Take the 2-minute tour ×
MathOverflow is a question and answer site for professional mathematicians. It's 100% free, no registration required.

Suppose you have a real positive definite matrix $A$ who eigenvalues are $\lambda_{1} \leq \lambda_{2} \leq \ldots \leq \lambda_{n}$. I am interested in bounding from below $\lambda_{2}-\lambda_{1}$. (It is known that $\lambda_{1}$ is simple). Are there any known methods for this kind of problem?

share|improve this question
    
Do you have any other information about A? As stated, the difference that you want to bound can be made arbitrarily small. –  Vidit Nanda Oct 22 '12 at 11:19
    
Well, $A$ is in fact a signless Laplacian of a graph. –  Felix Goldberg Oct 22 '12 at 11:37
1  
Searching google for "arbitrarily small spectral gap" produces this paper: <www.mis.mpg.de/fileadmin/jjost/abj21-6-04.pdf> which claims to construct graphs of arbitrarily small spectral gap given mild conditions on the degrees of the vertices. What is your starting point? If you know each entry of $A$ then you can presumably compute the eigenvalues and hence the gap. On the other hand if you want a theorem that bounds the gap for some general $A$, then there is no hope without more assumptions on the graph. –  Vidit Nanda Oct 22 '12 at 12:43
1  
One could get bounds depending on some feature of $A$. For instance, we all know that $a^2+b^2$ can get arbitrarily small for positive $a$ and $b$, but nevertheless it is interesting to discover the bound $2ab\leq a^2+b^2$. –  Federico Poloni Oct 22 '12 at 13:50
    
Vidit Nanda, OP is talking about the signless Laplacian, you are talking about the usual Laplacian, it seems. Their spectrum is quite different. For instance, it is known that the multiplicity of 0 as an eigenvalue of the signless Laplacian agrees with the number of connected components that are bipartite. In other words, if the graph is bipartite and connected, then OP's question simply reads: "Are there bounds on the second smallest eigenvalue?" –  Delio Mugnolo Oct 25 '13 at 9:13
add comment

2 Answers

You mention that $A$ is the Laplacian of a graph. The field of spectral graph theory is well established, and I think there are several techniques that can be used to bound the spectral gap of a graph Laplacian for an undirected weighted graph.

Perhaps the most straightforward is known as Cheeger's inequality which controls the spectral gap, $\lambda_2 - \lambda_1$ by the Cheeger constant of the graph. See the wikipedia entry for Cheeger constant: http://en.wikipedia.org/wiki/Cheeger_constant_(graph_theory)

If you might also consider trying a Log sobolev inequality: http://projecteuclid.org/DPubS?service=UI&version=1.0&verb=Display&handle=euclid.aoap/1034968224

I hope this helps.

share|improve this answer
    
But I am interested in the signless Laplacian - there is a Cheeger-type result for them (Desai & Rao, 1994) but I haven't seen anything for the gap. Thanks anyway! –  Felix Goldberg Oct 25 '13 at 12:32
add comment

There are many articles on spectral theory for signless Laplacians $Q$, but the most complete is as far as I know Towards a spectral theory of graphs based on the signless Laplacian, II

In that paper, which is three years old, Cvetkovic and Simic point out why in their opinion not the second smallest is important, but the second largest $Q$-eigenvalue is important (hint: it corresponds to the spectral gap). Most researchers seem to share their opinion, because as far as I know the answer to your question is unknown.

The best available piece of information is, to my knowledge, the bound on the smallest $Q$-eigenvalues based on the largest (signed) Laplacian eigenvalue and the maximal and minimal degrees of the graph that is presented in Bipartite subgraphs and the signless laplacian matrix.

share|improve this answer
    
Thanks - I am familiar with these papers, though, and with this one too: arxiv.org/abs/1307.7749 :) –  Felix Goldberg Oct 25 '13 at 12:30
add comment

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.