MathOverflow is a question and answer site for professional mathematicians. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Various sources claim that a maximum norm $||A||_{max}=\max_{i,j}|a_{ij}|$ is not submultiplicative, i.e. $||AB||_{max}\not\leq||A||_{max}||B||_{max}$.

Where can I find what norm a,b satisfy $||AB||_{max}\leq||A||_{a}||B||_{b}$?

share|cite|improve this question
It's not difficult to show directly that the max norm is not submultiplicative: just let $A$ and $B$ be $2 \times 2$ matices with all entries equal to $1$. – Ian Morris May 28 '10 at 10:25
Well, take a look at . – Wadim Zudilin May 28 '10 at 10:38
up vote 4 down vote accepted

The inequality $\|AB\|_{\max} \leq \|A\|_{a}\|B\|_{b}$ for all $A$, $B$ can be achieved or destroyed just by rescaling the norms $\|\cdot\|_a$ and $\|\cdot\|_b$. Let's suppose that we're considering $d \times d$ matrices. If we just make sure that the two norms $\|\cdot\|_a$ and $\|\cdot\|_b$ are scaled so that both of them have the property $\|C\|_i \geq \sqrt{d}\|C\|_{max}$ for all $d \times d$ matrices $C$, then the desired inequality follows from the elementary inequality $\|AB\|_{\max} \leq d.\|A\|_{\max}\|B\|_{\max}$. Conversely, if the norms are rescaled so that both of them give norm $\frac{1}{2}$ to the identity matrix, then the inequality clearly cannot hold since $\|Id\|_{max}=1$. The fact that such rescalings exist follows from the fact that norms on a finite-dimensional space are pairwise equivalent.

The point of this is that there are a lot of norms on the space of matrices if we don't make any additional requirements on them. Is this the kind of answer you were looking for? Or do you want the two norms to have additional properties?

edited: there was a typo on the main inequality

share|cite|improve this answer
This is what I was looking for, thank you. Furthermore, can you refer me to any book on this? – user6358 May 28 '10 at 13:08
To be honest, I actually can't recommend any books on this! In the absence of any other recommendations, you might try the books referred to in the Wikipedia article on matrix norms linked to by Wadim above. In particular, the book by Horn and Johnson seems to be widely appreciated. – Ian Morris May 28 '10 at 13:26

The Hilbert-Schmidt norm, $||A||_F= (\sum_{j=1}^n a_{i,j}^2)^{1/2}$ is clearly always larger than $||A||_{max}$ and is also submultiplicative.

Hence, $||AB||_{max} \leq ||AB||_F \leq ||A||_F ||B||_F$

share|cite|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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