0

I have to proove that $(K_{p,p}⊗ I_p)(\vec{a}⊗\vec{a}⊗\vec{a})$ equals $(\vec{a}⊗\vec{a}⊗\vec{a})$, where $\vec{a}$ is px1 vector, $I_p$ is unit matrix pxp and $K_{p,p}$ is commutation matrix ppxpp. I have some ideas, but I'm not sure, do I think right. I tried to proove it from right to left. Starting from $(\vec{a}⊗\vec{a}⊗\vec{a})$ = $((\vec{a}⊗\vec{a})⊗\vec{a})$

I used to fact, that $(\vec{a}⊗\vec{a})= K_{p,p} \cdot(\vec{a}⊗\vec{a}) \cdot K_{1,1}$. And I know, that commutation matrix is orthogonal, so $K_{p,p}'\cdot K_{p,p}=I_{p,p}$. So I have $K_{p,p}\cdot I_{p,p}\cdot (\vec{a}⊗\vec{a})$. But are $K_{p,p}\cdot I_{p,p}$ and $K_{p,p}⊗I_{p}$ equal?

  • Commutation matrix is block matrix K_{m,n} : mn x mn that has n x m blocks when (K_{m,n})_{(i,j)(g,h)} is 1 when g=j and i=h, 0 otherwise, where i,h=1,...,m and j,g=1,...,n – Vida Beach Apr 16 '23 at 08:35
  • Thanks. Is it possible to write this commutation matrix as a Kronecker product ? See interesting things there. – Jean Marie Apr 16 '23 at 08:39
  • As seen in the answer I have indicated you, the easiest way to work with a commutation matrix is to consider it as the operator sending $vec(A)$ onto $vec(A^T)$. 2) Besides (your last line) $K_{p,p}\cdot I_{p,p}$ and $K_{p,p}⊗I_{p}$ cannot be equal because they haven't the same dimensions.
  • – Jean Marie Apr 16 '23 at 12:25
  • ... unless $I_{p,p}$ mean $I_{(p^2)}$. – Jean Marie Apr 17 '23 at 16:26
  • @Ben Grossmann Thank you. Indeed, it is a good reference. – Jean Marie Apr 17 '23 at 18:13