Let $X = A^{\top} D A$ where $A$ is a rectangular matrix such that $A^{\top}A$ is positive definite (can assume strictly positive definite) and $D$ is a diagonal matrix (feel free to assume all elements of $D$ are positive and less than $1$).

I am interested in lower bounds on $\lambda_{\min}\left(X\right)$ that depend on any norm of the diagonal of $D$ (in the problem I am working on, the norm of the diagonal of $D$ will approach $\infty$ and I want to show the minimum eigenvalue diverges).

Without further restriction this is not possible because letting $A = I$, the identity matrix,

$\lambda_{\min}\left(X\right) = \min D_{ii}$.

Which is not a norm of $D$.

Let $n$ be the number of rows of $A$ and $m$ be the number of columns (assume $n > m$) and let $a_{i}$ be the $i$'th row of $A$. My basic thought about why there might be a solution is that:

$X = \sum_{i=1}^{n} a_{i} a_{i}^{\top} d_{i}$.

Intuitively: if we required that $a_{i} a_{i}^{\top}$ had to be a "full" matrix with only non-zero elements than it seems that any $d_{i}$ becoming large would make each element of the $X$ matrix large and so the smallest eigenvalue would be large.

Bottom line: I am looking for a lower bound on the minimum eigenvalue such that the bound becomes large as the norm of the diagonal of $D$ becomes large. Is there any such lower bound for some conditions on $A$?

Additionally: or if the lower bound can depend only $||A^{\top} D||$, that would answer my question as well.


There cannot be a better lower bound than $\lambda_\min(X) = \min D_{ii}$: If there is a vector $x\ne0$ such that $DAx=\lambda_\min(D)Ax$, then the estimate is sharp.


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.