3

Given two matrices $A, B \in \mathbb{R}^{n \times m}$, where $A = \{a_{i,j}\}$ and $B = \{b_{i,j}\}$, the Hadamard product (or point-wise product) is a matrix $$C = A \circ B$$ such that $C = \{c_{i,j}\} = \{a_{i,j} \cdot b_{i,j}\}$.

I'm not so skilled with this operator. I googled around and I found few things, just definitions and some relevant properties.

At the moment, I'm working on this kind of problem:

Consider $A, B \in \mathbb{R}^{N \times N}$ and $C \in \mathbb{R}^{N \times N}_{\geq 0}$. Find $x \in [0,1]^N$ such that $$M x = Qy,$$

where $M = (A+B)\circ C$, $Q=B \circ C$ and $y = [1 ~ 1 \ldots 1]^\top \in \mathbb{R}^{N}$.

Of course, I'm looking for the condition on matrices $A$, $B$ and $C$ such that one can solve the previous problem.

It can be also useful to solve a relaxed version of the problem:

Consider $A, B \in \mathbb{R}^{N \times N}$ and $C \in \mathbb{R}^{N \times N}_{\geq 0}$. Find $x \in \mathbb{R}^N$ such that $$M x = Qy,$$

where $M = (A+B)\circ C$, $Q=B \circ C$ and $y = [1 ~ 1 \ldots 1]^\top \in \mathbb{R}^{N}$.

I was able to find some condition for the existence of a solution $x$ when $A$ and $B$ have constant rows. But I would like to find general conditions.

The problem is too broad, and therefore I would like just some hints for the case $C \in \mathbb{R}^{N \times N}_{\geq 0}$, i.e. $c_{i,j} \geq 0 ~\forall i, j$.

the_candyman
  • 14,234
  • 4
  • 37
  • 65
  • 1
    In essence, you want to compute $M^{-1}$ and its action on $Qy$. Finding such an inverse is discussed in this earlier question: http://math.stackexchange.com/q/1321879/137524. (Daniel's answer is especially helpful, I think). – Semiclassical Jun 09 '16 at 19:51
  • 1
    One sufficient condition that may be of interest to you is that it's enough to have $(A + B)$ and $C$ be (symmetric and) positive definite (in this case, we can guarantee a unique solution). – Ben Grossmann Jun 09 '16 at 22:27
  • Are you looking for a better condition, than for all $i=1,\dots,N$ it holds that $$ \sum_{j=1}^N\left(a_{i,j}+b_{i,j}\right)c_{i,j}x_j=\sum_{j=1}^N b_{i,j}c_{i,j} ,,? $$ – user153012 Nov 23 '16 at 16:10
  • @user153012 No. I'm asking some conditions on $A$, $B$ and $C$ such that there is a solution $x$ and that $x \in [0, 1]^N$. – the_candyman Nov 23 '16 at 17:19
  • @the_candyman For example, if $A=0$, $B$ and $C$ are arbitrary, then $x=y$ is a solution. – user153012 Nov 23 '16 at 17:32
  • @user153012 I don't think this is enough. The case you are depicting is very trivial. – the_candyman Nov 23 '16 at 17:43
  • 1
    @the_candyman I know. Although, I don't think that we can expect a much better condition than the definition property itself. It is because the Hadamard product is not as powerful tool as the ordinary matrix multiplication. A standard reference for the Hadamard product is the 5th Chapter of Topics in Matrix Analysis by Roger A. Horn and Charles R. Johnson. – user153012 Nov 23 '16 at 17:57
  • @user153012 Thanks a lot! I appreciate your efforts and the reference. – the_candyman Nov 23 '16 at 18:05

0 Answers0