In one of the research paper I read that $$\mathbb{a}^T(\mathbb{b}\odot \mathbb{c})=\mathbb{b}^T \text{diag}(\mathbb{a})\mathbb{c}$$ where $\mathbb{a},
\mathbb{b},\mathbb{c}$ are vectors of same size. What is the name of above property and what does $\mathbb{b}\odot \mathbb{c}$ means? Thanks in advance. Here is the snap shot of the paper I am reading
-
2$\begingroup$ Care to share the link to the research paper you are reading? $\endgroup$– Jose Arnaldo Bebita DrisJun 5, 2018 at 2:50
-
1$\begingroup$ And if it’s behind a paywall just attach a screenshot of the context. $\endgroup$– rschwiebJun 5, 2018 at 2:56
-
1$\begingroup$ Yes, I agree with @rschwieb. The article as it is is behind a paywall. $\endgroup$– Jose Arnaldo Bebita DrisJun 5, 2018 at 2:57
-
1$\begingroup$ @FrankMoses it’s available for free at researchgate, for those of us who made accounts. $\endgroup$– rschwiebJun 5, 2018 at 3:05
-
1$\begingroup$ @FrankMoses I believe you now. Looks like it could be domain specific. I don’t think I’ve ever seen it. $\endgroup$– rschwiebJun 5, 2018 at 3:09
1 Answer
The notation ${\rm diag}({\bf a})$ is defined in the paper you linked, page 4027, column 1, last line. I have never used the notation $\odot$ but suspect that it is the elementwise product of two vectors, in which case for ${\bf a}=(a_1,a_2,\ldots,a_n)$ etc we have $$\eqalign{ {\bf a}^T({\bf b}\odot{\bf c}) &=\pmatrix{a_1&\cdots&a_n\cr}\pmatrix{b_1c_1\cr\vdots\cr b_nc_n\cr}\cr &=a_1b_1c_1+\cdots+a_nb_nc_n\cr &=\pmatrix{b_1&\cdots&b_n\cr}\pmatrix{a_1c_1\cr\vdots\cr a_nc_n\cr}\cr &=\pmatrix{b_1&\cdots&b_n\cr}\pmatrix{a_1&\cdots&0\cr\vdots&\ddots&\vdots\cr0&\cdots&a_n\cr}\pmatrix{c_1\cr\vdots\cr c_n\cr}\cr &={\bf b}^T{\rm diag}({\bf a}){\bf c}\ .\cr}$$
-
$\begingroup$ Thank you so much for your answer. I also think this is the case but they should have mentioned it in the paper $\endgroup$ Jun 5, 2018 at 3:19
-
$\begingroup$ Good detective work. I've seen a couple things denoting the Hadamard/Schur/elementwise product, but not this. Actually, it did turn up on stack exchange in a solution once before $\endgroup$– rschwiebJun 5, 2018 at 10:34