Connected graphs with large multiplicity of −1 in the

spectrum of the eccentricity matrix

Xinghui Zhao, Lihua You*

School of Mathematical Sciences, South China Normal University, Guangzhou, 510631,

P. R. China, e-mail: [email protected], [email protected].
∗
Corresponding author
arXiv:2409.07198v1 [math.CO] 11 Sep 2024

Abstract The eccentricity matrix of a simple connected graph is obtained from the dis-
tance matrix by only keeping the largest distances for each row and each column, whereas
the remaining entries become zero. This matrix is also called the anti-adjacency matrix,
since the adjacency matrix can also be obtained from the distance matrix but this time
by keeping only the entries equal to 1. It is known that, for λ ̸∈ {−1, 0} and a fixed
i ∈ N, there is only a finite number of graphs with n vertices having λ as an eigenvalue of
multiplicity n − i on the spectrum of the adjacency matrix. This phenomenon motivates
researchers to consider the graphs has a large multiplicity of an eigenvalue in the spectrum
of the eccentricity matrix, for example, the eigenvalue −2 [X. Gao, Z. Stanić, J.F. Wang,
Grahps with large multiplicity of −2 in the spectrum of the eccentricity matrix, Discrete
Mathematics, 347 (2024) 114038]. In this paper, we characterize the connected graphs
with n vertices having −1 as an eigenvalue of multiplicity n − i (i ≤ 5) in the spectrum
of the eccentricity matrix. Our results also become meaningful in the framework of the
median eigenvalue problem [B. Mohar, Median eigenvalues and the HOMO-LUMO index
of graphs, Journal of Combinatorial Theory Series B, 112 (2015) 78-92].

Keywords: Eccentricity matrix; Rank; Median eigenvalue; Multiplicity; Spectrum

1 Introduction
Throughout this paper we consider the graphs as simple and connected graphs. Let
G = (V (G), E(G)) be a simple graph with vertex set V (G) = {v1 , · · · , vn } and edge set
E(G). If the vertices vi and vj are adjacent, we write vi ∼ vj and vi vj ̸∼ E(G) otherwise.
The set of the neighborhood of vi in G is denoted by NG (vi ) (for short N (vi )). Then the
degree of the vertex v is equal |N (v)| (the cardinality of the set N (v)), denoted by dG (v),
and △(G) = max{dG (v)|v ∈ V (G)}. As usual, Kn , Cn , Pn is the complete graph, cycle,
path with order n, respectively. Let G ∪ H be the disjoint union of graphs G and H. The
join of two disjoint graphs G and H, denoted by G ∨ H, is the graph obtained by joining
each vertex of G to each vertex of H. The complement of graph G is denoted by G.

1
 The distance dG (u, v) (for short d(u, v)) between two vertices u and v is the minimum
length of the paths joining them. The diameter of G, denoted by diam(G), is the greatest
distance between any two vertices in G. The eccentricity εG (v) (for short ε(v)) of the
vertex v ∈ V (G) is given by εG (v) = max{d(u, v)|u ∈ V (G)}. Then the eccentricity
matrix A(G) = (ϵuv (G)) of G is defined as follows:
(
d(u, v), if d(u, v) = min{εG (u), εG (v)},
ϵuv (G) =
0, others.

Let M be an n × n matrix, P (M, λ) be characteristic polynomial of M , ξ1 (M ) ≥

ξ2 (M ) ≥ · · · ≥ ξn (M ) be roots of P (M, λ) = 0. We also call ξ1 (M ), ξ2 (M ), · · · , ξn (M )
the eigenvalues of M . The spectrum of M is a multiset of all eigenvalues of M , denoted
by Spec(M ). The multiplicity of an eigenvalue ξ in the Spec(M ) is denoted by mM (ξ).
Since the eccentricity matrix A(G) is a real symmetric matrix, all its eigenvalues,
called A-eigenvalues of G, are real. Then we can arrange the A-eigenvalues of G as
ξ1 ≥ ξ2 ≥ · · · ≥ ξn . The A-spectrum of G is a multiset consisting of the A-eigenvalues,
′ ′ ′
denoted by SpecA (G). If ξ1 > ξ2 > · · · > ξk are the pairwise distinct A-eigenvalues of G,
then the A-spectrum of G can be written as
 ′ ′ ′ 
ξ1 ξ2 . . . ξk
SpecA (G) = {ξ1 , ξ2 , . . . , ξn } = ,
m1 m2 . . . mk
′ ′
where mi = mA(G) (ξi ) is the algebraic multiplicity of the eigenvalue ξi (1 ≤ i ≤ k).
Other symbols not introduced in this paper can refer to [1] and [4].
Clearly, the matrix A(G) is obtained from the distance matrix by only keeping the
largest distances for each row and each column, whereas the remaining entries become 0.
That is why A(G) can be interpreted as the anti-adjacency matrix, where the adjacency
matrix obtained from the distance matrix by keeping only distances equal to 1 on each
row and each column. From this point of view, A(G) and A(G) are extremal among all
possible distance-like matrices. As a contrast with A(G), the eccentricity matrix has some
fantastic properties, one of which is that A(G) of a connected graph is not necessarily
irreducible, see the published [13, 14, 17].
In the last few decades, the eccentricity matrices of graphs inspired much interest
and attracted the attention of many researchers, which caused more than forty papers
published. For example, Wang et al. [17] showed that the eccentricity matrix of trees is
irreducible, Wei, He and Li [18] determined the n-vertex trees with minimum A-spectral

2
radius and all trees with given order and diameter having minimizing A-spectral ra-
dius, which solved two conjectures in [17], Wang et al. [16] studied the A-spectral radius
properties of graphs, Lei, Wang and Li [7] characterized the graphs whose second least
p √
A-eigenvalue is greater than − 15 − 193, Mahato and Kannan [10] studied the inertia
and spectral symmetry of the eccentricity matrix of trees, Wang et al. [15] characterized
all connected graphs with exactly one positive A-eigenvalue and so on.
Here is the motivation for the research reported in this paper. An experienced reader
knows that the eigenvalues 0 and −1 can attain a very large multiplicity in the spectrum
of the adjacency matrix of some connected graphs. For example, the multiplicity of −1 in
the spectrum of a complete graph Kn is n − 1, while the multiplicity of 0 in the spectrum
of a complete bipartite graph (of order n) is n − 2. Recently, Gao, Stanić and Wang [6]
characterized graphs with mA(G) (−2) = n − i with i ≤ 5. This motivates us to consider
the graphs for which a particular eigenvalue has a large multiplicity in the spectrum of
the eccentricity matrix.
In this paper, we will characterize graphs with mA(G) (−1) = n − i with i ≤ 5, and
obtain the following results, which will become meaningful in the framework of the median
eigenvalue problem [11, 12] since, for sufficiently large n, the median eigenvalue of the
obtained graphs is always −1.
For convenience, we use m(ξ) to denoted mA(G) (ξ) in the following.

Theorem 1.1 Let G be a connected graph with order n, G1 = {Kn−4 ∨4K1 , Kn−4 ∨(2K1 ∪
K2 ), Kn−4 ∨ (P3 ∪ K1 ), Kn−4 ∨ 2K2 , Kn−4 ∨ P4 , Kn−4 ∨ (K3 ∪ K1 ), Kn−4 ∨ C4 }. Then we
have

(i) m(−1) = n − 1 if and only if G ∼

= Kn ;

(ii) m(−1) ̸= n − 2;

(iii) m(−1) = n − 3 for n ≥ 4 if and only if G ∈ {Kn−2 ∨ 2K1 , P4 };

(iv) m(−1) = n − 4 for n ≥ 9 if and only if G ∈ {Kn−3 ∨ (K2 ∪ K1 ), Kn−3 ∨ 3K1 };

(v) m(−1) = n − 5 for n ≥ 16 if and only if G ∈ G1 ∪ {Kn−5 ∨ C5 , Kn−5 ∨ (K1 ∪ P4 ), Kn−5 ∨

H1 }, where H1 is illustrated in Figure 1.

3
 Figure 1: H1 .

2 Preliminaries
In this section, we recall some definitions and results which will be used in next
section.

Definition 2.1 ([2]) Let M be a complex matrix of order n described in the following
block form  
M11 · · · M1l
M =  ... .. .. 

. . 
Ml1 · · · Mll
where the blocks Mij are ni × nj matrices for any 1 ≤ i, j ≤ l and n = n1 + · · · + nl .
For 1 ≤ i, j ≤ l, let qij denote the average row sum of Mij , i.e. qij is the sum of all
entries in Mij divided by the number of rows. Then Q(M ) = (qij ) (or simply Q) is called
the quotient matrix of M . If, in addition, for each pair i, j, Mij has a constant row
sum, i.e., Mij ⃗enj = qij ⃗eni , then Q is called the equitable quotient matrix of M , where
⃗ek = (1, 1, · · · , 1)T ∈ C k , and C denotes the field of complex numbers.

Lemma 2.1 ([20]) Let M be defined as Definition 2.1 such that Mij = sij Jni ,nj for i ̸= j,
and Mii = sii Jni ,ni + pi Ini . Then the equitable quotient matrix of M is Q = (qij ) with
qij = sij nj if i ̸= j, and qii = sii ni + pi for 1 ≤ i, j ≤ l. Moreover,
 
p1 ··· pl
Spec(M ) = Spec(Q) ∪ .
n1 − 1 · · · nl − 1

Lemma 2.2 ([3]) Let M be an n×n Hermitian matrix, M ∗ be its k×k principal submatrix,
and ξi (M ) be the i-th largest eigenvalue of M . Then for i ∈ {1, 2, ..., k}, we have

ξi (M ) ≥ ξi (M ∗ ) ≥ ξn−k+i (M ).

A graph is H-free if it does not contain H as an induced subgraph, where H can be a

graph or a set of graphs.

Lemma 2.3 ([9]) A graph is a complete multipartite graph if and only if it is P3 -free.

4
Lemma 2.4 ([8]) Let G be a connected graph with order n ≥ 3. Then G ∼
= Kl ∨ (Ks ∪ Kt )
if and only if G is {K1,3 , P4 , C4 }-free, where s, t ≥ 0 and l ≥ 1.

Lemma 2.5 ([6]) The set {v1 , v2 , · · · , vk } of either duplicate or co-duplicate vertices of
a graph G gives rise to an eigenvalue ξ of multiplicity at least k − 1 in the spectrum of
A(G), along with the following:

−2, if ε(v1 ) = 2


and v1 , · · · , vk are duplicate;
0, if ε(v1 ) ̸= 2 and v1 , · · · , vk are duplicate;
ξ=


 −1, if ε(v1 ) = 1 and v1 , · · · , vk are co-duplicate;
if ε(v1 ) ̸= 1 v1 , · · ·

0, and , vk are co-duplicate.
Let G be a graph with vertex set V (G) = {v1 , · · · , vn }, U1 , · · · , Un be mutually
nonempty finite sets, the graph H with vertex set U1 ∪ · · · ∪ Un as follows: for each
i, the vertices of Ui are either all mutually adjacent (Ui is a clique), or all mutually
nonadjacent(Ui is an independent set); for 1 ≤ i, j ≤ n and i ̸= j, each vertex of Ui
is adjacent to each vertex of Uj if and only if vi and vj are adjacent in G. We call H
is a mixed extension of G. Furthermore, we represent a mixed extension by an n-tuple
(t1 , ..., tn ) of nonzero integers, where ti > 0 indicates that Ui is a clique of order ti , and
ti < 0 means that Ui is an independent set of order −ti . For example, Sk+1 is a star with
the vertex v0 of degree k and the other vertices v1 , v2 , ..., vk , S(t0 , t1 , · · · , tk ) is the mixed
extension of the star Sk+1 of the type (t0 , t1 , · · · , tk ), where ti (0 ≤ i ≤ k) is an integer.

Lemma 2.6 ( [15]) Let G be a connected graph. Then G has exactly one positive A-
eigenvalue if and only if G ∼
= S(t0 , −p, t1 , · · · , tq ) with p, q ≥ 0, t0 ≥ 1 and ti ≥ 2(1 ≤ i ≤
q). Furthermore, m(−1) = t0 − 1.

Lemma 2.7 ([18]) Let G be a connected graph with order n. Then ξn (G) = −1 if and
only if G ∼
= Kn .

Lemma 2.8 ([16]) Let G ∼

= Kr ∨ Kn1 ,n2 ,··· ,nk be a complete (r + k)-partite graph with
n1 ≥ n2 ≥ · · · nk ≥ 2, r ≥ 1 and k ≥ 2. Then m(−1) = r − 1.

Lemma 2.9 ([19]) Let G ∼

= Kr ∨Kn1 ,n2 ,··· ,nk be a complete (r +k)-partite graph with order
Pk
n=r+ ni , n1 ≥ n2 ≥ · · · nk ≥ 2. Then
i=1
(i) if r, k ≥ 1, then m(−2) = n − r − k if and only if (r, k) ̸∈ {(1, 4), (2, 3)}. Furthermore,
if ns = · · · = ns+t−1 (s ≥ 1, s + t − 1 ≤ k), then m(2ns − 2) = t − 1.
 
−2 2n1 − 2 · · · 2nk − 2
(ii) if r = 0, k ≥ 2, then SpecA (G) = .
n−k 1 ··· 1

5
3 The proof of Theorem 1.1
In this section, we will prove Theorem 1.1.

Lemma
( 3.1 Let A = (aij )n×n be a real symmetric matrix, a(≥ 2) be an integer, aij =
sij a, i ̸= j;
where sij ∈ {0, 1}. Then rank(A) = n.
1, i = j,

Proof. Firstly, we transform A to a row echelon form by Gauss transformation, which

takes n − 1 steps.
Step 1: Add −ai1 times thefirst row to the i-th row of A for 2 ≤ i ≤ n, and we obtain

1, i = j = 1;
1 − s(1) a, 2 ≤ i = j ≤ n;

(1) (1) ii (1)
A1 = (aij )n×n , where aij = sii = s21i a for 2 ≤ i ≤ n,


0, 2 ≤ i ≤ n, j = 1;
 (1)
sij a, i ̸= j, 2 ≤ i, j ≤ n,
(1)
sij = sij − si1 s1j a for i ̸= j and 2 ≤ i, j ≤ n.
(1) (1) (1)
Clearly, for 2 ≤ i ≤ n, aii = 1 − sii a ̸= 0 by sii is an integer and a ≥ 2.
(1)
si2 a
Step 2: Add − times the second row to the i-th row of A1 for 3 ≤ i ≤ n,
(1)
1−s22 a 
(1)


 aij , i ≤ 2 or j = 1;

0,

 3 ≤ i ≤ n, j = 2;
(2) (2) (2) (2) (1)
and we obtain A2 = (aij )n×n , where aij = 1−s a
3 ≤ i = j ≤ n; s22 = s22 ,
(2) ,
ii

 1−s22 a
(2)

 s a
 ij(2) , i ̸= j, 3 ≤ i, j ≤ n,


1−s22 a
(2) (1) (1) (1) (1) (1) (1) (2) (1) (1) (1) (1) (1)
sii = si2 s2i a + s22 + sii − sii s22 a for 3 ≤ i ≤ n, sij = sij − sij s22 a − si2 s2j a for
i ̸= j and 3 ≤ i, j ≤ n.
(2)
(2) 1−sii a (2)
Clearly, for 3 ≤ i ≤ n, aii = (2) ̸= 0 by sii is an integer and a ≥ 2.
1−s22 a
(2)
si3 a
Step 3: Add − times the third row to the i-th row of A2 for 4 ≤ i ≤ n, and
(2)
1−s33 a 
(2)


 aij , i ≤ 3 or j ≤ 2;

0,

 4 ≤ i ≤ n, j = 3;
(3) (3) (3) (3) (2) (2)
we obtain A3 = (aij )n×n , where aij = 1−sii a , 4 ≤ i = j ≤ n; s33 = s22 + s33 −
(3)

 1−s33 a
(3)

 s a
 ij(3) , i ̸= j, 4 ≤ i, j ≤ n,


1−s33 a
(2) (2) (3) (2) (2) (2) (2) (2) (2) (3) (2) (2) (2) (2) (2)
s22 s33 a, sii = si3 s3i a+s33 +sii −sii s33 a for 4 ≤ i ≤ n, sij = sij −sij s33 a−si3 s3j a
for i ̸= j and 4 ≤ i, j ≤ n.
(3)
(3) 1−sii a (3)
Clearly, for 4 ≤ i ≤ n, aii = (3) ̸= 0 by sii is an integer and a ≥ 2.
1−s33 a
(n−1)
Continuing the above steps, we get row echelon form An−1 = (aij )n×n of A. Clearly,
(
(n−1) 1, i = 1; (n−1)
we have aii = (i−1) which implies aii ̸= 0 for 1 ≤ i ≤ n. Thus
aii , 2 ≤ i ≤ n,

6
rank(A) = rank(An−1 ) = n. ■

Lemma 3.2 Let M be an n×n real symmetry matrix, M ∗ be its k×k principal submatrix,
mM ∗ (ξ) = s with s ≥ 0. Then mM (ξ) ≤ n − k + s.

Proof. We will complete the proof by the following three cases.

Case 1: mM (ξ) = 0.
Clearly, mM (ξ) ≤ n − k + s by 0 ≤ s ≤ k ≤ n.
Case 2: mM (ξ) > 0 and s ≥ 1.
Without loss of generality, we take ξt+1 (M ∗ ) = · · · = ξt+s (M ∗ ) = ξ for 0 ≤ t ≤ k − s.
If 1 ≤ t ≤ k − s − 1, then by Lemma 2.2, we have ξ1 (M ) ≥ ξ2 (M ) ≥ · · · ≥ ξt (M ) ≥
ξt (M ∗ ) > ξ > ξt+s+1 (M ∗ ) ≥ ξn−k+t+s+1 (M ) ≥ · · · ≥ ξn−1 (M ) ≥ ξn (M ), which implies
mM (ξ) ≤ n − t − [n − (n − k + t + s + 1) + 1] = n − k + s.
If t = 0 or t = k − s, similar to the above proof, we have mM (ξ) ≤ n − k + s.
Case 3: mM (ξ) > 0 and s = 0.
By s = 0, we have ξ > ξ1 (M ∗ ) or ξ1 (M ∗ ) > ξ > ξk (M ∗ ) or ξ < ξk (M ∗ ).
If ξ1 (M ∗ ) > ξ > ξk (M ∗ ), then there exists 1 ≤ l ≤ k − 1 satisfying ξl (M ∗ ) >
ξ > ξl+1 (M ∗ ). By Lemma 2.2, we have ξ1 (M ) ≥ ξ2 (M ) ≥ · · · ≥ ξl (M ) ≥ ξl (M ∗ ) >
ξ > ξl+1 (M ∗ ) ≥ ξn−k+l+1 (M ) ≥ · · · ≥ ξn−1 (M ) ≥ ξn (M ), which implies mM (ξ) ≤
n − l − [n − (n − k + l + 1) + 1] = n − k.
If ξ > ξ1 (M ∗ ) or ξ < ξk (M ∗ ), similar to the above proof, we have mM (ξ) ≤ n − k. ■

The following Proposition is very useful, but is so easy that we ignore its proof.

Proposition 3.3 Let G be a connected graph with diam(G) = d, Vi = {v ∈ V (G)|ε(v) =

i}. Then

(i) V (G) = V1 ∪ V2 ∪ · · · ∪ Vd is a partition of the vertex set of G;

(ii) if V1 ̸= ∅, then 1 ≤ d ≤ 2;

(iii) d = 1 if and only if G ∼

= Kn .

Lemma 3.4 Let G(̸∼

= Kn ) be a connected graph with order n, V1 ̸= ∅, m(−1) = n − k.
Then |V1 | = n − k + 1 or n − k.

Proof. Clearly, |V1 | ≤ n − k + 1 by V1 ̸= ∅, m(−1) = n − k and Lemma 2.5. Now we show

|V1 | ≥ n − k.

7
 By G ∼
̸= Kn , we have d ≥ 2, and d ≤ 2 by V1 ̸= ∅. Then d = 2 and V (G) = V1 ∪ V2 is
a partition of V (G).
If |V1 | ≤ n − k − 1, then |V2 | ≥ k + 1 by |V1 | + |V2 | = n. Let B be the principal
submatrix of A(G) + I indexed by V2 . Then rank(B) = |V2 | by Lemma 3.1, and thus
mB−I (−1) = mB (0) = |V2 | − rank(B) = 0 by matrix theory. Therefore, m(−1) ≤
n − |V2 | + mB−I (−1f ) = n − |V2 | ≤ n − k − 1 by Lemma 3.2, which implies a contradiction.
■

Lemma 3.5 Let G be a connected graph with order n and diam(G) = d. If V1 = ∅, and
n ≥ k(d − 1) + 1, then m(−1) ≤ n − k − 1.

Proof. By V1 = ∅ and diam(G) = d, we have V (G) = V2 ∪ V3 ∪ · · · ∪ Vd is a partition of

the vertex set of G. Clearly, there exists |Vi | ≥ k + 1 by n ≥ k(d − 1) + 1.
Let B be the principal submatrix of A(G) + I indexed by Vi . Then rank(B) = |Vi |
by Lemma 3.1, and thus mB−I (−1) = mB (0) = |Vi | − rank(B) = 0 by matrix theory.
Therefore, m(−1) ≤ n − |Vi | + mB−I (−1) = n − |Vi | ≤ n − k − 1 by Lemma 3.2. ■

Theorem 3.6 Let G be a connected graph with order n. Then m(−1) = n − 1 if and only
if G ∼
= Kn . In addition, G is determined by its A-spectrum.

Proof. Clearly, we have mA(Kn ) (−1) = n − 1. On the other hand, if m(−1) = n − 1, then
ξ2 = · · · = ξn = −1 and ξ1 = n − 1 by tr(A(G)) = 0, and thus G ∼
= Kn by Lemma 2.7.
Therefore, m(−1) = n − 1 if and only if G ∼
= Kn .
It is obvious that G is determined by its A-spectrum. ■

Theorem 3.7 There is no connected graph G with order n and m(−1) = n − 2.

Proof. Suppose there exists a connected graph G with order n satisfying m(−1) = n − 2,
then A(G) exactly has a positive A-eigenvalue or two positive A-eigenvalues.
If A(G) exactly has a positive A-eigenvalue, then G ∼
= S(t0 , −p, t1 , · · · , tq ) with p, q ≥
0, t0 ≥ 1, ti ≥ 2 for 1 ≤ i ≤ q and m(−1) = t0 −1 by Lemma 2.6. Thus we have t0 = n−1,
and G ∼
= Kn , which implies a contradiction by Theorem 3.6.
If A(G) exactly has two positive A-eigenvalues, then ξn = −1, and thus G ∼
= Kn by
Lemma 2.7, which implies m(−1) = n − 1, a contradiction. ■

Lemma 3.8 Let G be a connected graph with order n and diam(G) = d ≥ 4. Then
m(−1) ≤ n − 5.

8
Proof. Let d(v0 , vd ) = d and Pd+1 = v0 v1 v2 · · · vd be a path with length d. Then ε(v0 ) =
ε(vd ) = d, d − 1 ≤ ε(v1 ), ε(vd−1 ) ≤ d, d − 2 ≤ ε(v2 ) ≤ d. Clearly, the principal submatrix
of A(G) + I indexed by {v0 , v1 , v2 , vd−1 , vd } is
 
1 0 a b d
0 1 0 0 c
 
B1 =  a 0 1 0 e,
b 0 0 1 0
d c e 0 1
where a ∈ {0, 2}, b, c ∈ {0, d−1}, e ∈ {0, d−2}. By direct calculation, we have P (B1 , 0) =
−a2 c2 − b2 c2 − b2 e2 − 2ade + a2 + b2 + c2 + d2 + e2 − 1.
Case 1: ε(v2 ) = d − 2, ε(v1 ) = ε(vd−1 ) = d − 1.
Then b = c = d − 1, e = d − 2, thus P (B1 , 0) = −a2 d2 − 2d4 + 2a2 d − 2ad2 + 10d3 +
4ad − 15d2 + 8d < 0 by a ∈ {0, 2} and d ≥ 4.
Case 2: ε(v2 ) = d − 2, ε(v1 ) = d − 1, ε(vd−1 ) = d.
Then b = 0, c = d − 1, e = d − 2, thus P (B1 , 0) = −a2 d2 + 2a2 d − 2ad2 + 4ad + 3d2 −
6d + 4 ̸= 0 by a ∈ {0, 2} and d ≥ 4.
Case 3: ε(v2 ) = d − 2, ε(v1 ) = d, ε(vd−1 ) = d − 1.
Then b = d−1, c = 0, e = d−2, thus P (B1 , 0) = −d4 −2ad2 +6d3 +a2 +4ad−10d2 +6d ̸=
0 by a ∈ {0, 2} and d ≥ 4.
Case 4: ε(v2 ) = d − 2, ε(v1 ) = ε(vd−1 ) = d.
Then b = c = 0, e = d − 2, thus P (B1 , 0) = −2ad2 + 4ad + a2 + 2d2 − 4d + 3 ̸= 0 by
a ∈ {0, 2} and d ≥ 4.
Case 5: ε(v2 ) ≥ d − 1.
Then a = e = 0, thus P (B1 , 0) = −b2 c2 + b2 + c2 + d2 − 1 ̸= 0 by b, c ∈ {0, d − 1} and
d ≥ 4.
Combining the above arguments, we have P (B1 , 0) ̸= 0, which implies mB1 −I (−1) = 0.
Therefore, m(−1) ≤ n − 5 by Lemma 3.2. ■

By Proposition 3.3, Theorem 3.6 and Lemma 3.8, we have the following conclusion
immediately.

Remark 3.9 Let G be a connected graph with order n, diam(G) = d and n − 3 ≤

m(−1) ≤ n − 4. Then 2 ≤ d ≤ 3.

Lemma 3.10 Let G ∼

= Kr ∨ (n − r)K1 be a complete (r + 1)-partite graph with 1 ≤ r ≤
n − 2. Then m(−1) = r − 1.

9
Proof. Clearly, the eccentricity matrix A(G) can be written as
 
Jr×r − Ir×r Jr×(n−r)
,
J(n−r)×r 2J(n−r)×(n−r) − 2I(n−r)×(n−r)

and the equitable quotient matrix of A(G) is

 
r−1 n−r
Q= .
r 2(n − r − 1)

Then P (A(G), λ) = (λ + 1)r−1 (r + 2)n−r−1 P (Q, λ) by Lemma 2.1. By direct calculation,

we have P (Q, −1) = r(n − r − 1) ̸= 0, which implies m(−1) = r − 1. ■

Theorem 3.11 Let G be a connected graph with order n ≥ 4. Then m(−1) = n − 3 if

and only if G ∈ {P4 , Kn−2 ∨ 2K1 }. In addition, G is determined by its A-spectrum.

Proof. By direct calculation, we have P (A(P4 ), λ) = (λ + 1)(λ − 1)(λ + 4)(λ − 4), then
mA(P4 ) (−1) = 1. If G ∼
= Kn−2 ∨ 2K1 , then mA(G) (−1) = n − 3 by Lemma 3.10 and
r = n − 2.
Let m(−1) = n − 3. Now we show G ∈ {P4 , Kn−2 ∨ 2K1 }.
Clearly, we have 2 ≤ d ≤ 3 by Remark 3.9. Then we complete the proof by the
following two cases.
Case 1: d = 3.
Let d(v0 , v3 ) = 3 and P4 = v0 v1 v2 v3 be a path with length 3. Then ε(v0 ) = ε(v3 ) = 3,
2 ≤ ε(v1 ), ε(v2 ) ≤ 3. Firstly, we show ε(v1 ) = ε(v2 ) = 2.
Let ε(v1 ) = 3, and B2 be the principal submatrix of A(G)+I indexed by {v0 , v1 , v2 , v3 },
where x ∈ {0, 2} and  
1 0 x 3
0 1 0 0
B2 = 
x
.
0 1 0
3 0 0 1
By direct calculation, we have P (B2 , 0) = −x2 − 8 < 0, which implies mB2 −I (−1) = 0.
Thus m(−1) ≤ n − 4 by Lemma 3.2, a contradiction. Therefore, ε(v1 ) = 2.
Similar to the above proof, we have ε(v2 ) = 2.
Now we complete the proof of Case 1 by the following two Subcases.
Subcase 1.1: n = 4.
It is obvious that G ∼
= P4 .
Subcase 1.2: n ≥ 5.

10
 Then there exists a vertex, say, v4 ∈ V (G) \ {v0 , v1 , v2 , v3 }, satisfying ε(v4 ) ∈ {2, 3}
and there exists some vi (0 ≤ i ≤ 3) such that vi ∼ v4 . Then the induced subgraph
G[{v0 , v1 , v2 , v3 , v4 }] must be one of H2 , H3 , · · · , H7 , as shown in Figure 2.

H2 H3 H4 H5 H6 H7

Figure 2: Graphs H2 , H3 , H4 , H5 , H6 , H7 .

Let Bi (3 ≤ i ≤ 8) be the principal submatrix of A(G)+I indexed by {v0 , v1 , v2 , v3 , v4 }

of Hi−1 . Then
     
1 0 2 3 0 1 0 2 3 b 1 0 2 3 0
0 1 0 2 2 0 1 0 2 0 0 1 0 2 0
   
 
2
B3 =  0 1 0 2, B4 = 2 0 1 0 2, B5 = 2 0 1 0 2,
 
3 2 0 1 a  3 2 0 1 a  3 2 0 1 a
0 2 2 a 1 b 0 2 a 1 0 0 2 a 1

     
1 0 2 3 0 1 0 2 3 b 1 0 2 3 0
0 1 0 2 2 0 1 0 2 0 0 1 0 2 0
   
 
2
B6 =  0 1 0 0, B7 = 2 0 1 0 0, B8 = 2 0 1 0 0,
 
3 2 0 1 b  3 2 0 1 b  3 2 0 1 b
0 2 0 b 1 b 0 0 b 1 0 0 0 b 1

where a ∈ {0, 2, 3}, b ∈ {0, 2}. By direct calculation, for 3 ≤ i ≤ 8, we have the
coefficient of λ in P (Bi , λ) is not equal to 0, then mBi (0) ≤ 1, which impliesmBi −I (−1) ≤ 1
and thus m(−1) ≤ n − 4 by Lemma 3.2, a contradiction. For example, if i = 3, then
P (B3 , λ) = λ5 − 5λ4 − 15λ3 + (65 − 4a)λ2 + (50 − 4a)λ + 24a − 192. Clearly, we have
50 − 4a ̸= 0 by a ∈ {0, 2, 3}.
Case 2: d = 2.
Firstly, we show G is P3 -free. Otherwise, there exist three vertices v1 , v2 , v3 , such that
G[{v1 , v2 , v3 }] = P3 . Without loss of generality, we suppose v1 ∼ v2 , v1 ̸∼ v3 and v2 ̸∼ v3 .
Clearly, by diam(G) = 2, there exists v4 ∈ V (G), such that v4 ∼ v2 and v4 ∼ v3 . Let
B9 be the principal submatrix of A(G) + I indexed by {v1 , v2 , v3 , v4 }. Then
 
1 0 2 c
0 1 2 e
B9 = 
2
,
2 1 e
c e e 1

11
where c ∈ {0, 1, 2}, e ∈ {0, 1}.
By direct calculation, we have P (B9 , λ) = λ4 − 4λ3 − (c2 + 2e2 + 2)λ2 + (2c2 − 4ce +
12)λ+3c2 −4ce+6e2 −7, and P (B9 , 0) = 3c2 −4ce+6e2 −7 ̸= 0 by c ∈ {0, 1, 2}, e ∈ {0, 1}.
Then mB9 (0) = 0, which implies mB9 −I (−1) = 0 and thus m(−1) ≤ n − 4 by Lemma 3.2,
a contradiction. Therefore, G is P3 -free.
By Lemma 2.3 and G is P3 -free, G is a complete multipartite graph, say, G ∼
=
Kn1 ,n2 ,··· ,nk , where n1 ≥ n2 ≥ · · · ≥ nk ≥ 1 and k ≥ 2.
Subcase 2.1: nk ≥ 2.
Then m(−1) = 0 by (ii) of Lemma 2.9, a contradiction.
Subcase 2.2: nk = 1.
Without loss of generality, we assume that nk−r+1 = · · · = nk = 1, and n1 ≥ n2 ≥
· · · ≥ nk−r ≥ 2 if k − r > 0.
k−r
P
If k − r ≥ 2, then m(−1) = r − 1 = (r − ni ) − 1 ≤ (n − n1 − nk−r ) − 1 ≤ n − 5 by
i=1
Lemma 2.8, a contradiction.
If k−r = 1, then G = Kn−r,1,··· ,1 ∼
= Kr ∨(n−r)K1 with 1 ≤ r ≤ n−2 and m(−1) = r−1
by Lemma 3.10, and thus r = n − 2 by m(−1) = n − 3. Therefore, G ∼ = Kn−2 ∨ 2K1 .
If k − r = 0, then G = K1,1,··· ,1 ∼
= Kn , and thus m(−1) = n − 1 by Theorem 3.6, a
contradiction.
Combining the above two subcases, we have G ∼
= Kn−2 ∨ 2K1 if d = 2, and m(−1) =
n − 3 if and only if G ∈ {P4 , Kn−2 ∨ 2K1 } by the above two cases.
Clearly, G is determined by its A-spectrum. ■

graph G P (A(G), λ)
Kn−4 ∨ 4K1 (λ + 1)n−5 (λ + 2)3 [λ2 − (1 + n)λ + 2n − 14]
Kn−4 ∨ (2K1 ∪ K2 ) (λ + 1)n−5 (λ + 2)λ[λ3 − (n − 3)λ2 − (2n + 10)λ + 4n − 32]
Kn−4 ∨ (P3 ∪ K1 ) (λ + 1)n−5 (λ + 2)[λ4 − (n − 3)λ3 − (2n + 6)λ2 + (4n − 20)λ + 8]
Kn−4 ∨ 2K2 (λ + 1)n−5 λ2 (λ + 2)[λ2 − (n − 3)λ − 2n + 6]
(λ + 1)n−5 [λ5 − (n − 5)λ4 − (4n − 4)λ3 − 12λ2
Kn−4 ∨ P4
+(8n − 16)λ + 16]
Kn−4 ∨ (K3 ∪ K1 ) (λ + 1)n−5 λ2 [λ3 − (n − 5)λ2 − (4n − 4)λ − 12]
Kn−4 ∨ C4 (λ + 1)n−5 (λ + 2)2 [λ3 + (1 − n)λ2 + 4n − 12]

Table 1: the characteristic polynomial of A(G), where G ∈ G1

.

12
Lemma 3.12 Let G be a connected graph with order n(≥ 5) and |V1 | = n − 4, G1 =
{Kn−4 ∨ 4K1 , Kn−4 ∨ (2K1 ∪ K2 ), Kn−4 ∨ (P3 ∪ K1 ), Kn−4 ∨ 2K2 , Kn−4 ∨ P4 , Kn−4 ∨ (K3 ∪
K1 ), Kn−4 ∨ C4 }. Then G ∈ G1 and m(−1) = n − 5.

Proof. Clearly, we have d = 2 by |V1 | = n − 4 and Proposition 3.3. Thus V (G) = V1 ∪ V2 ,

G[V1 ] = Kn−4 , the maximal degree of G[V2 ], △(G[V2 ]) ≤ 2, and the number of edges of
G[V2 ], 0 ≤ |E(G[V2 ])| ≤ 4. Therefore, G ∈ G1 . By direct calculation and Lemma 2.1, we
have P (A(G), λ) as shown in Table 1, and thus m(−1) = n − 5. ■

Theorem 3.13 Let G be a connected graph with order n ≥ 9. Then m(−1) = n − 4 if

and only if G ∈ {Kn−3 ∨ (K2 ∪ K1 ), Kn−3 ∨ 3K1 }. In addition, G is determined by its
A-spectrum.

Proof. By direct calculation and Lemma 2.1, we have

P (A(Kn−3 ∨ (K2 ∪ K1 )), λ) = (λ + 1)n−4 λ[λ3 − (n − 4)λ2 − (3n − 1)λ − 8],

P (A(Kn−3 ∨ 3K1 ), λ) = (λ + 1)n−4 (λ + 2)2 (λ2 − nλ + n − 7),

which implies m(−1) = n − 4 if G ∈ {Kn−3 ∨ (K2 ∪ K1 ), Kn−3 ∨ 3K1 }.

( = n − 4. Then 2 ≤ diam(G) = d ≤ 3 by Remark 3.9. If |V1 | = ∅, then

Let m(−1)
n − 5, if d = 2
m(−1) ≤ by n ≥ 9 and Lemma 3.5, a contradiction with m(−1) =
n − 9, if d = 3
n − 4. Therefore, |V1 | =
̸ ∅, and thus d = 2, V (G) = V1 ∪ V2 by Proposition 3.3. Moreover,
we have n − 4 ≤ |V1 | ≤ n − 3 by Lemma 3.4.
If |V1 | = n − 4, then G ∈ G1 and m(−1) = n − 5 by Lemma 3.12, a contradiction.
If |V1 | = n − 3, it is obvious that G ∼
= Kn−3 ∨ (K2 ∪ K1 ) or G ∼
= Kn−3 ∨ 3K1 .
Clearly, G is determined by its A-spectrum. ■

Lemma 3.14 Let G be a connected graph with order n and diam(G) = d, m(−1) = n−5.
Then 2 ≤ d ≤ 4.

Proof. Clearly, we have d = 1 if and only if G ∼

= Kn , then m(−1) = n − 1 > n − 5 by
Theorem 3.6, a contradiction. So d ≥ 2. Now we show d ≤ 4.
Let d(v0 , vd ) = d ≥ 5 and Pd+1 = v0 v1 v2 · · · vd−2 vd−1 vd be a path with length d. Then
ε(v0 ) = ε(vd ) = d, d − 1 ≤ ε(v1 ), ε(vd−1 ) ≤ d, d − 2 ≤ ε(v2 ), ε(vd−2 ) ≤ d. Clearly, the

13
principal submatrix of A(G) + I indexed by {v0 , v1 , v2 , vd−2 , vd−1 , vd } is
 
1 0 0 a b d
0 1 0 0 0 c
 
0 0 1 0 0 e
B10 = 
a
,
 0 0 1 0 0 
b 0 0 0 1 0
d c e 0 0 1
where a, e ∈ {0, d − 2}, b, c ∈ {0, d − 1}. By direct calculation and d ≥ 5, we have

P (B10 , 0) = (a2 + b2 )(c2 + e2 ) − a2 − b2 − c2 − d2 − e2 + 1 ̸= 0,

which implies mB10 −I (−1) = mB10 (0) = 0, Thus m(−1) ≤ n − 6 by Lemma 3.2, a
contradiction. Therefore, d ≤ 4. ■

Lemma 3.15 Let G be a connected graph with order n ≥ 16, m(−1) = n − 5, G1 as

defined in Lemma 3.12. Then G ∈ G1 ∪ {Kn−5 ∨ C5 , Kn−5 ∨ (K1 ∪ P4 ), Kn−5 ∨ H1 }.

Proof. Let m(−1) = n − 5. Then 2 ≤ diam(G) = d ≤ 4 by Lemma 3.14.

n − 16, d = 2

If |V1 | = ∅, then m(−1) ≤ n − 8, d = 3 by n ≥ 16 and Lemma 3.5, a contra-

n − 6, d=4

diction with m(−1) = n − 5. Therefore, |V1 | =
̸ ∅, and thus d = 2, V (G) = V1 ∪ V2 by
Proposition 3.3. Moreover, we have n − 5 ≤ |V1 | ≤ n − 4 by Lemma 3.4.
Case 1: |V1 | = n − 4.
It is obvious that G ∈ G1 by Lemma 3.12.
Case 2: |V1 | = n − 5.
Then |V2 | = 5 by V (G) = V1 ∪ V2 , and thus G ∼
= Kn−5 ∨ G[V2 ], the maximal degree
of G[V2 ], △(G[V2 ]) ≤ 3. Without loss of generality, we assume that v1 ∈ V1 and V2 =
{v2 , v3 , v4 , v5 , v6 }. Now we show G ∈ {Kn−5 ∨ C5 , Kn−5 ∨ (K1 ∪ P4 ), Kn−5 ∨ H1 } by the
following two subcases.
Subcase 2.1: G[V2 ] is a connected graph.
Subcase 2.1.1: K1,3 is an induced subgraph of G[V2 ].
Let G[{v2 , v3 , v4 , v5 }] = K1,3 , where v2 ∼ v3 , v2 ∼ v4 , v2 ∼ v5 . Then the principal
submatrix of A(G) + I indexed by {v1 , v2 , v3 , v4 , v5 , v6 } is
 
1 1 1 1 1 1
1 1 0 0 0 2 
 
 1 0 1 2 2 a
B11 = 
1 0 2 1 2 b  ,

 
1 0 2 2 1 c 
1 2 a b c 1

14
where a, b, c ∈ {0, 2}. By direct calculation, we have P (B11 , 0) = 2(ab + ac + bc − a2 −
b2 − c2 − a − b − c + 2) ̸= 0, then mB11 (0) = 0, and thus rank(A(G) + I) ≥ rank(B11 ) =
6 − mB11 (0) = 6, which implies m(−1) = mA(G)+I (0) = n − rank(A(G) + I) ≤ n − 6, a
contradiction with m(−1) = n − 5.
Subcase 2.1.2: P4 is an induced subgraph of G[V2 ], and G[V2 ] is K1,3 -free.
Let P4 = v2 v3 v4 v5 . Then the principal submatrix of A(G)+I indexed by {v1 , v2 , v3 , v4 , v5 , v6 }
is  
1 1 1 1 1 1
1 1 0 2 2 e
 
1 0 1 0 2 f
B12 =
1
,
 2 0 1 0 g
1 2 2 0 1 h
1 e f g h 1
where e, f, g, h ∈ {0, 2}. By direct calculation, we have P (B12 , 0) = 2(f 2 + g 2 + ef + 3f g +
gh + 5e + 5h − e2 − h2 − eg − 3eh − f h − 5f − 5g).

H8 H9 H10 H11 H12

Figure 3: Graphs H8 , H9 , H10 , H11 , H12 .

Now we show G[V2 ] ∼

= C5 or G[V2 ] ∼
= H1 . Otherwise, G[V2 ] must be one of H8 , H9 , H10 , H11 ,
which is shown in Figure 3 by G[V2 ] is K1,3 -free, and the values of P (B12 , 0) is shown in
Table 2.

G[V2 ] e f g h P (B12 , 0)
H8 2 2 2 0 12
H9 0 0 2 2 8
H10 0 0 0 2 12
H11 0 0 2 0 −12

Table 2: the values of P (B12 , 0)

.
Then mB12 (0) = 0 by P (B12 , 0) ̸= 0, and thus rank(A(G) + I) ≥ rank(B12 ) =
6 − mB12 (0) = 6, which implies m(−1) = mA(G)+I (0) = n − rank(A(G) + I) ≤ n − 6, a
contradiction with m(−1) = n − 5. Therefore, G[V2 ] ∼
= C5 or G[V2 ] ∼
= H1 in this subcase.
Subcase 2.1.3: C4 is an induced subgraph of G[V2 ], and G[V2 ] is {K1,3 , P4 }-free.

15
 Without loss of generality, we assume that G[{v2 , v3 , v4 , v5 }] ∼
= C4 and then G[V2 ] ∼
=
H12 by G[V2 ] is {K1,3 , P4 }-free, where H12 as shown in Figure 3. Let B13 be the principal
submatrix of A(G) + I indexed by {v1 } ∪ V2 , where
 
1 1 1 1 1 1
1 1 0 2 0 0
 
1 0 1 0 2 0
B13 = 
1 2 0 1 0
.
 0

1 0 2 0 1 2
1 0 0 0 2 1

Then P (B13 , 0) = −8 ̸= 0 by direct calculation, and thus rank(A(G) + I) ≥ rank(B13 ) =

6, which implies m(−1) = mA(G)+I (0) = n − rank(A(G) + I) ≤ n − 6, a contradiction.
Subcase 2.1.4: G[V2 ] is {K1,3 , P4 , C4 }-free.
Then G[V2 ] ∼
= Kl ∨ (Ks ∪ Kt ) by Lemma 2.4, where s, t ≥ 0 and l ≥ 1, and thus
△(G[V2 ]) = (l − 1) + s + t = |V2 | − 1 = 4, a contradiction with △(G[V2 ]) ≤ 3.
Subcase 2.2: G[V2 ] is a disconnected graph.
Subcase 2.2.1: G[V2 ] has five connected components.
Clearly, G[V2 ] ∼
= 5K1 . Let B14 be the principal submatrix of A(G) + I indexed by
{v1 } ∪ V2 , where  
1 1 1 1 1 1
1 1 2 2 2 2
 
1 2 1 2 2 2
B14 =
1
.
 2 2 1 2 2

1 2 2 2 1 2
1 2 2 2 2 1
Then P (B14 , 0) = 4 by direct calculation, and thus rank(A(G) + I) ≥ rank(B14 ) = 6,
which implies m(−1) = mA(G)+I (0) = n − rank(A(G) + I) ≤ n − 6, a contradiction.
Subcase 2.2.2: G[V2 ] has four connected components.
Clearly, G[V2 ] ∼
= 3K1 ∪ K2 , where v2 ∼ v3 . Let B15 be the principal submatrix of
A(G) + I indexed by {v1 } ∪ V2 , where
 
1 1 1 1 1 1
1 1 0 2 2 2
 
1 0 1 2 2 2
B15 =
1
.
 2 2 1 2 2

1 2 2 2 1 2
1 2 2 2 2 1

Then P (B15 , 0) = −8 by direct calculation, and thus rank(A(G) + I) ≥ rank(B15 ) = 6,

which implies m(−1) = mA(G)+I (0) = n − rank(A(G) + I) ≤ n − 6, a contradiction.

16
 Subcase 2.2.3: G[V2 ] has three connected components.
Clearly, G[V2 ] ∈ {2K1 ∪ P3 , 2K1 ∪ K3 , 2K2 ∪ K1 }.
If G[V2 ] ∼
= 2K1 ∪P3 or 2K1 ∪K3 , where G[{v2 , v3 }] ∼
= 2K1 , then the principal submatrix
of A(G) + I indexed by {v1 } ∪ V2 is
 
1 1 1 1 1 1
1 1 2 2 2 2
 
1 2 1 2 2 2
B16 =
1
,
 2 2 1 0 a1 

1 2 2 0 1 0
1 2 2 a1 0 1

where a1 ∈ {0, 2}. By direct calculation, we have P (B16 , 0) = −2(a21 + 3a1 − 4) ̸= 0. Thus
rank(A(G)+I) ≥ rank(B16 ) = 6, which implies m(−1) = mA(G)+I (0) = n−rank(A(G)+
I) ≤ n − 6, a contradiction.
If G[V2 ] ∼
= 2K2 ∪ K1 , where v3 ∼ v4 , v5 ∼ v6 , then the principal submatrix of A(G) + I
indexed by {v1 } ∪ V2 is  
1 1 1 1 1 1
1 1 2 2 2 2
 
1 2 1 0 2 2
B17 =
1
.
 2 0 1 2 2

1 2 2 2 1 0
1 2 2 2 0 1
By direct calculation, we have P (B17 , 0) = 12. Thus rank(A(G) + I) ≥ rank(B17 ) = 6,
which implies m(−1) = mA(G)+I (0) = n − rank(A(G) + I) ≤ n − 6, a contradiction.
Subcase 2.2.4: G[V2 ] has two connected components.
Clearly, we have G[V2 ] ∈ {K1 ∪ P4 , K1 ∪ C4 , K1 ∪ K1,3 , K1 ∪ P3 ∪ K1 , K1 ∪ (K2 ∨
2K1 ), K1 ∪ K4 , K2 ∪ P3 , K2 ∪ K3 }. Now we show G[V2 ] ∼
= K 1 ∪ P4 .
If G[V2 ] ∼
= K1 ∪ C4 , where C4 = v3 v4 v5 v6 v3 ; or G[V2 ] ∼
= K1 ∪ P3 ∪ K1 , where v3 ∼
v4 , v3 ∼ v5 , v3 ∼ v6 , v4 ∼ v5 , then the principal submatrix of A(G) + I indexed by
{v1 , v2 , v3 , v4 , v5 , v6 } is  
1 1 1 1 1 1
1 1 2 2 2 2
 
1 2 1 0 a2 0 
B18 =
1
,
 2 0 1 0 2 
1 2 a2 0 1 b2 
1 2 0 2 b2 1
(
(2, 0), if G[V2 ] ∼= K1 ∪ C4
where (a2 , b2 ) = ∼ . By direct calculation, we have
(0, 2), if G[V2 ] = K1 ∪ P3 ∪ K1
P (B18 , 0) = 2a2 b2 + 2b22 + 8 − 2a22 − 6a2 − 2b2 ̸= 0.

17
 If G[V2 ] ∼
= K1 ∪ K1,3 , where v2 ∼ v3 , v2 ∼ v4 , v2 ∼ v5 , then the principal submatrix of
A(G) + I indexed by {v1 , v2 , v3 , v4 , v5 , v6 } is B11 in Subcase 2.1.1 with a = b = c = 2, and
thus P (B11 , 0) ̸= 0.
If G[V2 ] ∼
= K1 ∪ (K2 ∨ 2K1 ) or K1 ∪ K4 , where G[{v3 , v4 , v5 , v6 }] ∼
= K2 ∨ 2K1 or K4 ,
then the principal submatrix of A(G) + I indexed by {v1 , v2 , v3 , v4 , v5 , v6 } is
 
1 1 1 1 1 1
1 1 2 2 2 2
 
1 2 1 0 0 0
B19 = 
1 2 0 1 0
,
 0
1 2 0 0 1 a3 
1 2 0 0 a3 1

where a3 ∈ {0, 2}. By direct calculation, we have P (B19 , 0) = 2(a23 + a3 − 2) ̸= 0.

If G[V2 ] ∼
= K2 ∪ P3 or K2 ∪ K3 , where G[{v4 , v5 , v6 }] = P3 or K3 , then the principal
submatrix of A(G)+I indexed by {v1 , v2 , v3 , v4 , v5 , v6 } is obtained from B16 by replacing 2
by 0 in the (2, 3), (3, 2) position, denoted by B20 , and we have P (B20 , 0) = 2(a21 +a1 −2) ̸= 0
by direct calculation.
Combining the above arguments, we get rank(A(G) + I) ≥ 6, which implies m(−1) ≤
n − 6, a contradiction. Thus G[V2 ] ∼
= K1 ∪ P4 in this subcase.
By Case 1 and Case 2, we complete the proof. ■

graph G P (A(G), λ)
Kn−5 ∨ C5 (λ + 1)n−5 (λ2 + 2λ − 4)2 (λ − n + 1)
Kn−4 ∨ (K1 ∪ P4 ) (λ + 1)n−5 (λ2 + 2λ − 4)[λ3 − (n − 3)λ2 − (2n + 10)λ + 4n − 36]
Kn−4 ∨ H1 (λ + 1)n−5 (λ2 + 2λ − 4)[λ3 − (n − 3)λ2 − (2n + 2)λ + 4n − 20]

Table 3: the characteristic polynomial of A(G)

.

Theorem 3.16 Let G be a connected graph with order n ≥ 16, G1 = {Kn−4 ∨4K1 , Kn−4 ∨
(2K1 ∪ K2 ), Kn−4 ∨ (P3 ∪ K1 ), Kn−4 ∨ 2K2 , Kn−4 ∨ P4 , Kn−4 ∨ (K3 ∪ K1 ), Kn−4 ∨ C4 }. Then
m(−1) = n − 5 if and only if G ∈ G1 ∪ {Kn−5 ∨ C5 , Kn−5 ∨ (K1 ∪ P4 ), Kn−5 ∨ H1 }. In
addition, G is determined by its A-spectrum.

Proof. If m(−1) = n − 5, then G ∈ G1 ∪ {Kn−5 ∨ C5 , Kn−5 ∨ (K1 ∪ P4 ), Kn−5 ∨ H1 } by

Lemma 3.15.
Conversely, if G ∈ {Kn−5 ∨C5 , Kn−5 ∨(K1 ∪P4 ), Kn−5 ∨H1 }, then by direct calculation
and Lemma 2.1, we have P (A(G), λ) as shown in Table 3. Thus m(−1) = n − 5 and G is

18
determined by its A-spectrum if G ∈ G1 ∪ {Kn−5 ∨ C5 , Kn−5 ∨ (K1 ∪ P4 ), Kn−5 ∨ H1 } by
Table 1 and Table 3. ■

By Theorems 3.6, 3.7, 3.11, 3.13, 3.16, the results of Theorem 1.1 hold, and we com-
plete the proof. Clearly, if there exists v ∈ V (G) such that ε(v) = 1, then A(G) is
irreducible, and thus we have the following conclusion by Theorem 1.1 immediately.

Remark 3.17 Let G be a connected graph with order n ≥ 16 and m(−1) ≥ n − 5. Then
A(G) is irreducible.

Conclusion
In this paper, by using equitable quotient matrix, interlacing theorem, the relationship
between rank of matrix and the multiplicity of eigenvalue 0 etc., we characterize graphs
with m(−1) = n − i, where i ≤ 5. Actually, we can also characterize graphs with
m(−1) = n − 6 in a similar way, but the calculation would be complicated.
Fowler and Pisanski [5] introduced HL-index of the adjacency matrix of a graph. It is
related to the HOMO-LUMO separation studied in the theoretical chemistry. Similarly,
Wang et al. [15] defined the HL-index RA (G) = max{|ξH |, |ξL |} with respect to the
eccentricity matrix A(G) of a connected graph G with order n, where H = ⌊ n+1
2
⌋, L =
⌈ n+1
2
⌉. Usually, ξH and ξL are called median eigenvalues. Clearly, for sufficiently large n,
we always have RA (G) = 1 for a connected graph G with m(−1) = n − i, where i ≤ 5.
After careful consideration and demonstration, we propose the following conjecture
and two problems about the multiplicity of −1 and 0 in the spectrum of the eccentricity
matrix for further research.
Conjecture 1. Let G be a connected graph with order n, m(ξ) = n − i for 1 ≤ i ≤
n − ⌈ n+1
2
⌉. Then ξ ∈ {−2, −1, 0} for a large enough n.
Problem 1. Determine the connected graphs with order n and m(−1) = n − i, where
6 ≤ i ≤ n − ⌈ n+1
2
⌉.
Problem 2. Determine the connected graphs with order n and m(0) = n − i, where
i ≤ 4.

19
Funding
This work is supported by the National Natural Science Foundation of China (Grant
Nos. 12371347, 12271337).

References
[1] J.A. Bondy, U.S.R. Murty, Graph Theory, Spring, London, 2008.

[2] A.E. Brouwer, W.H. Haemers, Spectra of Graphs, Springer, New York, 2012.

[3] D.M. Cvetković, M. Doob, H. Sachs, Spectra of Graphs: Theory and Applications,
1980.

[4] D. Cvetković, P. Rowlinson, S.K. Simić, An Introduction to the Theory of Graph

Spectra, Cambridge Univ. Press, Cambridge, 2010.

[5] P.W Fowler, T. Pisanski, HoMO-LUMO maps for fullerenes, Acta Chimica Slovenica,
57 (2010) 513-517.

[6] X. Gao, Z. Stanić, J.F. Wang, Grahps with large multiplicity of −2 in the spectrum
of the eccentricity matrix, Discrete Mathematics, 347 (2024) 114038.

[7] X.Y. Lei, J.F. Wang, G.Z. Li, On the eigenvalues of eccentricity matrix of graphs,
Discrete Applied Mathematics, 295 (2021) 134-147.

[8] H.Q. Lin, M.Q. Zhai, S.C. Gong, On graphs with at least three distance eigenvalues
less than −1, Linear Algebra and its Application, 458 (2014) 548-558.

[9] V. Lozin, Graph theory notes, 2017, https://homepages.warwick.ac.uk/masgax/graph-

Theory-notes. Accessed 27 September 2017.

[10] I. Mahato, M.R. Kannan, On the eccentricity matrix of trees: Inertia and spectral
symmetry, Discrete Mathematics, 345 (2022) 113067.

[11] B. Mohar, Median eigenvalues and the HOMO-LUMO index of graphs, Journal of
Combinatorial Theory Series B, 112 (2015) 78-92.

[12] B. Mohar, Median eigenvalues of bipartite subcubic graphs, Combinatorics Proba-

bility & Computing, 25 (2016) 768-790.

20
[13] Z.P. Qiu, Z.K. Tang, Q.Y. Li, Eccentricity spectral radius of t-clique trees with given
diameter. Discrete Applied Mathematics, 337 (2023) 202-217.

[14] S. Sorgun, H. Hüçük, On two problems related to anti-adjacency (eccentricity) ma-

trix, Discrete Applied Mathematics, 328 (2023) 1-9.

[15] J.F. Wang, X.Y. Lei, M. Lu, S. Sorgun, et al., On graphs with exactly one anti-
adjacency eigenvalue and beyond, Discrete Mathematics, 346 (2023) 113373.

[16] J.F. Wang, X.Y. Lei, W. Wei, X.B. Luo, S.C. Li, On the eccentricity matrix of
graphs and its applications to the boiling point of hydrocarbons, Chemometrics and
Intelligent Laboratory Systems, 207 (2020) 104173.

[17] J.F. Wang, M. Lu, F. Belardo, M. Randić, The anti-adjacency matrix of a graph:
eccentricity matrix, Discrete Applied Mathematics, 251 (2018) 299–309.

[18] W. Wei, X.C. He, S.C. Li, Solutions for two conjectures on the eigenvalues of the
eccentricity matrix, and beyond, Discrete Mathematics, 343 (2020) 111925.

[19] W. Wei, S.C. Li, On the eccentricity spectra of complete multipartite graphs, Applied
Mathematics and Computation, 424 (2022) 127036.

[20] L.H. You, M. Yang, W. So, W.G. Xi, On the spectrum of an equitable quotient
matrix and its application, Linear Algebra and its Application, 577 (2019) 21-40.

21