I'm doing the following problem to prepare for my exam tomorrow:

Suppose that $A$ is a $n \times m$ matrix with $n \geq m$ and that the singular values of $A$ are $\sigma_1, \sigma_2, \dots, \sigma_m$. Now consider the augmented matrix

$$ B = \left[ \begin{array}{cc} A \\ I \end{array} \right], $$

where $I$ is the $m \times m$ identity matrix. We want to show that the singular values of $B$ are $\tilde{\sigma}_i = \sqrt{1 + \sigma_i^2}$ for each $i$.

I'm currently lost on where to begin. I know that $A = U\Sigma V^T$ for orthogonal matrices $U$ and $V$ with $\Sigma$ containing the singular values on the diagonal, by definition. However, I'm not seeing how to extend this to show the desired result for $B$.

Can someone lend a hand?


Let $\lambda_1 \geq \dots \geq \lambda_m \geq 0$ be the eigenvalues of the positive semidefinite matrix $A^TA$ so $\sigma_i = \sqrt{\lambda_i}$. Consider

$$ B^T B = \left[ \begin{matrix} A^T & I_m^T \end{matrix} \right] \left[ \begin{matrix} A \\ I_m \end{matrix} \right] = A^TA + I_m. $$

Thus, the eigenvalues of $A^TA + I_m$ are $\lambda_i + 1$ and so the singular values of $B$ are $\sqrt{\lambda_i + 1} = \sqrt{\sigma_i^2 + 1}$.

  • $\begingroup$ Why is $A^TA$ positive semi-definite? Does that hold for any choice of $A$? $\endgroup$ – Jeff Koff Feb 16 '16 at 10:43
  • 1
    $\begingroup$ Yes. The matrix $A^TA$ is symmetric and $ \left< A^TAv, v \right> = (A^T Av)^T v = v^T A^T A v = (Av)^T Av = \left< Av, Av \right> \geq 0$. $\endgroup$ – levap Feb 16 '16 at 10:45

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.