3
$\begingroup$

This is a follow up question on my previous question here that was on solved in the deterministic setting by Denis Serre, when the perturbation can be separated. Therefore, I decided to split the deterministic case from the random question initially posed in the very same question. So let me state the random case:

We consider a symmetric matrix $A \in \mathbb R^{n \times n}$ defined by

$$A_{ij}= i \delta_{ij} - X_{ij}\varepsilon \,.$$

Here $\delta_{ij}$ is the Kronecker delta and $X_{ij}$ are iid Bernoulli (but of course $X_{ij}=X_{ji}$ to make it hermitian).

We first note that this matrix is not diagonally dominant if the matrix dimension $n$ is large enough, independent of what $\varepsilon$ is.

This is because $\lim_{n \rightarrow \infty} \sum_{i=1}^{n} \vert A_{i,1}\vert=\infty >\vert A_{1,1} \vert.$

Numerical experiments with matrix size $n=50$ or $n=200$ show that the lowest eigenvalue stays far above $0$, if $\varepsilon>0$ is sufficiently small.

Question: How can one show that $A$ is positive definite independent of the dimension if $\varepsilon$ is sufficiently small but fixed ?

To elaborate on the numerical experiments. We find that the lowest eigenvalue of $A$ in some interval $[0,96,1.08]$ if we choose $\varepsilon=0.1$ and $n=50$(upper plot) and $n=200$(lower plot), as I illustrate in the following plot where I sampled the lowest eigenvalue for 100 realizations:

k=50

k=200

Please let me know if you have any questions!

$\endgroup$
6
$\begingroup$

Let $u\in \mathbb{R}^n$ as $u_i=i^{-1}$. Then $$ \mathbb{E}\big[\langle u Au\rangle \big]=\sum_{i=1}^n i^{-1}-\frac{\epsilon}{2}\sum_{i,j\leq n}\frac{1}{ij}\approx \log n -\frac{\epsilon}{2}(\log n)² $$ Which is stricly negative for $\log n > 2\epsilon^{-1}$. (Remark : In order to see it numerically one should choose $n$ exponentially large with $\epsilon^{-1}$.)

$\endgroup$
0
$\begingroup$

You can find other counter-examples along the vein of $v\in \mathbb{R}^n$ $v_i=i^a$ for certain $a$, making sure that the sum of the negative terms outweighs that of the positive components of the diagonal elements. Also, a matrix may have all positive eigenvalues yet not be positive definite; see <https://math.stackexchange.com/questions/4336/if-eigenvalues-are-positive-is-the-matrix-positive-definite>.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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