2
$\begingroup$

Can I make the first eigenvalue of a weighted Laplacian matrix $L$ greater than zero by adding a matrix $E=\operatorname{diag}[1,0 \dots 0]$?

I mean, how can I prove that

$$\lambda_0(L+E)>0\text{ ?}$$

By simulation, it seems to be true.

Thanks.

$\endgroup$

1 Answer 1

1
$\begingroup$

If the associated graph is not connected, there are easy counterexamples (take $4$ vertices, the first and second connected by a single edge, the third and fourth connected by another; then $v = (0,0,1,1)^T$ is in the kernel of both $L$ and $L + E$). So let's assume the graph is connected.

If $0$ is an eigenvalue of $L + E$, that means there is a vector $v\neq 0$ such that

$$(L + E)v = Lv + (v_1,0,\dots,0)^T = 0$$

so $Lv = (-v_1,0,\dots,0)^T$.

Now, we have $(\textrm{Im}T)^\perp = \textrm{Ker}T^*$ for every linear operator $T$. Since $L = L^*$, we have $(\textrm{Im}L)^\perp = \textrm{Ker}L$. We know $\textrm{Ker}L = \textrm{span}(1,1,\dots,1)^T$ (this is where we use connectedness), this tells us that $\textrm{Im}L$ consists of vectors whose entries sum to zero. Thus in writing $Lv = (-v_1,0,\dots,0)^T$ above, we must have $v_1 = 0$, and so actually $Lv = 0$. But as we have just noted $\textrm{Ker}L = \textrm{span}(1,1,\dots,1)^T$, so $v = 0$ and we have a contradiction.

There's probably a cleaner way to observe all this, but that should get you started. To show that this eigenvalue must be positive, note that

$$ v^T(L + E)v = v^TLv + v^TEv = v^TLv + v_1^2 \geq v^TLv \geq 0 $$

so $L + E$ is positive semi-definite (and since you know $\lambda = 0$ is not an eigenvalue, you can conclude it is actually positive definite).

$\endgroup$
4
  • $\begingroup$ Thank you. Yes, it can works for proving that the first eigenvalue is not equal to zero. But what about the sign? I didn't get the last comment about the original entries of L $\endgroup$
    – user48161
    Dec 4, 2014 at 17:39
  • $\begingroup$ Ok, I thought about it some more, and you can conclude the sign is positive, it's been added above. $\endgroup$
    – BaronVT
    Dec 4, 2014 at 17:43
  • $\begingroup$ Can the $ImL$ be the set of only vectors whose entries sum to zero?? $\endgroup$
    – user48161
    Dec 5, 2014 at 9:32
  • $\begingroup$ If the graph is connected, yes. If so, the kernel is the "ones" vector above (you should know some facts along these lines, e.g. zero is an eigenvalue of mult. 1, etc.). The image is then the orthogonal complement, and a vector is orthogonal to "ones" if and only if its entries sum to zero. $\endgroup$
    – BaronVT
    Dec 5, 2014 at 18:49

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .