0
$\begingroup$

Suppose that we have symmetric positive definite matrices $A,B,D \in R^{n \times n}$. Further, we have $$\begin{bmatrix}A&C\\C^T &B \end{bmatrix} > 0$$ and also $$\begin{bmatrix}A B^{-1} A &C\\C^T & B A^{-1} B \end{bmatrix} > 0$$ holds.

Does anyone knows whether $$\begin{bmatrix}(A+D) (B+D)^{-1} (A+D) & C+D\\C^T +D & (B+D) (A+D)^{-1} (B+D) \end{bmatrix} > 0$$ holds? Thank you!

$\endgroup$
2
  • $\begingroup$ Have you done any computer experiment to see that if this conjecture is convincing or not? $\endgroup$
    – user1551
    Jan 13, 2015 at 20:27
  • $\begingroup$ I did some tests and it seems to work, but I do not know how to show this. $\endgroup$
    – Ben
    Jan 14, 2015 at 14:15

1 Answer 1

0
$\begingroup$

The conjecture is wrong. It should be easy to generate a random counterexample using a computer. E.g. I obtained the followings: $$ \begin{bmatrix}A&C\\ C^T&B\end{bmatrix}= \left[\begin{array}{cc|cc} 7&1&0&0\\ 1&1&0&2\\ \hline 0&0&3&1\\ 0&2&1&7 \end{array}\right], \ D=\begin{bmatrix}5&-3\\ -3&3\end{bmatrix}. $$ you may verify that $A,B,D$ as well as the first two block matrices in your question are positive definite, but the last block matrix in your question has a negative eigenvalue.

$\endgroup$
1
  • $\begingroup$ Thanks! I made a mistake in the experiments. $\endgroup$
    – Ben
    Jan 14, 2015 at 16:39

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.