1
$\begingroup$

The Centering Matrix $C:=1_N-\iota \iota'$ has eigenvalue $1$ of multiplicity $n − 1$ and eigenvalue $0$ of multiplicity $1$. Therefore a matrix $S$ with columns consisting of eigenvectors of $C$ can be chosen such that $$CS=S\left(\begin{array}{cc} 1_{N-1} & 0 \\ 0 & 0 \end{array} \right).$$

I understand that for the first $N-1$ eigenvectors $s$ it must hold that $$Cs=s-\frac{1}{N}\iota\iota's=s \Leftrightarrow \iota's=0$$ and for the last eigenvector $s_N$ it must hold that $$Cs_N=s_N-\frac{1}{N}\iota\iota's_N=0 \Leftrightarrow s_N=\alpha\iota, \alpha\in\mathbb{R}$$

Therefore to obtain $S'S=1_N$ it must hold that $s_N=\frac{1}{\sqrt{N}}\iota$. However, how can I choose $S=(s_1,\ldots,s_{N-1})$ such that $S'S=1_N$?

$\endgroup$
0
$\begingroup$

Found one possible characterization of $S$. As already mentioned $s_N=\frac{1}{\sqrt{N}}\iota$ to fullfill $S'S$ (Otherwise the last entry of $S'S$ is not equal to $1$.) If have chosen the following vectors $s_i$: $$s_1=\left(\begin{array}{c} \frac{1}{k_i}\\-\frac{1}{k_i} \\ 0\end{array} \right),s_2=\left(\begin{array}{c} \frac{1}{k_2}\\\frac{1}{k_2} \\ -\frac{2}{k_2}\\0\end{array} \right),\ldots,s_i=\left(\begin{array}{c} \frac{1}{k_i}\\\vdots \\ -\frac{i}{k_i}\end{array} \right).$$ These $s_i$ fullfill $\iota's_i$ and by construction $s_j's_k=0$. Therefore the $s_i$ form a basis of the subspace of Dimension $N-1$ spanned by range$(C)$. For $S'S=1_N$ to hold we need $s_i's_i=1$ therefore we have to choose $$k_i=\sqrt{i+i^2}$$.

$\endgroup$

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.