1. Introduction
All graphs considered here are finite and simple. Undefined notations and terminologies will conform to those in [
1].
For a graph G with vertices and edges, and are used to denote the vertex set and the edge set of G, respectively. For , and (or simple by ) be the degree of a vertex v in G. Let , set . If , take . Let , and be a cycle of G, use , and to denote the graphs obtained from G by deleting the vertex v, the edge e and all vertices of , respectively. Let H be a subgraph of G, and denotes the subgraph of G induced by the edge set . Let G and H be two graph which we denote by as the disjoint union of G and H and by the disjoint union of l copies of H.
For a graph G with n vertices, its adjacency matrix is the matrix with th entry equaling to 1 if vertices i and j are adjacent and equaling to 0 otherwise. denotes the degrees matrix corresponding to vertices of G on the main diagonal. The Laplacian matrix of G is denoted by , where . The characteristic polynomial of the adjacency matrix (respectively, Laplacian matrix ) is denoted by (respectively, ). The eigenvalues of () are also called the adjacency (Laplacian) eigenvalues of G. Since both matrices and are real symmetric matrices, their eigenvalues are all real numbers. We can therefore assume that and are adjacency eigenvalues and the Laplacian eigenvalues of G, respectively. The multiset of eigenvalues of and are called the adjacency spectrum and Laplacian spectrum of G, respectively. The maximum eigenvalue of is called the index of G. Two graphs are said to be cospectral with respect to the adjacency (respectively, Laplacian) matrix if they have the same adjacency (respectively, Laplacian) spectrum. A graph is said to be determined (DS for short) by its adjacency (respectively, Laplacian) spectrum if there is no other non-isomorphic graph with the same spectrum with respect to the adjacency (respectively, Laplacian) matrix.
To date, numerous examples of cospectral but non-isomorphic graphs have been reported. For example, Schwenk showed that the proportion of trees on
n vertices which are characterized by their spectra converges towards zero as
n increases in [
2] and Godsil and Mckay in [
3,
4] gave some constructions for pairs of cospectral graphs. Recently, many graphs with special structures have been proved to be determined by their spectrum, as can be seen in [
5,
6,
7,
8,
9,
10,
11,
12,
13,
14,
15,
16,
17,
18]. The authors in [
6,
7,
11] investigated the cospectrality of graphs up to order 11 and gave a survey on the spectral characterizations of graphs. Some results on the Laplacian spectral characterizations of graphs can be found in [
19,
20,
21].
Here, we list some connected graphs determined by their adjacency spectrum as following:
(i) The path with
n vertices
and its complement, where
, the complete graph
, the regular complete bipartite graph
, the cycle
and their complements, some graphs obtained by deleting some edges from
are determined by their adjacency spectrum [
6,
7,
8,
22]. We write
and
.
(ii) Let
denote a tree with a vertex
v of degree 3 such that
. Then,
is determined by its adjacency spectrum if and only if
for any
[
14,
17]. Write
.
(iii) A lollipop graph, denoted by
, is obtained by identifying a vertex of
and a vertex of
. Then, the lollipop graph is determined by its adjacency spectrum [
5,
10].
(iv) All connected graphs with an index in the interval (2,
are determined by their adjacency spectrum [
9].
(v) The sandglass graph is obtained by appending a triangle to each pendant vertex of a path. Then, the sandglass graph is determined by its adjacency spectrum [
12].
(vi) The
graph, denoted by
, is a graph consisting of the two given vertices joined by three paths whose order is
,
, and
, respectively, with any two of these paths only having the given vertices in common. The dumbbell graph, denoted by
, consists of two vertex-disjoint cycles
,
and a path
joining them having only its end-vertices in common with the cycles. An
∞-graph
is a graph consisting of two cycles
and
with just a vertex in common. The authors in [
15,
16,
23,
24,
25] gave some DS-graphs among
-graphs and all DS-graphs of the dumbbell graphs and
∞-graphs. In [
26], the authors proved the
-graphs without
as its subgraphs are DS.
In this paper, we give two parameters
and
related to the characteristic polynomial of
G. Some properties of
and
were obtained in
Section 3. We characterized all connected graphs with
and
. From
Section 3, one can see that
for all graphs
G in (i)–(vi) as shown above, except for
,
,
and
. By applying the results of the parameters
and
in
Section 4, we give the necessary and sufficient conditions for two classes of graphs to be DS with respect to their adjacency spectrum. We also give a method to characterize the graphs determined by their adjacency spectrum in
Section 4.
Section 5 gives a summary.
3. Two Invariants Related to Characteristic Polynomials
In this section, we investigate two invariants related to some coefficients of the characteristic polynomials of G. Then, some properties of the invariants are given.
Definition 1. Let G be a graph, the parameter is defined by the following It is easy to see that if , then and . Thus, we have:
Theorem 1. Let G and H be two graphs with . Then Theorem 2. Let G be a graph with components . Then Proof. It is sufficient to prove the case
. Let
,
and
, for
. By Lemmas 2 and 5, we have
By Definition 1, it follows
□
Theorem 3. Let G be a graph with n vertices and . Thenwhere (respectively, denotes the number of triangles (respectively, ) containing the edge in G. Proof. Since
(namely,
) and
, by Lemma 5, we have
Note that
,
,
and
if
. Since
, we have
This implies that . □
Note that if G is a connected graph, , except for . From Theorem 3, we have:
Corollary 1. Let G be a connected graph with at least three vertices and . Then
- (1)
, the equality holds if and only if is a pendent edge with and .
- (2)
If is a pendent edge of G and , then
Theorem 4. Let G be a connected graph. Then:
- (1)
and the equality holds if and only if
- (2)
if and only if .
Proof. (1) By induction on . Since and , we have (1) holds if .
Suppose . Choose such that or . So, and . By the induction hypothesis, . By Corollary 1, it is easy to get that and if and only if and .
By the induction hypothesis, we have . Therefore .
Conversely, since and , we can show that .
(2) By induction on . Since and , it is clear that (2) holds if .
Suppose
. Choose
such that
H is connected. By Theorem 3, we have
Note that . We consider only the following cases:
Case 1. .
Note that and . If and , we have by (1). Hence, . If and , we know that by the induction hypothesis. Therefore .
Case 2. .
Since and , we know that . By the induction hypothesis, . Therefore, .
Conversely, since and , we can show that if , and and by Theorem 3. □
By Theorems 1–3, one can construct all connected graphs
G with
. Let
be a connected graph and
. By Theorem 4,
and
. Now, we give an algorithm to construct all connected graphs in
as following.
Algorithm 1: Construction of all connected graphs with . |
(i) Take and . (ii) Let i be an integer, , for to i (step −1) for each for each if then end for for each if then , where . end for end for end for Note. In the front steps, (respectively, denotes the number of triangles (respectively, ) containing the edge in graph .
|
Lemma 6. Algorithm 1 achieves completeness, that is, if G is any connected graph with and , then by Algorithm 1.
Proof. By induction on and , by Theorem 4, it is true for . Suppose that the algorithm achieves completeness for . When , we shall prove the completeness by induction on . □
By the proof of Theorem 4, we have that it is true for . Suppose , and we consider the following cases:
Case 1. G has a pendant edge, say
and
. Then,
, where
H is connected. By Corollary 1
If , then and , so by the induction hypothesis on , . If , then , say , and so by the induction hypothesis on . Hence, G must be obtained from H by adding the pendant edge from Algorithm 1.
Case 2. G has no any pendant edge. Then, there exists at least an edge
such that
is connected. Clearly,
. By Theorem 3,
Hence, , say . By the induction hypothesis on , . Therefore, G must be obtained from H by adding the edge by Algorithm 1. This completes the proof of the completeness. □
Note that if and , then . By Lemma 6 and Algorithm 1, we obtain the following theorem. Here, the proof is omitted.
Theorem 5. Let G be a connected graph. Then:
- (i)
if and only if
- (ii)
if and only if
- (iii)
if and only if , where all graphs are listed in Figure 1.
Note: in
Figure 1, for
and
, does not contain
as its subgraphs and all the dot lines of the graphs denote a path with at least two vertices.
Now we give another invariant as follows:
Definition 2. Let G be a connected graph. Set .
From Definitions 1 and 2 and Theorems 1–3, we can easily prove the following:
Theorem 6. (1) Let G and H be two graphs such that . Then (2) Let G be a graph with k components . Then (3) Let G be a connected graph and . Then Theorem 7. Let G be a connected graph. Then, , and the equality holds if and only if .
Proof. By induction in , since and , we have (1) which holds when .
Suppose . Choose such that is connected. Clearly, . We distinguish the three following cases:
Case 1. e is a pendant edge.
Obviously,
, where
. By Theorem 5 and the induction hypothesis, we have
and
Note that if and only if and . By the induction hypothesis, . Since and G is connected, we have .
Case 2. e is an edge in a triangle.
We know that
is connected and
. By Theorem 6 (3),
and
if and only if
and
. By the induction hypothesis,
. Thus,
(only if
).
Case 3. e is an edge in a cycle whose length is greater than 3.
Clearly,
. By Theorem 6 (3), we have
and
if and only if
and
. By the induction hypothesis,
. Hence,
.
Conversely, by Theorem 4 and Definition 2, we can directly verify our findings. □
Let be a connected graph and . Similarly to , we construct all connected graphs in by the following algorithm.
Lemma 7. Algorithm 2 is completeness, that is, if G is any connected graph with and , then by Algorithm 2.
Proof. By induction in and . By Theorem 7, it is true for . Suppose that the algorithm is completeness for . When , we shall prove the completeness by induction in .
By the proof of Theorem 7, we have that it is true for . Suppose that , and we consider the following cases:
Case 1. G has a pendant edge, say
and
. Then,
, where
H is connected. Note that
. By the proof of case 1 of Theorem 7, we have
If , then and , so by the induction hypothesis on , . If , then , say , and so by the induction hypothesis on . Hence, G must be obtained from H by adding the pendant edge from Algorithm 2, where .
Case 2. G has no any pendant edge. Then, there exists at least an edge
such that
is connected. Since
, by Theorem 6 (3) we have
If , then and , so by the induction hypothesis on , . If , then , say , and so by the induction hypothesis on . Hence, G must be obtained from H by adding the edge from Algorithm 2. This completes the proof of the completeness. □
Algorithm 2: Construction of all connected graphs with . |
(i) Take and . (ii) Let i be a negative integer, , for to i (step −1) for each for each if then end for for each if then , where . end for end for end for Note. In the front steps, (respectively, denotes the number of triangles (respectively, ) containing the edge in the graph .
|
From Lemma 7, Theorems 6 and 7 as well as Algorithm 2, we can prove the following. Here, the proof is omitted.
Theorem 8. Let G be a connected graph. Then:
- (1)
if and only if
- (2)
if and only if
- (3)
if and only if , where , for and , does not contain as its subgraph.
Remark 1. The invariants and are important for determining DS graphs. By Theorems 6 and 7, for each component of the cospectral graphs of G, we have that , which gives all possible cospectral graphs of G. It is therefore easy to find all possible cospectral graphs of G. Then, by considering the invariants and , it can be ascertained whether G is determined by its adjacency spectrum.
Remark 2. By Algorithms 1 and 2, one can characterize all connected graphs with and . For example, . Since there are many classes of graphs with and for a large positive integer k, it is difficult to characterize all graphs with and .
Remark 3. In this section, we define two invariants for graph G, which satisfies two important properties: component additivity (Theorems 2 and 6 (2)) and boundedness (Theorems 4 and 7). In fact, we define some new invariants with component additivity and boundedness. For example, and , which have similar properties and applications of and . The readers interested by new invariants may study their properties and find more than DS graphs.
4. Some Graphs Determined by Their Adjacency Spectrum
For convenience, we denote by
the set
, for
, and by
the set
, for
. In
Section 1, we list nearly all connected graphs
G determined by their adjacency spectrum, which
, except for
,
,
and
. In fact, the necessary and sufficient conditions for all connected
G with
to be DS were found. However, there are a few results on the DS-graphs of the connected graphs
G with
, which are some special classes in
,
and
, seeing [
9,
15,
16,
26]. In this section, by using the results of previous sections, we shall give the necessary and sufficient conditions for one class of trees
T with
and one class of graphs
G with
to be DS with respect to their adjacency spectrum. In the following text, we use
instead of
.
Lemma 8 ([
27])
. All roots of and are the following:
Let , set , and then it is useful to write the characteristic polynomial of and in the following form:
- (1)
,
- (2)
.
The following graphs and lemmas are frequently used in following text.
Remark 4. The parameters of each graph in Figure 2 take the following values: for , for and for . Lemma 9 ([
27])
. (1) Let H be a proper subgraph of a connected graph G, then(2) For a graph G of n vertices with , let , then An internal path of
G is a walk
(
) that the vertices
,
, ⋯,
are distinct (
do not need to distinct),
,
and
for
[
27].
Lemma 10 ([
27])
. Let G be a connected graph that is not isomorphic to and let be the graph obtained from G by subdividing the edge of G. If lies on an internal path of G, then . Lemma 11 ([
27])
. (1) The connected graphs with index 2 are precisely the graphs below: , , and for .(2) The connected graphs with an index of less than 2 precisely include the following graphs: and for .
Woo and Neumaier in [
29] gave the structure of graphs with
. An open quipu is a tree
G of maximum degree 3 such that all vertices of degree 3 lie on a path and a closed quipu is a connected graph
G of maximum degree 3 such that all vertices of degree 3 lie on a circuit and no other circuit exists. The following result can be found in [
29].
Lemma 12 ([
29])
. Let G be a connected graph with . Then, , where denotes the maximum degree of G, and:- (1)
If , then G is an open quipu or a closed quipu.
- (2)
If , then
Note that if
G is an open quipu or a closed quipu, then
may not belong to the interval
, as can be seen in [
29]. For a connected
G with
and
, by Lemma 12 and Theorem 8, we have
,
. The inverse is not true. For example,
, where
is the graph obtained by adding a pendent edge to
and
. Now, we consider the spectral characterizations of two classes of trees
T such that
and
, that is
and
, as can be seen in
Figure 2.
Lemma 13. Let T be a tree and H be a graph with and . If and H has no as its subgraph, , then we have In particular, if
H and
G have the same degree sequence, then
Proof. Since , and , by Lemma 5, we have ; thus, the result holds. □
Remark 5. The parameters of each graph in Figure 3 take the following values: for , for , and for , for , and for , , and for . Sometimes, we shall give the parameters of the graphs above, say, in Lemma 4.9 and the proof of Theorem 4.1, which are and instead of the parameters and of , respectively. For a graph G with , is said to be the product degree of the edge . We denote by the sequence of the product degree of G, that is, , where .
Lemma 14. Let H be a graph with and . Let , , then we have:
- (1)
If , then , the equality holds if and only if H does not contain as its subgraph;
- (2)
If , then , the equality holds if and only if H does not contain as its subgraph;
- (3)
If and H is not a graph in (1) and (2), where and , then ;
- (4)
If , where and , then .
Proof. It is easy to see that each H of (1) and (2) and have the same number of vertices and edges, as well as the same degree sequence. From Lemma 13, by direct computation, we see that (1) and (2) are true.
(3) Here, let , , which do not contain as its subgraph. For each graph , by Lemma 4.6 we have and for all , where , . One can see that for , and , where . Therefore, we have the following:
If , it follows that for , which implies that (3) is true.
If , it follows that for . By the (1) and (2) of the lemma, (3) holds.
If , , it follows that for . By (2) of the lemma, (3) holds.
If with , or with , by the arguments above, we have that (3) holds.
(4) If and and , one sees that and have the same number of edges, the same degree sequence, and the distinct product degree sequence. By the computation of the sum of the product degree, (4) follows from Lemma 13. □
Lemma 15. Let and , then we have:
- (1)
If , then , for ;
- (2)
If , then , for and ;
- (3)
If , then .
Proof. Note that for , and for and , and for . By direct computation, (1) follows from Lemma 13.
For each , by Lemma 13, we have that and for . By lemma (1), it is easy to see that (2) and (3) hold. □
Lemma 16. (1) If , then H and are not cospectral, where , , .
(2) If with and , then and .
Proof. (1) Suppose that and H are cospectral. Then, , and H does not contain odd circles as its subgraphs. By Theorem 5 and Lemma 14 (1), H does not contain and as its subgraphs, which implies that .
By Lemmas 1 and 2, the characteristic polynomials of
and
H can be computed as follows:
Since , by Lemma 4.1 and eliminating the same terms from and , we can write (using Mathematica5.0) and , denoted by and , as follows:
,
Now, we prove the necessity of lemma (1) by comparing the coefficients of the corresponding terms of and .
Case 1. If , we have . Note that , and one sees . Then, the least term of is while the first term of is . Hence, other terms of the same exponents must exist in such that the sum of the coefficients of which plus -2 equals -1. Since and , it is easy to see that all possible candidates are , or . However, the least term of and are different, which is a contradiction.
Case 2. If , substituting into and by eliminating the same terms of and , we have
Note that and . Clearly, the least term of is . Therefore, other terms of the same exponents must exist in such that the sum of the coefficients of which plus 1 equals 2. Since and , it is not hard to see that all possible candidates are , , and . All possible combinations of the terms are: , or . For each combination above, we have that the least term of is and does not vanish from , which is a contradiction.
Case 3. If , the least term of is . Since the first term of is , other terms of the same exponents must exist in such that the sum of the coefficients of which plus -2 equals -1. Since and , it is easy to see that all possible candidates are and . However, we combine the terms, the coefficient of which plus -2 is not -1, which completes the proof of (1).
(2) Since , from , we get
: ,
: .
Therefore, it follows that and . □
Theorem 9. Let . Then, is determined by its adjacency spectrum.
Proof. Let
H be cospectral with
. Suppose that
H has
k connected components
,
, ⋯,
. By Theorems 3.6–3.8, it is not hard to see that
By Theorem 7,
, therefore we have
for
. Choose the middle vertex
v of degree 3 of
such that
. By Lemma 11,
. Thus, by Lemmas 9 and 12 (2), one can see that
and
. On the other hand, it is easy to obtain that
. By Lemma 10,
for all
. Hence, we know that
H only has one component, say
, such that
and
for
. Note that
, by Lemmas 11, 12 and Theorem 8, we have
and for
,
Clearly, for each , for each and for each . Since , we have the following:
(i) , where and ; or
(ii) .
Now we distinguish the following cases.
Case 1. . By Theorems 4 and 5, and . Then, by Theorem 2, , which contradicts .
Case 2. . By Theorems 4 and 5, . By Theorems 2 and 4, . By Lemma 14, we have that if and only if , where does not contain as its subgraphs. From Lemma 16 (1), .
Case 3. . By Theorems 4 and Theorem 5, . Since , by Theorems 2 and 4, , that is . By Lemma 14 (4), we have that , which implies that .
Case 4. H is a tree, that is . By Theorem 3.5, . It therefore follows that . By Lemma 15, , where . Therefore, from Lemma 16 (2). We complete the proof of the theorem. □
From the proof of Theorem 4.1, we obtain a method to find DS graphs by the following steps.
(i) Characterize all connected graphs with and , where and .
(ii) For a given graph G, by applying the properties of invariants, such as , and the number of triangles etc., we find all possible cospectral graphs of G.
(iii) Using the results of the eigenvalues, by considering the coefficients of the characteristic polynomial, we give a proof to whether G is determined by its adjacency spectrum.
Here, we give an example to determine whether the graph
(as can be seen in
Figure 4) is DS by using the above method.
Lemma 17 ([
21])
. For a graph G, let denote the trace of ( or the number of close walks of length k in G), then:- (i)
;
- (ii)
, where is the graph obtained from by adding an edge.
Suppose that H and are cospectral. At first, we give some invariants. From Lemmas 2, 11 and 17 and Theorems 1 and 6, one can obtain the following invariants.
(i) H and F have the same number of vertices, edges and triangles.
(ii) .
(iii) , .
(iv) Two is an eigenvalue of F if and only if .
Then, by using some properties (Theorems 3.2–3.8) of
and the invariants above one give all possible cospectral graphs of
F. Note that
, for
, we have
Finally, using the properties of the eigenvalues (Lemmas 9–11), by considering coefficients of the characteristic polynomial (Lemmas 5 and 8), one can determine the cospectral graphs of F.
From Lemmas 9 and 10, it is not hard to see that for all . By Lemma 5, we have for all and for if and only if , where . Hence, H must be with . Let and , by Lemmas 9 and 10, and it follows that , which implies .
When , by direct computation, we have that is DS. Therefore, we obtain the following result.
Theorem 10. Let and . Then, is determined by its adjacency spectrum.