How to show that the relationship $\left | \lambda_{min} \right | \geq \sigma_{min}$ holds between the minimum eigenvalue and singular value of a square matrix $A \in \mathbb{C}^{n \times n}$?

  • 2
    $\begingroup$ I think it is worth mentioning that you implicitly order the eigenvalues by their modulus. $\endgroup$ – Julien May 4 '13 at 3:18

On one hand, if $v$ is a unit eigenvector of $A$ corresponding to the eigenvalue $\lambda_\min$, then $v^\ast A^\ast Av=\|Av\|^2=|\lambda_\min|^2$. On the other hand, $A^\ast A$ is positive semidefinite. Hence it can be unitarily diagonalised as $U^\ast DU$, where $D$ is a nonnegative diagonal matrix. The minimum diagonal entry of $D$, i.e. the minimum eigenvalue of $A^\ast A$, is equal to $\sigma_\min^2$. Therefore $$ |\lambda_\min|^2 =v^\ast A^\ast Av \ge\min_{\|x\|=1}x^\ast A^\ast Ax =\min_{\|x\|=1}(Ux)^\ast D (Ux) =\min_{\|y\|=1}y^\ast D y =\sigma_\min^2. $$

  • 1
    $\begingroup$ I like how you always give a full and concise argument, +1. $\endgroup$ – Julien May 4 '13 at 3:22
  • $\begingroup$ +1, this argument is nice! and should have been accepted as an answer rather than the other bookish one :). $\endgroup$ – dineshdileep May 4 '13 at 5:55

This is related to the so called weyl-horn theorem on the relation between eigenvalues and singular values of an arbitrary matrix $A$. Let $\lambda_1,\lambda_2,\dots,\lambda_n$ be the eigenvalues, such that $|\lambda_1|>|\lambda_2|>\dots>|\lambda_n|$ (your $\lambda_{min}=\lambda_n$) and $\sigma_1,\sigma_2,...,\sigma_n$ (your $\sigma_{min}=\sigma_n$) be the singular values. Then \begin{align} &\prod_{j=1}^{k}|\lambda_j|\leq \prod_{j=1}^{k}\sigma_j~~,k=1,2,...,n-1 \\ &\prod_{j=1}^{n}|\lambda_j|= \prod_{j=1}^{n}\sigma_j \\ \end{align} Let $a=\prod_{j=1}^{n-1}|\lambda_j|$ and $b=\prod_{j=1}^{n-1}\sigma_j$, then we have $a \leq b$ and $a|\lambda_{min}|=b\sigma_{min}$ , thus $|\lambda_{min}|\geq \sigma_{min}$

  • $\begingroup$ These are called Weyl's inequalities. What Horn did is that he proved that such sequences always occur as eigenvalues and singular values of some complex $n\times n$ matrix. Note: your argument requires to treat the null case separately. This being said, +1. $\endgroup$ – Julien May 4 '13 at 3:13

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.