0
$\begingroup$

Let $\mathbf{A},\mathbf{B}\in\mathbb{R}^{n\times n}$ be two positive semidefinite matrices. Also let $\mathbf{A}\circ \mathbf{B}$ denote the Hadamard product of $\mathbf{A}$ and $\mathbf{B}$. A classical result states that \begin{align*} \lambda_{\min}(\mathbf{A}\circ \mathbf{B})\ge \lambda_{\min}(\mathbf{A})\lambda_{\min}(\mathbf{B}) \end{align*} I am wondering whether a "subspace-restricted" version of this result is true. More specifically let $\mathbf{U}\in\mathbb{R}^{n\times k}$ be an orthonormal matrix obeying $\mathbf{U}^T\mathbf{U}=\mathbf{I}_k$ with $k\le n$. I am wondering if the following inequality holds

\begin{align*} \lambda_{\min}(\mathbf{U}^T(\mathbf{A}\circ \mathbf{B})\mathbf{U})\ge \lambda_{\min}(\mathbf{U}^T\mathbf{A}\mathbf{U})\lambda_{\min}(\mathbf{U}^T\mathbf{B}\mathbf{U}) \end{align*}

$\endgroup$
1
$\begingroup$

As stated the assertion is false e.g. let first column of A equal to 0 and second column of B equal to 0. take $\mathbf{u}=\frac{(\mathbf{e}_1+\mathbf{e}_2)}{\sqrt{2}}$

$\endgroup$

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.