Theorem: Suppose $U,V$ are orthogonal subspaces spans $X$, let $A:U\to U,B:V\to V$. then the operator $C=\begin{pmatrix}A&0\\0&B\end{pmatrix}:X\to X$

and $\|C\|=\max\{\|A\|,\|B\|\}$

using above theorem I am told to show the following

$A$ be any linear operator on finite dimensional vector spaces then $\|A\|=\max\{|s_1,\dots,\|s_k\|\}$ where $s_i$'s are singular values of $A$

from svd of $A$ we get $A=U\Sigma V^T \Rightarrow \|A\|=\|U\Sigma V^T\|\le\|U\|\|\Sigma\|\|V^T\|=\|\Sigma\|$ since unitary matrices has rank $1$...please help how to proceed


In a finite dimensional space $A$ has an adjoint $A^*$ and $||A^*|| = ||A||$

Let $B = AA^*$ then $||B|| = ||A A^*|| = ||A||^2$

$B$ is self adjoint and non-negative therefore has a spectrum of real non-negative eigenvalues {$\lambda_i$}, corresponding orthogonal eigenspaces {$E_i$}, and projections {$P_i$} onto those eigenspaces. Furthermore $B = \Sigma \lambda_i P_i$.

Then by your given initial result, $||B|| = max(|| \lambda_i P_i||) = max(|\lambda_i|.|| P_i||) $. But for each projection, $||P_i|| = 1$.

So, $||B|| = max(|\lambda_i| )$ and $||A|| = ||B||^{1/2} = (max(|\lambda_i|) ^{1/2} = max(|\lambda_i|^{1/2}) $.

Since the eigenvalues of $B$ are non-negative then $||A|| = max(\lambda_i^{1/2})$ which by definition is the maximum of the singular values of $A$


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.