Trees with a large Laplacian eigenvalue multiplicity
S. Akbari, E.R. van Dam, M.H. Fakharan

TL;DR
This paper investigates the multiplicities of non-integer Laplacian eigenvalues in trees, providing bounds and characterizing trees with high eigenvalue multiplicities, highlighting the importance of algebraic connectivity.
Contribution
It offers new bounds on the multiplicities of non-integer Laplacian eigenvalues in trees and characterizes trees with near-maximal multiplicities, focusing on the role of algebraic connectivity.
Findings
Upper bounds on eigenvalue multiplicities are established.
Trees with high multiplicities are characterized.
The role of algebraic connectivity in eigenvalue multiplicities is emphasized.
Abstract
In this paper, we study the multiplicity of the Laplacian eigenvalues of trees. It is known that for trees, integer Laplacian eigenvalues larger than are simple and also the multiplicity of Laplacian eigenvalue has been well studied before. Here we consider the multiplicities of the other (non-integral) Laplacian eigenvalues. We give an upper bound and determine the trees of order that have a multiplicity that is close to the upper bound , and emphasize the particular role of the algebraic connectivity.
Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsGraph theory and applications · Synthesis and Properties of Aromatic Compounds · Interconnection Networks and Systems
Trees with a large Laplacian eigenvalue multiplicity
S. Akbari
Department of Mathematical Sciences, Sharif University of Technology, Iran
,
E.R. van Dam
Department of Econometrics and O.R., Tilburg University, The Netherlands
and
M.H. Fakharan
Department of Mathematical Sciences, Sharif University of Technology, Iran
Abstract.
In this paper, we study the multiplicity of the Laplacian eigenvalues of trees. It is known that for trees, integer Laplacian eigenvalues larger than are simple and also the multiplicity of Laplacian eigenvalue has been well studied before. Here we consider the multiplicities of the other (non-integral) Laplacian eigenvalues. We give an upper bound and determine the trees of order that have a multiplicity that is close to the upper bound , and emphasize the particular role of the algebraic connectivity.
Key words and phrases:
Laplacian spectrum, trees, multiplicities of eigenvalues
2010 Mathematics Subject Classification:
05C50
1. Introduction
In this paper, we study the multiplicity of the Laplacian eigenvalues of trees. In particular, we are interested in upper bounds for these multiplicities, and trees with multiplicities that are close to these upper bounds.
Let be a graph with Laplacian eigenvalues . Fiedler [8] showed that if and only if is connected, where denotes the multiplicity of Laplacian eigenvalue . Grone, Merris, and Sunder [10, Prop. 2.2] proved that if is a connected bipartite graph (in particular, if it is a tree). On the other hand, for the complete graph , the multiplicity of is large and indeed is . How large can the multiplicities be when we restrict to trees? It is shown by Grone et al. [10, Thm. 2.1] that if an integer is a Laplacian eigenvalue of a tree of order , then and . Interestingly, it is different for the Laplacian eigenvalue , because for all graphs , where and are the number of pendant vertices and quasipendant vertices, respectively, see [7, p. 263]. For example, it follows that if is a star (), then . Guo, Feng, and Zhang [12] showed that if is a tree of order , then and for every integer and every there exists a tree of order with . Also Barik, Lal, and Pati [2] studied the multiplicity of Laplacian eigenvalue in trees. In this paper we consider the multiplicities of the other (non-integral) Laplacian eigenvalues.
Our paper is further organized as follows. In Section 2.1, we introduce some notation and definitions. We then collect some relevant results from the literature in Section 2.2. In Section 3, we introduce the family of trees that we will call spiders, as they will play a key role in the main part of the paper, Section 4. In this main part we will give an upper bound on the multiplicities of a Laplacian eigenvalue of a tree of order , in particular for the algebraic connectivity, and characterize the trees that are close to this upper bound. One main result (Theorem 4.1) — and starting point of several refinements — is that for a tree of order , the multiplicity of a Laplacian eigenvalue is at most , and we characterize the trees that attain this bound. We will also characterize the trees that have a multiplicity that is close to the upper bound (in Theorems 4.5 and 4.6), and show that there are finitely many trees besides a specific family of spiders for which the algebraic connectivity has a multiplicity within a fixed range from the upper bound (in Theorem 4.4).
2. Preliminaries
2.1. Notation
Let be a graph, where is the vertex set and is the edge set. Throughout this paper all graphs are simple, that is, without loops and multiple edges. Let be the degree of . We denote the non-increasing vertex degree sequence of by . The maximum and minimum degree of are denoted by and , respectively.
The adjacency matrix of , denoted by , is an matrix whose -entry is if and are adjacent and [math] otherwise. We call the Laplacian matrix of , where is the diagonal matrix with . The eigenvalues of are called the (Laplacian) eigenvalues111Whenever we write about eigenvalues of a graph, we mean Laplacian eigenvalues (unless explicitly stated otherwise). of , and we denote these in increasing order by . The multiset of eigenvalues of is called the (Laplacian) spectrum of . Fiedler [8] called the algebraic connectivity of . The multiplicity of a Laplacian eigenvalue in a graph is denoted by ; the number of Laplacian eigenvalues of in an interval is denoted by . We say that two (or more) Laplacian eigenvalues are conjugates (of each other) if they are roots of the same irreducible factor of the characteristic polynomial of the Laplacian (over the rationals). Conjugate eigenvalues have the same multiplicity.
A star (graph) is a complete bipartite graph , for some positive integer . We let be the path of order . The diameter of is denoted by . A vertex of degree one is called a pendant vertex and a vertex that is adjacent to at least one pendant vertex is called a quasipendant vertex. The number of pendant and quasipendant vertices of a graph are denoted by and , respectively.
In all of the above notation, we remove the additional or if there is no ambiguity; for example instead of , or instead of .
2.2. A collection of elementary results
In this section we collect and extend some relevant basic lemmas from the literature. We start with some general ones. For other basic results, we refer to the books by Brouwer and Haemers [4], Cvetković, Doob, and Sachs [5], and Cvetković, Rowlinson, and Simić [6].
Lemma 2.1**.**
[6, Prop. 7.5.6].* If is a connected graph with distinct Laplacian eigenvalues, then .*
Lemma 2.2**.**
[3, Thm. 1]. Let be a graph of order , with vertex degrees and Laplacian eigenvalues . If is not , then for . In particular, if has at least one edge, then and if has at least two edges, then .
We recall that a sequence interlaces another sequence with whenever , for . It is well known [13, Thm. 1] that the eigenvalues of a principal submatrix of a Hermitian matrix interlace the eigenvalues of . For the Laplacian eigenvalues, we moreover have the following two specific results.
Lemma 2.3**.**
[10, Thm. 4.1].* Let be a graph of order and let be a (spanning) subgraph of obtained by removing just one of its edges. Then the largest Laplacian eigenvalues of interlace the Laplacian eigenvalues of .*
A consequence of this is the following.
Lemma 2.4**.**
[10, Cor. 4.2].* Let be a pendant vertex of and let . Then the Laplacian eigenvalues of interlace the Laplacian eigenvalues of .*
The following results concern multiplicities.
Lemma 2.5**.**
[7, p. 263]**. Let be a graph with pendant vertices and quasipendant vertices. Then .
This result follows from the observation that for every pair of pendant vertices that is adjacent to the same quasipendant vertex, the difference of the corresponding characteristic vectors is an eigenvector of for eigenvalue . This gives linearly independent eigenvectors.
Finally, we have some specific results for trees.
Lemma 2.6**.**
[10, Thm. 2.3].* Let be a tree with pendant vertices. If is a Laplacian eigenvalue of , then .*
The following is a clear generalization to non-integral eigenvalues of a result by Grone, Merris, and Sunder [10, Thm. 2.15]. It uses the Matrix-Tree theorem, which states that any cofactor of the Laplacian matrix of equals the number of spanning trees of . In case of trees, this is equivalent to the fact that the product of all non-zero Laplacian eigenvalues equals the number of vertices; we shall use this also later.
Lemma 2.7**.**
Let be a Laplacian eigenvalue of a tree with . Then the product of and its conjugate eigenvalues equals . In particular, if is an integer, then .
Proof.
Let be a pendant vertex. Then the Laplacian matrix of has the form
[TABLE]
where is the principal submatrix corresponding to . Since , there is an eigenvector of for such that its last component is [math]. If is the vector obtained from by deleting the last component, then it is not hard to see that . So is an eigenvalue of as well (and so are its conjugates). By the Matrix-Tree Theorem however, we have that , and the result follows. ∎
To conclude this section, we mention a bound for the multiplicity of the algebraic connectivity of a tree. It follows easily from a result of Grone and Merris [11].
Proposition 2.8**.**
Let be a tree with . Then .
Proof.
Let , and suppose that . Because the multiplicity is at least , there is an eigenvector that has value [math] corresponding to one of the vertices, and so is a so-called type I tree (see [11]). By [11, Thm. 2], the so-called characteristic vertex of has degree at least , but this is at least , which is a contradiction. ∎
3. Spiders and their spectra
In this section we define two families of trees that are most relevant to our results. We start with the main one, i.e., the family of trees that have large multiplicities for some non-integral Laplacian eigenvalues.
The spider , with , is defined as in Figure 1. It is obtained from the star by extending of its rays (legs) by an extra edge, and has vertices. We say that the spider has legs.
Proposition 3.1**.**
The Laplacian spectrum of is
[TABLE]
where and , are the roots of .
Proof.
In Figure 1, an eigenvector for eigenvalue is shown222Whenever we use and in this paper, we mean the specific values and .. It is clear that there are linearly independent such eigenvectors corresponding to the eigenvalue . Similarly, we find linearly independent eigenvectors corresponding to eigenvalue . Because , there are only two other eigenvalues . Because the product of all non-zero eigenvalues equals , it follows that . Noting the trace of (which equals twice the number of edges) it follows that , and the result follows. ∎
Note that if we let , then and , so for the algebraic connectivity equals and . Proposition 3.1 thus shows that the upper bound of Proposition 2.8 is sharp.
Proposition 3.2**.**
The Laplacian spectrum of with is
[TABLE]
where and , , are the roots of .
Proof.
As before, we have eigenvalues and with multiplicities at least , and eigenvalue [math]. In addition, we have eigenvalue with multiplicity at least by Lemma 2.5. Thus, three eigenvalues are left. Again, from the product of all non-zero eigenvalues being equal to , it follows that . Moreover, . A quadratic equation easily follows from the trace of :
[TABLE]
for every graph of order with edges, and degrees . For , we obtain that , and from this we conclude that . From all of this, the spectrum follows. ∎
Note that for a fixed integer , one can use induction to show that and , for , starting from Proposition 3.1 and by inductively applying Lemma 2.4. On the other hand, for fixed , it also follows by induction that for all . Consequently, we have the following.
Lemma 3.3**.**
Let , with . Then and .
It also follows from the above that for . Below we will show that the spider graphs , with are extremal regarding the eigenvalues and .
Observe first however that if then and . So in this case, and are the roots of and the spectrum of is
[TABLE]
A second family of trees that we will use is shown in Figure 2. Such a tree is obtained by attaching pendant vertices to one end point of the path on vertices, and pendant vertices to its other end point, for some positive integers . We note that for , we obtain a star (a tree with diameter ). The Laplacian spectrum of this graph is . Also, every tree of order with diameter is a graph for some integers such that . The spectrum of this graph can easily be obtained in a similar way as in Proposition 3.2.
Lemma 3.4**.**
[9, Prop. 1]**. Let be of order . Then the characteristic polynomial of is .
In order to prove some of the results in the next section, we need the following two half-known characterizations of trees with extremal Laplacian eigenvalues . The first is a result by Li, Guo, and Shiu [14]. For completeness, we provide a shorter proof of this result.
Proposition 3.5**.**
[14, Thm. 3.4].* Let be a connected graph of order . Then if and only if for some positive integers with .*
Proof.
One side is clear by Lemma 3.3. To show the other side, suppose that . If , then is a path or a cycle. Since , it follows from interlacing (Lemmas 2.3 and 2.4) that the order of is less than , and then it can be checked that . Next, assume that is a vertex of with degree . If there exists a vertex at distance from with degree at least , then the graph can be obtained from the fork in Figure 3 by adding some pendant vertices and some edges. But and so again by interlacing, we have that , which is a contradiction. Thus, every vertex of degree at least is adjacent to . By Lemma 2.2, we have , hence . So for some integers . By Proposition 3.2, it is necessary to have . ∎
We note that the proof actually shows that if , then is a star or a spider. For more details and the full classification of graphs with , we refer to Li, Guo, and Shiu [14]. The following similar result is a strengthening of a result by Zhang [15, Thm. 2.12].
Proposition 3.6**.**
Let be a tree of order . If , then is a star, a spider, , or , and equality holds if and only if for some positive integers , with .
Proof.
Suppose that . By interlacing (Lemma 2.4), the tree cannot have as a subgraph since , so . If is not a star, then it must have diameter or . If , then is for some positive integers . Using Lemma 3.4, we find however that and , so it follows again by interlacing that in this case can only be one of the trees , , and , for some positive integer . Indeed, , , and .
Finally, suppose that . For the fork (see Figure 3), we have . So cannot have as a subgraph, and thus is a spider. The case of equality now follows from Lemma 3.3 and the fact that a star, , , and all have . ∎
4. Trees with a large multiplicity
We are now ready for our main results, that is, to bound the multiplicities of non-integer Laplacian eigenvalues of trees, and to characterize the trees with large Laplacian eigenvalue multiplicities.
Theorem 4.1**.**
Let be a tree of order with a Laplacian eigenvalue . If , then , and equality holds if and only if and . In particular, if equality holds, then is the algebraic connectivity of or its conjugate.
Proof.
Suppose that and . Then . So by Lemma 2.7, the eigenvalue is not an integer, so it has at least one conjugate eigenvalue with . So . If is odd, then , so has exactly distinct eigenvalues. This implies that the diameter of is at most (by Lemma 2.1), so is a star, with spectrum , which is a contradiction.
So is even and . Hence has spectrum
[TABLE]
for some eigenvalue . Now, by Lemma 2.1, has diameter at most , and it is not a star. So has pendant vertices and quasipendant vertices. Thus, , by Lemma 2.5, which is a contradiction, so .
Now, suppose that . Again, by Lemma 2.7, the eigenvalue is not an integer. If , then possibly can have two conjugate eigenvalues. However, this would imply that has four distinct eigenvalues, which would give a contradiction in the same way as in the above case of even. So and has exactly one conjugate eigenvalue . By Lemma 2.7, we have . So the spectrum of is
[TABLE]
for some eigenvalues and . Since the product of all non-zero eigenvalues of a tree equals , we find that . Because the only tree on vertices with an eigenvalue is the star (which easily follows by observing that the complement of such a tree should be disconnected), it follows that , and hence by Lemma 2.5, we deduce that . Moreover, by Lemma 2.1 the diameter of is at most . Because is odd, it now easily follows that is isomorphic to . As observed in Proposition 3.1, this graph indeed has an eigenvalue (in fact, the two conjugates with multiplicity . ∎
Note that if is a tree but not a star, then . Because , it then follows by interlacing that , so we have the following.
Corollary 4.2**.**
If is a tree of order but not a star, then .
The multiplicity of in a star equals and by Corollary 4.2, for every other tree . So there is a huge gap between these multiplicities and . Lemma 3.3 and Proposition 3.2 show that there are no other gaps. In fact, for any , the tree has .
In Theorem 4.4, we will show that for every positive integer , all except finitely many trees with are spiders. But first, we determine the typical values of the Laplacian eigenvalues with large multiplicity.
Lemma 4.3**.**
Let be an integer with . If is a tree of order and for a Laplacian eigenvalue , then .
Proof.
Suppose that and . Because cannot be integer, there exists another (conjugate) eigenvalue with . Since there can be no other (conjugate) eigenvalue with such a large multiplicity. Thus, and are the roots of a quadratic factor of the Laplacian polynomial, so equals a positive integer , say. Because by Lemma 2.7, it follows that . Since , we have that
First suppose that . If is even, then it follows that , which is a contradiction. Similarly, if is odd, then ; again a contradiction. Next, suppose that . Note that by Proposition 2.8 and Lemma 2.2, we have and , and hence . Note also that is a simple eigenvalue by [10, Prop. 2.2], so it is not equal to or . If is even, then
[TABLE]
and so , which is a contradiction. Similarly, if is odd, then
[TABLE]
and so ; again a contradiction.
Thus, and hence . ∎
Theorem 4.4**.**
Let be an integer with and let be a tree of order . Then if and only if .
Proof.
If , then is not a star, so , and hence by Lemma 4.3, we have . Now, by Proposition 3.6, it is clear that for some suitable and . These integers are now determined by the multiplicity of (see Propositions 3.1 and 3.2), which finishes the proof. ∎
Observe the contrast between Theorems 4.1 and 4.4 in the sense that in the latter we restrict to eigenvalue . Indeed, the following examples show that this restriction is necessary, at least for .
Let and let be the tree obtained from by adding one pendant vertex to one of the legs. See Figure 4 for the case . It is clear from Proposition 3.6 that , and it follows from interlacing (Lemma 2.4) that equals or . We claim that this multiplicity equals . In order to show this, we first consider the case and show that is not an eigenvalue of for any .
Indeed, suppose that is an eigenvalue of with eigenvector . By normalizing the entry of the top right vertex in Figure 4, and applying the equation , we recursively find the entries of as given in the figure (we omit the technical details; note also that it easily follows that cannot be zero). But then finally for the bottom two right vertices (where we obtained ) we should have the equation , which gives a contradiction.
Next, consider the general case , and suppose that it has eigenvalue with multiplicity , for . By interlacing it then follows that has eigenvalue with multiplicity at least , and by repeating this, we find that has eigenvalue with multiplicity at least , which is a contradiction, and hence confirms our claim.
For even , we can now take for , to obtain . For odd , we take for , to obtain . Thus, for every there are infinitely many trees with such multiplicities.
In the given examples, we cannot take . Indeed, for this case we can prove something stronger than in Theorem 4.4, and obtain a result that is similar to Theorem 4.1. Here we will show that there are unique trees of order with multiplicities and for an eigenvalue .
Theorem 4.5**.**
Let be a tree of order and be a Laplacian eigenvalue of . Then if and only if and . In particular, in this case is the algebraic connectivity of or its conjugate.
Proof.
One side of the equivalence is clear from Proposition 3.2. To show the other side, we will first show by induction on (with even) the claim that if , then and that . We checked that this is true for by enumerating all trees of this order with the Sage computer package. For , consider a tree with . Consider one of its extremal quasipendant vertices (in the sense that becomes pendant in the tree that is obtained from by removing all its pendant vertices). Now, remove the edge that connects to its (unique) non-pendant neighbor. The remaining graph is a disjoint union of a star (with center ) and a tree on vertices, say. By interlacing (Lemma 2.3), it follows that . But by Theorem 4.1, so it follows that . Because , we have two cases. If , then , and it follows from Theorem 4.1 that indeed and , which implies by interlacing that . If , then , so by induction and , but the latter (and interlacing) again implies that , which finishes the proof of our claim.
Just like in the proof of Theorem 4.4, we can now apply Proposition 3.6 to finish the proof. ∎
We omit the proof of the following similar result, as the proof is very similar, although we also have to use Theorem 4.5 now.
Theorem 4.6**.**
Let be a tree of order and be a Laplacian eigenvalue of . Then if and only if and . In particular, in this case is the algebraic connectivity of or its conjugate.
We note that one could try to extend this result further, and indeed, the induction steps work over and over again, so induction would give a theorem about multiplicity for each , if only the bases of the induction steps would be true. Of course, this is where things go wrong. For example, we cannot prove a similar result for multiplicity and because we have counterexamples such as on vertices. Indeed, there are five such counterexamples, and we depict them in Figure 5. Besides (Fig. 5a), there are (exactly) two other trees of order that have eigenvalue with multiplicity and (Fig. 5b and c). Moreover, there is one tree of order with eigenvalues with multiplicity (Fig. 5d), and one tree of order that has the roots of as eigenvalues with multiplicity (Fig. 5e). The variety of these examples shows that further classification of trees with an eigenvalue other than with multiplicity for (on top of Lemma 4.3 and Theorem 4.4) seems unfeasible.
Acknowledgement. The research of the first author was partly funded by Iran National Science Foundation (INSF) under the contract No. 96004167. Also, part of the work in this paper was done while the third author was visiting Tilburg University.
The reference list from the paper itself. Each links out to its DOI / PubMed record.
- 1[1]
- 2[2] S. Barik, A.K. Lal, S. Pati, On trees with Laplacian eigenvalue one, Linear and Multilinear Algebra. 56:6 (2008) 597-610.
- 3[3] A.E. Brouwer, W.H. Haemers, A lower bound for the Laplacian eigenvalues of a graph-Proof of a conjecture by Guo, Linear Algebra Appl. 429 (2008) 2131-2135.
- 4[4] A.E. Brouwer, W.H. Haemers, Spectra of Graphs, Springer, New York, 2012; http://homepages.cwi.nl/~aeb/math/ipm/ .
- 5[5] D. Cvetković, M. Doob, H. Sachs, Spectra of Graphs, Academic Press, New York, 1980.
- 6[6] D. Cvetković, P. Rowlinson, S. Simić, An Introduction to the Theory of Graph Spectra, Cambridge University Press, New York, 2010.
- 7[7] I. Faria, Permanental roots and the star degree of a graph, Linear Algebra Appl. 64 (1985) 255-265.
- 8[8] M. Fiedler, Algebraic connectivity of graphs, Czechoslovak Mathematical Journal. 23 (2) (1973) 298-305.
