3
$\begingroup$

Assume that $B$ and $A$ are two positive definite matrices. Take $B^*$ a block diagonal matrix with block $B_{11}$ and $B_{22}$ of $B$. This means the following: $$ B=\left[\begin{array}{ll} B_{11}& B_{12}\\ B_{21}&B_{22} \end{array}\right], B^*=\left[\begin{array}{ll} B_{11}& 0\\ 0 &B_{22} \end{array}\right] $$ It can be seen that $2B^*\succ B$ and therefore: $$ \det(I+2AB^*)\geq \det(I+AB). $$ So I managed to prove that, however I have another conjecture that the following is true: $$ 2\det(I+AB^*)\geq \det(I+AB). $$ My question is about the proof of this conjecture.

$\endgroup$

2 Answers 2

7
$\begingroup$

A quick counterexample to your conjecture is

\begin{equation*} A = \begin{pmatrix} 13 & 3 & -13 & -5\\ 3 & 4 & -3 & 4\\ -13 & -3 & 13 & 5\\ -5 & 4 & 5 & 10\\ \end{pmatrix},\quad B = \begin{pmatrix} 15 & 3 & -14 & -1\\ 3 & 18 & -6 & 3\\ -14 & -6 & 19 & -1\\ -1 & 3 & -1 & 1 \end{pmatrix} \end{equation*}

For this choice, $2\det(I+AB^*) \approx 6\times 10^4$, while $\det(I+AB)\approx 8\times 10^4$. The $A$ above is semidefinite; if that displeases you, then just add a trivial $\epsilon I$ to it, the counterexample will still hold.

$\endgroup$
2
  • $\begingroup$ Many thanks for the answer. Unfortunately I forgot to write an important assumption about $B$. All diagonal elements are equal: $b_{11}=...=b_{nn}=K$. I am not sure if this is significant or not for the problem. As a matter of fact I was trying to find an upper bound like $C\det(I+AB^*)\geq \det(I+AB)$ for some $C$ independent of the dimension of $B$. I wonder if this is possible. My numerical evaluation of the current problem shows that with this additional assumption, we can have $C=2$. I might have mistaken somewhere. Should I post this as question, if it can be an independent problem? $\endgroup$
    – Arash
    Sep 26, 2013 at 10:45
  • $\begingroup$ Even with the all diagonal elements equal this does not hold; take the matrix $B$ that I have above, set its diagonal to 15 (it still remains posdef); the conjecture still fails. Perhaps for it to hold $C$ has to depend on $n$ (the size of the matrices)---but am not sure about that. You might want to pose it as a separate problem. $\endgroup$
    – Suvrit
    Sep 26, 2013 at 16:15
1
$\begingroup$

From the comments, a true version is $$2^n\det(I+AB^*)\geq \det(I+AB), \quad (1)$$ where $n$ is the dimension of $A, B$.

The proof is easy, as $2^n\det(I+AB^*)\geq\det(I+2AB^*)$. Can $2$ be replaced by a smaller constant in (1)?

$\endgroup$

Your Answer

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

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