4
$\begingroup$

Let $A\in\mathbb{R}^{n\times n}$ be a positive semidefinite matrix and $D\in\mathbb{R}^{n\times n}$ be a diagonal positive definite matrix. Let $\mathrm{diag}(X)\in\mathbb{R}^{n\times n}$ denote the diagonal part of a matrix $X\in\mathbb{R}^{n\times n}$.

My question: Does the following inequality hold true $$ \mathrm{diag}\left[(A+D)^{-1}\right] \ge \left[\mathrm{diag}(A)+D\right]^{-1}\ \ ? $$

For the sake of clarity, I describe below a simple example. Consider $$ A=\begin{bmatrix}2 & 1 \\ 1 & 3\end{bmatrix}, \quad D=\begin{bmatrix}0.5 & 0 \\ 0 & 2\end{bmatrix}. $$ We have $$ \mathrm{diag}\left[(A+D)^{-1}\right] = \begin{bmatrix}0.2927 & 0\\0 & 0.1341 \end{bmatrix}, \quad \left[\mathrm{diag}(A)+D\right]^{-1} = \begin{bmatrix}0.1818 & 0 \\0 & 0.0833 \end{bmatrix}, $$ so that $\mathrm{diag}\left[(A+D)^{-1}\right] \ge \left[\mathrm{diag}(A)+D\right]^{-1}$.

(If true, the above inequality perhaps is just a well-know inequality but I wasn't able to find it in any "classical" matrix theory textbooks.)

Improve this question
$\endgroup$
4
  • $\begingroup$ The inequality is equivalent to $\operatorname{diag}\left(B^{-1}\right) \geq \left(\operatorname{diag} B \right)^{-1}$ for all positive definite $B$, right? $\endgroup$
    – darij grinberg
    May 19, 2017 at 15:36
  • $\begingroup$ @darijgrinberg: Yes, if you take $B=A>0$ and $D=0$. Otherwise I don't see the equivalence. $\endgroup$
    – Ludwig
    May 19, 2017 at 15:40
  • 2
    $\begingroup$ To get yours from mine, set $B = A + D$. To get mine from yours, set $A = B$ and $D = 0$. Setting $D = 0$ is harmless, since positive semidefinite matrices are limits of positive definite matrices. $\endgroup$
    – darij grinberg
    May 19, 2017 at 15:46
  • $\begingroup$ @darijgrinberg: Yes, absolutely right. $\endgroup$
    – Ludwig
    May 19, 2017 at 15:54

1 Answer 1

Reset to default
9
$\begingroup$

This is true. As Darij suggests, denote $B=A+D$, let $e=e_i$ be a basic unit vector. We have to prove $(B^{-1}e,e)\geqslant (Be,e)^{-1}$. Denote further $C=B^{1/2}$, this rewrites $\|Ce\|\cdot \|C^{-1}e\|\geqslant 1$. This follows from Cauchy-Bunyakovsky-Schwarz, as $(Ce,C^{-1}e)=(e,e)=1$.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

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