I just found this related question in here Q1.

Given a positive definite matrix $\mathbf{A}$, consider its eigendecomposition $(\mathbf{A}\mathbf{V} = \mathbf{V}\mathbf{D})$. Consider an arbitrary matrix $\mathbf{B}$,

Is there any closed-form expression of the eigendecomposition of $\mathbf{A} \circ \mathbf{B}$?

$\circ$ denotes the Hadamard product.

Edition 29/05/2015

Since considering an arbitrary matrix $\mathbf{B}$ does not lead to a closed-form expression, let us consider that all elements of $\mathbf{B}$ can be defined such as $[\mathbf{B}]_{m,n}= e^{i\theta_{m,n}}$ where $\theta_{m,n} \in [0, 2\pi]$ for $m = 1, \ldots , M$ $n = 1, \ldots , N$. Of course, $\mathbf{A}$ has the same dimensions of $\mathbf{B}$.

Thank you in advance

  • $\begingroup$ What data are allowed to appear in your "closed-form expression"? Of course we need to know the eigendecomposition of $B$... $\endgroup$ – Yemon Choi May 28 '15 at 12:36
  • 2
    $\begingroup$ the obvious answer is: no there won't be a nontrivial closed form. $\endgroup$ – Suvrit May 28 '15 at 13:00

There won't be a nontrivial closed form in general. A trivial "closed-form" is mentioned below.

The Hadamard product $A \circ B$ is a principal submatrix of the Kronecker product $A\otimes B$. Thus, let $A=U\Lambda U^{-1}$ and $B=VDV^{-1}$ (I assumed $B$ is diagonalizable and not fully arbitrary as required in the question). Then, we have \begin{equation*} A \circ B = E^*(U\Lambda U^{-1} \otimes V D V^{-1})E = E^*[(U\otimes V)(\Lambda\otimes D)(U\otimes V)^{-1}]E. \end{equation*} Thus $A\circ B$ is a compression of the matrix on the right hand side. This is about as far as we get.

  • 1
    $\begingroup$ OK, then I assume no further insights can be obtained. $\endgroup$ – mikitov May 28 '15 at 13:32
  • 2
    $\begingroup$ @mikitov: if $B$ is special, then surely one can say more; e.g., if $B=11^T$ :-) One can say more precise things about rougher quantities such as $\|A\circ B\|$ etc., or singular value inequalities, and so on; but a fine-grained description of the spectrum seems too hard. $\endgroup$ – Suvrit May 28 '15 at 15:58

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.