edges with nonzero weight are exactly the four edges incident with those vertices, and their ordinary endpoints are distinct. All remaining ordinary–ordinary core edges have weight zero, while the pendant weights depend only on whether a core vertex is exceptional. Hence the weighted matrices are permutation similar for all such placements. In particular, H { v 0 , v 3 } and H { v 0 , v 5 } are cospectral although their minimum separations are 3 and 5. If b 2 = a 2 , the displayed critical-moment difference vanishes directly.
The theorem gives an exact critical-moment comparison rather than a claim that the use of perturbations, spectral moments, or closed walks is new. In a specific unweighted sun-like family, Koval and Kwan [41, Claims 4.7–4.8] use spectral moments to determine the minimum hub separation and the number of hub pairs attaining it. They also note that adjacency moments give the direct closed-walk formulation [41, Remark 2.4]. The displayed formula gives the exact contribution at the first possible mixed moment for the degree-dependent thorn perturbations considered here. The ordinary Sombor weight satisfies a ≠ 0 and b 2 ≠ a 2 . The Randić weight φ ( x , y ) = ( x y ) − 1 / 2 satisfies the same condition, so the minimum-separation conclusion also applies to that decreasing weight. The following refinement gives the critical term for a single placement.
For an independent exceptional set U and an integer h ≥ 2 , write
N h ( U ) = { { x , y } ⊆ U : d G ( x , y ) = h } .
Corollary 4.
Let G be a connected walk-regular graph of degree r, let m , q ≥ 1 , and let φ be a real symmetric weight. Put
D = r + m , D ′ = r + m + q , a = φ ( D , D ) , b = φ ( D ′ , D ) .
Fix k ≥ 2 and 2 ≤ h ≤ diam ( G ) , choose any v ∈ V ( G ) , and set
B h , k = k tr A φ ( H { v } ) 2 h − ( k − 1 ) tr A φ ( G ( m ) ) 2 h .
By Lemma 1, B h , k is independent of the choice of v. Every independent k-set U with δ ( U ) ≥ h satisfies
tr A φ ( H U ) 2 h = B h , k + 2 h a 2 h − 4 ( b 2 − a 2 ) 2 Γ h ( U ) .
If G is distance-regular and p h denotes the common number of geodesics between two vertices at distance h, this becomes
tr A φ ( H U ) 2 h = B h , k + η h N h ( U ) , η h = 2 h p h 2 a 2 h − 4 ( b 2 − a 2 ) 2 .
As above, when h = 2 the factor a 2 h − 4 is 1. Thus η h ≥ 0 , and it is positive exactly when b 2 ≠ a 2 and either h = 2 or a ≠ 0 . In particular, if G is distance-regular, a ≠ 0 , and b 2 ≠ a 2 , then for fixed G , m , q , φ and k, the weighted spectrum determines both the minimum separation of an independent exceptional set and the number of exceptional pairs attaining that distance.
Proof. 
At order 2 h , the contribution of words using no perturbation label is tr ( A φ ( G ( m ) ) 2 h ) . By Lemma 1, each of the k single-label contributions equals
tr A φ ( H { v } ) 2 h − tr A φ ( G ( m ) ) 2 h ,
independently of the center v. The zero blocks appended so that all matrices are indexed by the same vertex set do not affect this positive-order moment, so the total contribution from words using at most one perturbation label is exactly B h , k .
If δ ( U ) ≥ h , the two-label bound and the three-center argument in the proof of Theorem 3 show that every mixed contribution at order 2 h comes from exactly two labels whose centers are at distance h. For each unordered pair { x , y } at distance h, the pairwise inclusion–exclusion computation in the proof of Theorem 3 gives the mixed contribution
2 h a 2 h − 4 ( b 2 − a 2 ) 2 g h ( x , y ) 2 .
Summing over all such pairs gives the first formula. If G is distance-regular, then it is walk-regular. Let c i denote its backward intersection number at distance i, so every vertex at distance i from x has exactly c i neighbors at distance i − 1 from x. Starting at y and moving one distance layer closer to x at each step gives g h ( x , y ) = c h c h − 1 … c 1 , so this number depends only on h and equals p h . This gives the second formula. The final assertion follows from Theorem 3, which recovers the minimum separation, and from the second formula with h equal to that minimum. □
Corollary 5.
Let G be a connected walk-regular graph of degree r, let m , q ≥ 1 , and let S , T ⊆ V ( G ) be independent exceptional sets with | S | = | T | ≥ 2 and δ ( S ) ≠ δ ( T ) . If φ is either the ordinary Sombor weight or the Randić weight φ R ( x , y ) = ( x y ) − 1 / 2 , then A φ ( H S ) and A φ ( H T ) are not cospectral. Thus the corresponding weighted spectrum determines the minimum separation of the exceptional set, although the associated scalar index is the same for all independent exceptional sets of that cardinality.
Proof. 
Put D = r + m , D ′ = r + m + q , a = φ ( D , D ) , and b = φ ( D ′ , D ) . Since D ′ > D > 0 , for the ordinary Sombor weight,
a 2 = 2 D 2 , b 2 = D ′ 2 + D 2 ,
so a ≠ 0 and b 2 ≠ a 2 . For the Randić weight,
a 2 = D − 2 , b 2 = ( D D ′ ) − 1 ,
and again a ≠ 0 and b 2 ≠ a 2 . After interchanging S and T if necessary, apply Theorem 3. □
Corollary 6.
Let G be a distance-transitive graph of degree r and diameter d ≥ 2 , and fix m , q ≥ 1 . For each 2 ≤ s ≤ d , let H s be obtained by making two core vertices at distance s exceptional. Assume φ ( r + m , r + m ) ≠ 0 and φ ( r + m + q , r + m ) 2 ≠ φ ( r + m , r + m ) 2 . Then the graphs H s for 2 ≤ s ≤ d have pairwise distinct φ-weighted spectra. For each fixed s, the isomorphism type is independent of the chosen pair. In particular, the conclusion holds for both the ordinary Sombor and Randić weights.
Proof. 
Every distance-transitive graph is vertex-transitive and hence walk-regular, so Theorem 3 applies. Distance transitivity makes any two ordered pairs at the same distance equivalent under an automorphism, and Proposition 2 gives isomorphic thorny graphs for the corresponding thorn vectors. Thus Theorem 3 shows that placements at distinct core distances at least 2 have distinct weighted spectra. □
Example 2.
Let G = Q 3 , take m = q = 1 , and use the ordinary Sombor weight. The core is cubic and distance-transitive. Choose the two exceptional vertices at distance 2 or at distance 3. In both cases the exceptional set is independent, so the scalar Sombor indices agree. Corollary 6 shows that the two Sombor spectra are distinct. In fact, Theorem 3 shows that the order-4 moment for distance 2 exceeds that for distance 3 by 1296. Thus the placement phenomenon is not specific to cycle cores.
The closest-pair refinement is already visible on a cycle with three exceptional vertices. On C 12 , still with m = q = 1 and the ordinary Sombor weight, let
S = { v 0 , v 2 , v 4 } , T = { v 0 , v 2 , v 6 } .
Both are independent 3-sets with minimum separation 2, so their scalar Sombor indices are equal. However, N 2 ( S ) = 2 and N 2 ( T ) = 1 . Corollary 4 therefore shows that their Sombor spectra are distinct. Here p 2 = 1 and η 2 = 196 , so the order-4 moment of H S exceeds that of H T by 196. Thus the critical moment can distinguish placements with the same minimum separation when the number of closest exceptional pairs differs.
Example 3.
The Sombor spectrum can distinguish placements with the same complete pairwise-distance multiset. On C 16 , take m = q = 1 , use the ordinary Sombor weight, and set
S = { v 0 , v 2 , v 4 , v 10 } , T = { v 0 , v 2 , v 6 , v 8 } .
Both are independent 4-sets, and both have pairwise-distance multiset { 2 , 2 , 4 , 6 , 6 , 8 } . The zero-label contributions agree because the common base matrix is fixed, and the one-label contributions agree by Lemma 1. The two-label contributions agree at every order up to 8, because a two-label subsystem on a cycle is determined up to permutation similarity by the cyclic distance between its centers, and the two placements have the same distance multiset.
No word of length at most 8 using all four labels can have nonzero trace. Suppose otherwise. Then some closed walk contributing to its trace visits all four exceptional centers. Choose one occurrence of each center on this walk and list them in cyclic order. The four intervening subwalks each have length at least 2. Equality at total length 8 would require all four successive center pairs to be at distance 2, whereas each of S and T contains only two pairs at distance 2. A three-label contribution can first occur at order 8. For three centers, choose one occurrence of each and list them in cyclic order. The three intervening subwalks then have total length at least the sum of the three pairwise core distances. The unique triple in S attaining sum 8 is { v 0 , v 2 , v 4 } , while every triple in T has pairwise-distance sum at least 12. For this triple, consider any length-8 closed walk visiting v 0 and v 4 , and choose one occurrence of each. The two cyclic subwalks between these occurrences each have length at least 4 and together have length 8, so both have length 4. Each is therefore a v 0 – v 4 geodesic. Since 4 < | V ( C 16 ) | / 2 , this geodesic is unique, namely v 0 v 1 v 2 v 3 v 4 . Hence, up to cyclic shift, the only length-8 closed tour is the walk along this path and back. Since v 0 occurs only once on the cyclic tour, its eight cyclic shifts are distinct based closed walks.
Here D = 3 and D ′ = 4 , so the Sombor weight gives a 2 = 18 and b 2 = 25 . For a fixed starting position on this tour, the all-three-label weight is b 8 , the sum of the three two-label weights is 2 a 2 b 6 + a 4 b 4 , the sum of the three one-label weights is 2 a 6 b 2 + a 4 b 4 , and the base weight is a 8 . These eight based walks therefore give the three-label inclusion–exclusion contribution
8 b 8 − 2 a 2 b 6 + 2 a 6 b 2 − a 8 = 8 ( b 2 − a 2 ) 2 ( b 4 − a 4 ) = 117992 .
At order 8, all contributions using at most two labels cancel between the two placements, and no other three- or four-label contribution occurs. The preceding length bounds also show that all spectral moments below order 8 agree, whereas
tr A φ ( H S ) 8 − tr A φ ( H T ) 8 = 117992 .
Thus the two Sombor spectra are distinct even though their complete pairwise-distance multisets agree. The complete pairwise-distance multiset therefore does not in general determine the higher spectral moments.

6.1. Cycle Specialization

For cycle cores with two exceptional vertices, the underlying graphs form a subfamily of the generalized sun graphs studied by Wu and Ma [38, Theorem 6]. Their work includes the Randić spectrum, another degree-based weighted adjacency spectrum. On the scalar side, the formula below specializes the two-level reassignment formula to a cycle with two exceptional vertices [42, Theorem 3].
Let n ≥ 4 and let C n have vertices v 0 , v 1 , … , v n − 1 in cyclic order. The cyclic distance is
d C n ( v i , v j ) = min { | i − j | , n − | i − j | } .
Fix integers m , q ≥ 1 . For 1 ≤ s ≤ ⌊ n / 2 ⌋ , let H n , s be the thorny graph obtained from C n by attaching m + q pendant vertices at v 0 and v s , and attaching m pendant vertices at every other core vertex. Thus d C n ( v 0 , v s ) = s . Put
D = 2 + m , D ′ = 2 + m + q , a = φ ( D , D ) , b = φ ( D ′ , D ) .
Proposition 5.
For the graphs H n , s above, the scalar index is
I φ ( H n , 1 ) = ( n − 3 ) φ ( D , D ) + 2 φ ( D , D ′ ) + φ ( D ′ , D ′ ) + ( n − 2 ) m φ ( D , 1 ) + 2 ( m + q ) φ ( D ′ , 1 ) ,
and, for 2 ≤ s ≤ ⌊ n / 2 ⌋ ,
I φ ( H n , s ) = ( n − 4 ) φ ( D , D ) + 4 φ ( D , D ′ ) + ( n − 2 ) m φ ( D , 1 ) + 2 ( m + q ) φ ( D ′ , 1 ) .
Thus the scalar index does not distinguish the different nonadjacent distances s.
Proof. 
It suffices to count the core edges and the pendant edges. If s = 1 , the two exceptional vertices are adjacent, so the cycle has one core edge of type ( D ′ , D ′ ) , two of type ( D , D ′ ) , and n − 3 of type ( D , D ) . If s ≥ 2 , there is no core edge of type ( D ′ , D ′ ) , and the two exceptional vertices are incident with four core edges of type ( D , D ′ ) in total. The remaining n − 4 core edges have type ( D , D ) . The pendant contribution is the same in all cases. Summing these terms gives the two formulas. □
Two exceptional pairs at the same cyclic distance lie in the same dihedral orbit of C n . Hence Proposition 2 shows that the isomorphism type of H n , s depends only on s.
Corollary 7.
If
2 ≤ s < t ≤ n 2 ,
then
tr A φ ( H n , s ) 2 s − tr A φ ( H n , t ) 2 s = 2 s a 2 s − 4 ( b 2 − a 2 ) 2 .
When s = 2 , the factor a 2 s − 4 is 1. Thus the two matrices are not cospectral whenever b 2 ≠ a 2 and either s = 2 or a ≠ 0 .
Proof. 
Apply Theorem 3. Because s < t ≤ ⌊ n / 2 ⌋ , one has s < n / 2 , so the two exceptional vertices at distance s are joined by a unique geodesic in C n . □
Corollary 8.
For either the ordinary Sombor weight or the Randić weight, the graphs H n , s with 2 ≤ s ≤ ⌊ n / 2 ⌋ have pairwise distinct weighted spectra. Thus, within this family, the full weighted spectrum determines the nonadjacent cyclic distance between the two exceptional core vertices, whereas Proposition 5 shows that the corresponding scalar index does not.
Proof. 
This is the cycle case of Corollary 5. □

7. Discussion

The block reduction in Theorem 1 specializes the generalized Fiedler block-coupling theorem of Cardoso et al. [34, Theorem 3]. The connected positive-weight non-uniform star case is also covered by Rojo and Medina [32, Theorems 4.2 and 4.6], with disconnected positive-weight cores handled componentwise. Uniform positive-weight attachments fall under Cardoso et al. [35, Theorem 2.1]. These are existing matrix principles. The present reduction makes their application explicit for arbitrary cores and real symmetric degree-dependent weights, without a positivity restriction on the induced weights. The spectral consequences and recovery statements are given above with their respective hypotheses.
The placement results address a different question: how exceptional thorn multiplicities affect the weighted spectrum on connected walk-regular cores. Independent exceptional sets of the same size have equal moments below twice the smaller minimum separation for any real symmetric weight. When their separations differ, the first possible moment difference is expressed through squared geodesic counts. This difference is strictly positive if b 2 ≠ a 2 and either the smaller minimum separation is 2 or a ≠ 0 . When the critical-moment coefficient vanishes, this comparison alone does not establish whether the full weighted spectra agree. Example 3 shows that identical pairwise-distance multisets need not imply identical higher moments. No complete reconstruction of an arbitrary exceptional placement is asserted.

8. Conclusions

For thorny graphs with positive pendant multiplicities, an explicit reduction of the degree-based weighted adjacency matrix yields a nullity bound, exact inertia for nonzero pendant couplings, interlacing, and a compression whose dimension depends only on the core size. For fixed r ≥ 1 , m, and φ , the nonzero weighted spectrum of a uniform attachment to an r-regular core determines the core adjacency spectrum whenever φ ( r + m , r + m ) ≠ 0 . For independent exceptional sets of equal cardinality on connected walk-regular cores, the first possible difference between spectral moments at unequal minimum separations is governed by a squared-geodesic statistic. For fixed G , m , q , φ and exceptional-set cardinality, the condition a ≠ 0 and b 2 ≠ a 2 makes the weighted spectrum determine the minimum separation and, on distance-regular cores, the number of closest exceptional pairs. Both the ordinary Sombor and Randić weights satisfy this condition. Future work may examine higher moments of placements with the same minimum separation and critical geodesic statistic. Another question is when the weighted spectrum recovers the full pairwise-distance multiset, or when distinct placements remain cospectral despite sharing the invariants identified here.

Author Contributions

Xing-Yu Hu is the sole author and is responsible for all aspects of the work.

Funding

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Data Availability Statement

Not applicable. No datasets were generated or analyzed in this theoretical study.

Conflicts of Interest

The author declares no competing interests.

References

  1. Gutman, I. Geometric approach to degree-based topological indices: Sombor indices. MATCH Commun. Math. Comput. Chem. 2021, 86, 11–16. [Google Scholar]
  2. Cruz, R.; Gutman, I.; Rada, J. Sombor index of chemical graphs. Appl. Math. Comput. 2021, 399, 126018. [Google Scholar] [CrossRef]
  3. Rada, J.; Rodriguez, J. M.; Sigarreta, J. M. General properties on Sombor indices. Discret. Appl. Math. 2021, 299, 87–97. [Google Scholar] [CrossRef]
  4. Liu, H.; Gutman, I.; You, L.; Huang, Y. Sombor index: review of extremal results and bounds. J. Math. Chem. 2022, 60, 771–798. [Google Scholar] [CrossRef]
  5. Gowtham, K. J.; Husin, M. N.; Siddiqui, M. K. Some new bounds on the modified symmetric division deg index. Malays. J. Math. Sci. 2024, 18(1), 39–50. [Google Scholar] [CrossRef]
  6. Gowtham, K. J. New vertex-edge Sombor, Nirmala and Misbalance indices. TWMS J. Appl. Eng. Math. 2024, 14(2), 597–604. [Google Scholar]
  7. Gowtham, K. J. Entropy measures of some nanotubes using Sombor index. NanoNEXT 2022, 3(3), 1–5. [Google Scholar] [CrossRef]
  8. Gowtham, K. J. On reverse Laplacian energy of a graph. Lett. Appl. NanoBioScience 2023, 12(1), 19. [Google Scholar] [CrossRef]
  9. Gowtham, K. J. Reverse Hyper-Zagreb indices of the Cartesian product of graphs. Int. J. Math. Comb. 2021, 4, 49–56. [Google Scholar]
  10. Jayanna, G. K.; Swamy, N. N.; Sooryanarayana, B.; Cangul, I. N. Some properties of the novel topological graph index atom-bond sum-connectivity index. Montes Taurus J. Pure Appl. Math. 2024, 6(3), 418–427. [Google Scholar]
  11. Gutman, I. Spectrum and energy of the Sombor matrix. Vojnoteh. Glas. Tech. Cour. 2021, 69, 551–561. [Google Scholar] [CrossRef]
  12. Jayanna, G. K.; Gutman, I. On characteristic polynomial and energy of Sombor matrix. Open J. Discret. Appl. Math. 2021, 4, 29–35. [Google Scholar] [CrossRef]
  13. Ghanbari, N. On the Sombor characteristic polynomial and Sombor energy of a graph. Comput. Appl. Math. 2022, 41, 242. [Google Scholar] [CrossRef]
  14. Lin, Z.; Zhou, T.; Miao, L. On the spectral radius, energy and Estrada index of the Sombor matrix of graphs. Trans. Comb. 2023, 12, 191–205. [Google Scholar] [CrossRef]
  15. Pirzada, S.; Rather, B. A.; Das, K. C.; Shang, Y.; Gutman, I. On spectrum of Sombor matrix and Sombor energy of graphs. Georgian Math. J. 2025, 32, 649–660. [Google Scholar] [CrossRef]
  16. Rather, B. A.; Imran, M.; Raza, Z.; Farooq, F. B. Spectral properties of Sombor matrix of threshold graphs. RAIRO Oper. Res. 2024, 58, 2845–2864. [Google Scholar] [CrossRef]
  17. Swamy, N. N.; Jayanna, G. K.; Sooryanarayana, B. Reverse Sombor energy of a graph. Biointerface Res. Appl. Chem. 2024, 14(3), 61. [Google Scholar] [CrossRef]
  18. Das, K. C.; Gutman, I.; Milovanović, I.; Milovanović, E.; Furtula, B. Degree-based energies of graphs. Linear Algebra Appl. 2018, 554, 185–204. [Google Scholar] [CrossRef]
  19. Liu, H.; You, L.; Huang, Y.; Fang, X. Spectral properties of p-Sombor matrices and beyond. MATCH Commun. Math. Comput. Chem. 2022, 87, 59–87. [Google Scholar] [CrossRef]
  20. Li, X. L.; Yang, N. Spectral properties and energy of weighted adjacency matrices for graphs with degree-based edge-weight functions. Acta Math. Sin. (Engl. Ser.) 2024, 40, 3027–3042. [Google Scholar] [CrossRef]
  21. Li, X.; Yang, N. Unified approach for spectral properties of weighted adjacency matrices for graphs with degree-based edge-weights. Linear Algebra Appl. 2024, 696, 46–67. [Google Scholar] [CrossRef]
  22. Li, X.; Yang, N. Some interlacing results on weighted adjacency matrices of graphs with degree-based edge-weights. Discret. Appl. Math. 2023, 333, 110–120. [Google Scholar] [CrossRef]
  23. Gao, J.; Li, X.; Yang, N.; Zheng, R. The Li–Feng transformation of weighted adjacency matrices for graphs with degree-based edge-weights. Discret. Math. 2025, 348(9), 114520. [Google Scholar] [CrossRef]
  24. Rather, B. A.; Ganie, H. A. Degree-based weighted adjacency matrices: spectra, integrality, and edge deletion effects. arXiv 2026, arXiv:2603.08895. [Google Scholar] [CrossRef]
  25. Ramane, H. S.; Kitturmath, D. V.; Bhajantri, K. Sombor polynomial and Nirmala polynomial of some graph transformations of regular graphs. J. Adv. Math. Stud. 2022, 15, 310–320. [Google Scholar]
  26. Gutman, I. Distance of thorny graphs. Publ. Inst. Math. (Beograd) 1998, 63(77), 31–36. [Google Scholar] [CrossRef]
  27. Frucht, R.; Harary, F. On the corona of two graphs. Aequationes Math. 1970, 4, 322–325. [Google Scholar] [CrossRef]
  28. Godsil, C. D.; McKay, B. D. A new graph product and its spectrum. Bull. Austral. Math. Soc. 1978, 18(1), 21–28. [Google Scholar] [CrossRef]
  29. Barik, S.; Pati, S.; Sarma, B. K. The spectrum of the corona of two graphs. SIAM J. Discret. Math. 2007, 21(1), 47–56. [Google Scholar] [CrossRef]
  30. Fiuj Laali, A. R.; Haj Seyyed Javadi, H.; Kiani, D. Spectra of generalized corona of graphs. Linear Algebra Appl. 2016, 493, 411–425. [Google Scholar] [CrossRef]
  31. Dai, M.; Shen, J.; Dai, L.; Ju, T.; Hou, Y.; Su, W. Generalized adjacency and Laplacian spectra of the weighted corona graphs. Physica A 2019, 528, 121285. [Google Scholar] [CrossRef]
  32. Rojo, O.; Medina, L. Spectra of weighted compound graphs of generalized Bethe trees. Electron. J. Linear Algebra 2009, 18, 30–57. [Google Scholar] [CrossRef]
  33. Cardoso, D. M.; Gutman, I.; Martins, E. A.; Robbiano, M. A generalization of Fiedler’s lemma and some applications. Linear Multilinear Algebra 2011, 59(8), 929–942. [Google Scholar] [CrossRef]
  34. Cardoso, D. M.; de Freitas, M. A. A.; Martins, E. A.; Robbiano, M. Spectra of graphs obtained by a generalization of the join graph operation. Discret. Math. 2013, 313, 733–741. [Google Scholar] [CrossRef]
  35. Cardoso, D. M.; Martins, E. A.; Robbiano, M.; Rojo, O. Copies of a rooted weighted graph attached to an arbitrary weighted graph and applications. Electron. J. Linear Algebra 2013, 26, 706–717. [Google Scholar] [CrossRef]
  36. Boulet, R. Spectral characterizations of sun graphs and broken sun graphs. Discret. Math. Theor. Comput. Sci. 2009, 11(2), 149–160. [Google Scholar] [CrossRef]
  37. Braga, R. O.; Rodrigues, V. M.; Silva, R. O. Locating eigenvalues of a symmetric matrix whose graph is unicyclic. Trends Comput. Appl. Math. 2021, 22(4), 659–674. [Google Scholar] [CrossRef]
  38. Wu, Y.; Ma, X. The spectra of sun graph and their energy. J. Xinjiang Univ. (Natural Science Edition in Chinese and English) 2023, 40(5), 550–559. [Google Scholar] [CrossRef]
  39. Godsil, C. D.; McKay, B. D. Feasibility conditions for the existence of walk-regular graphs. Linear Algebra Appl. 1980, 30, 51–61. [Google Scholar] [CrossRef]
  40. Dalfó, C.; van Dam, E. R.; Fiol, M. A. On perturbations of almost distance-regular graphs. Linear Algebra Appl. 2011, 435(10), 2626–2638. [Google Scholar] [CrossRef]
  41. Koval, I.; Kwan, M. Exponentially many graphs are determined by their spectrum. Q. J. Math. 2024, 75(3), 869–899. [Google Scholar] [CrossRef]
  42. Hu, X.-Y. Cut decomposition and induced-edge statistics for degree-based edge indices under thorn reassignments on regular cores. Prepr.>, Version 1 (2026). [CrossRef]
  43. Harary, F. Graph Theory; Addison-Wesley: Reading, 1969. [Google Scholar]
  44. Gutman, I. Sombor index of thorny graphs. Vojnoteh. Glas. Tech. Cour. 2025, 73, 413–422. [Google Scholar] [CrossRef]
  45. Shinde, S. S.; Ramane, H. S.; Gudimani, S. B.; Swamy, N. Degree sum adjacency polynomial of standard graphs and graph operations. Electron. J. Graph Theory Appl. 2022, 10(1), 15–32. [Google Scholar] [CrossRef]
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.
Copyright: This open access article is published under a Creative Commons CC BY 4.0 license, which permit the free download, distribution, and reuse, provided that the author and preprint are cited in any reuse.