Calculating the SVD of the following matrix (by hand)

421 Views Asked by At

I'm doing an exercise from an old exam that asks for calculating the Singular Value Decomposition (SVD) of the following matrix, for any $p\in\mathbb{R}$:

$$A_p=\begin{bmatrix} \ \ 1+p &\ \ 1-p & -1+p & -1-p\\ \ \ 1-p&\ \ 1+p&-1-p&-1+p\\ -1+p&-1-p&\ \ 1+p&\ \ 1-p\\ -1-p&-1+p&\ \ 1-p&\ \ 1+p \end{bmatrix}$$

Now I can see that $A_p={A_p}^T$ and thus by the spectral theorem there exists an $X\in O_4(\mathbb{R})$ such that $X^{-1}A_pX=\Lambda:=\mathrm{diag}(\lambda_1,\lambda_2,\lambda_3,\lambda_4)$, and thus the singular values equal the absolute values of the eigenvalues $\lambda_1,\lambda_2,\lambda_3,\lambda_4$. Also, it is easy to see that with $r_i$ for the $i$'th row of $A_p$, $r_1=-r_4$ and $r_2=-r_3$ and thus $\dim\ker(A_p)=2$, and thus $A_p$ has eigenvalue $0$ with multiplicity $2$ for any $p\in\mathbb{R}$.

Now from here everything I tried and expanded didn't lead me anywhere. Mathematica confirms my work this far and gives the remaining two eigenvalues which are $4$ and $4p$. However, I want to know how to calculate these eigenvalues myself. I'm quite sure there is some trick that I am not seeing since calculating $\det(A_p-\lambda I)$ is tedious and this old exam question does not yield that much points.

Just a note: A follow up question also asks for the best $\mathrm{rank}\ 1$ approximation so I would also need the eigenvectors corresponding to the largest eigenvalue, which can be either $4$ or $4p$.

Any help is much appreciated.

1

There are 1 best solutions below

1
On BEST ANSWER

I just found out that $$A_p=4\begin{bmatrix}\ \ \frac12\\ \ \ \frac12\\-\frac12\\-\frac12 \end{bmatrix}\begin{bmatrix}\frac12& \frac12&-\frac12&-\frac12 \end{bmatrix}+4p\begin{bmatrix}\ \ \frac12\\-\frac12\\\ \ \frac12\\-\frac12 \end{bmatrix}\begin{bmatrix}\frac12& -\frac12&\frac12&-\frac12 \end{bmatrix},$$ so this fully solves my question and directly yields the eigenvectors I was looking for.