8
$\begingroup$

For Hermitian matrices $X, Y$, I write $X\ge Y\ge 0$ to mean $X-Y$ and $Y$ are positive semidefinite.

In Lemma 2.5 of [Linear Algebra Appl. 452 (2014) 1-6] I proved that if $X + Y\ge W + Z$, $X\ge W\ge Y\ge 0$ and $X\ge Z \ge Y\ge 0$, then $$\det X+\det Y\ge \det W+\det Z.$$

I was thinking of relaxing the condition a little bit.

Is the same inequality true under only the condition $X + Y\ge W + Z$, $X\ge Y\ge 0$, $X\ge W\ge 0$, $X\ge Z\ge 0$?

$\endgroup$
2
  • 4
    $\begingroup$ WLOG, $X=I$. After that, WLOG, $Y$ is diagonal. After that, forget about determinants for $Z,W$, just look at the products of diagonal entries. Am I missing anything? $\endgroup$
    – fedja
    Jul 1, 2014 at 3:56
  • $\begingroup$ @fedja: It is right to assume $X=I$ and $Y=D$ to be diagonal. Oh, it seems that I have an answer. $\endgroup$
    – M. Lin
    Jul 1, 2014 at 19:10

1 Answer 1

2
$\begingroup$

The answer is yes. I thank fedja for discussion.

WLOG, assume $X=I$ (we may do so by pre-post multiplying both sides by $\det X^{-1/2}$, with a standard continuity argument).

After this, we may further assume $Y=D$ to be diagonal (by unitary similarity).

So the question is equivalent to showing if $I+D\ge W+Z$ and $I\ge D, W, Z\ge 0$, then $1+\det D\ge \det W+\det Z$.

Note that $I+D\ge W+Z$ implies $I+D\ge diag(W)+diag(W)$, where $diag(\cdot)$ means the diagonal part.

By a simple induction on the size of the matrix, one gets $1+\det D\ge \det (diag(W))+\det (diag(Z))$. The conclusion follows by applying the Hadamard inequality, $\det (diag(W))\ge \det W$.

$\endgroup$
1
  • $\begingroup$ Could you give more details of the induction step? Thanks in advance. $\endgroup$
    – Tadashi
    Jul 13, 2014 at 20:27

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.