1
$\begingroup$

I had been working on this problem for days and yet to reach a conclusion.

Given a matrix $M = 1^Ty\thinspace A^Tdiag(y_i)A - A^Tyy^TA$, $A \in \mathbb{R}^{2\times 2}, y \in \mathbb{R}^2_{++}$, $y$ strictly positive, nonzero

(Note: $diag(y_i)$ should more appropriately say $Iy$, where $I$ is the diagonal one matrix)

Define: $$A = \begin{bmatrix} a_{11} & a_{12} \\ a_{21} & a_{22} \end{bmatrix}$$ Explicitly:

$$M = (y_1 + y_2) \begin{bmatrix} a_{11} & a_{21} \\ a_{12} & a_{22} \end{bmatrix}\begin{bmatrix} y_1 & 0 \\ 0 & y_2 \end{bmatrix}\begin{bmatrix} a_{11} & a_{12} \\ a_{21} & a_{22} \end{bmatrix} - \begin{bmatrix} a_{11} & a_{21} \\ a_{12} & a_{22} \end{bmatrix} \begin{bmatrix} y_1 \\ y_2 \end{bmatrix}\begin{bmatrix} y_1 & y_2 \end{bmatrix} \begin{bmatrix} a_{11} & a_{12} \\ a_{21} & a_{22} \end{bmatrix}$$

Is there any method that can prove that the above expression is positive (semi)definite?


Some feeble attempt at chipping the problem:

$M = 1^Ty\thinspace A^Tdiag(y_i)A - A^Tyy^TA = 1^Ty\thinspace A^Tdiag(y_i)A - (y^TA)^Ty^TA$ Seems to be ripe for cauchy schwarz type of argument.

Further more I think the proof would go much easier if it were converted into component (i.e. explicit reference to element of A) form instead of matrix form as it is now.

$\endgroup$
  • $\begingroup$ If $A=0$, then $M=0$ which is not positive definite. $\endgroup$ – copper.hat Oct 27 '15 at 5:40
  • $\begingroup$ @copper.hat Sorry I meant positive (semi) definite $\endgroup$ – Olórin Oct 27 '15 at 5:42
  • $\begingroup$ $y$ is any non - zero vector and $A$ can be any non-zero matrix? $\endgroup$ – thanasissdr Oct 27 '15 at 6:03
  • $\begingroup$ @thanasissdr Yes and their entries are real for simplicity. Also let's make $y_i$ purely positive for simplicity $\endgroup$ – Olórin Oct 27 '15 at 6:05
2
$\begingroup$

Since $1^T y$ is a scalar, it is easy to rewrite your definition of $M$ as $M = A^T Z A$, where $Z = 1^T y \operatorname{diag}\left(y_i\right) - yy^T$. So it remains to prove that $Z$ is nonnegative definite. But $Z = y_1 y_2 \left(\begin{matrix} 1 & -1 \\ -1 & 1 \end{matrix}\right)$.

$\endgroup$
  • $\begingroup$ Thanks also I made $y_i$s purely positive for simplicity. $\endgroup$ – Olórin Oct 27 '15 at 6:09
  • $\begingroup$ Can you clarify what you meant by but? $\endgroup$ – Olórin Oct 27 '15 at 13:52
  • $\begingroup$ What I mean is that this $Z $ is clearly nonnegative semidefinite. $\endgroup$ – darij grinberg Oct 27 '15 at 14:02

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.