Suppose $A=(a_{ij})$ is an $n\times n$ complex matrix, $\operatorname{tr}(A)=\displaystyle \frac{1}{n}\sum_{i}a_{ii}$. I wonder whether there exists a relationship between $\|A\|$ (the operator norm) and $\operatorname{tr}(A^k)$, where $k=1,\cdots,n$?

  • in general the trace is not divided by $n$, this is a especial trace operator or it is a typo? – Masacroso Nov 8 at 19:09
  • @Masacroso: normalizing the trace so that it becomes a state is a very standard thing. – Martin Argerami Nov 9 at 0:17
up vote 1 down vote accepted

In general no, as nilpotent operators exist. For instance consider $$A=\begin{bmatrix}0&1\\0&0\end{bmatrix}. $$ Then $\|A\|=1$ and $\operatorname{tr}(A^k)=0$ for all $k$.

When $A$ is normal, it is easy to check that $$\tag1 \|A\|=\lim_{k\to\infty}|n\operatorname{tr}(A^k)|^{1/k}. $$ Indeed, you have that $\|A\|=\max\{|\lambda_j|:\ j=1,\ldots,n\}$, where $\{\lambda_j\}$ are the eigenvalues of $A$ counting multiplicities. If we take $\lambda_1$ to be the one with $|\lambda_1|=\|A\|$,
\begin{align} |\operatorname{tr}(A^k)|^{1/k}&=\exp\left({\tfrac1k\,\log|\sum_{j=1}^n\lambda_j^k|}\right) =\exp\left({\tfrac1k\,\log|\lambda_1^k|\,|1+\sum_{j=1}^n(\lambda_j/\lambda_1)^k|}\right)\\ \ \\ &=|\lambda_1|\,\exp\left({\tfrac1k\,\log |1+\sum_{j=1}^n(\lambda_j/\lambda_1)^k|}\right)\xrightarrow[k\to\infty]{}|\lambda_1|,\\ \ \\ \end{align} as the expression inside brackets is bounded above by $\tfrac1k\,\log(1+n)$.

  • why the second equality has no the term $1/n$? – mathrookie Nov 9 at 15:09
  • I forgot an $n$ in $(1)$. – Martin Argerami Nov 9 at 15:13
  • So in your proof,$tr$ denotes the sum of diagonal elements of a matrix? – mathrookie Nov 9 at 15:20
  • No. $ \ \ \ \ $ – Martin Argerami Nov 9 at 15:21
  • $|tr(A^k)|=1/n|\sum \lamda j k$? – mathrookie Nov 9 at 15:26

Your Answer

 

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Not the answer you're looking for? Browse other questions tagged or ask your own question.