Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

I'd like some help showing the equivalence of these two norms when $p = \log n$.

Recall the $p$-th Schatten norm of a linear operator $A$ acting on $\mathbb{R}^{n}$. In the particular case of $p = \log(n)$, Schatten norm should be equivalent to the spectral norm, see last line on p.18. Moreover, $\| A\|_{C^{n}_{\log n}} \leq \| A \|$.

On the other hand, consider an example of an operator on $\mathbb{R}^{3}$:

$$ A = \begin{pmatrix} 1 & 0 & 0\\ 0 & 2 & 0 \\ 0 & 0 & 3 \end{pmatrix} $$

Obviously, $\| A \| = 3$. However when $p = \log(3)$, the corresponding Schatten's norm is

$$ \| A \|_{C^{3}_{\log 3}} = \left( 1^{\log 3} + 2^{\log 3} + 3^{\log 3}\right)^{\frac{1}{\log 3}} \approx 5.48 $$

implying that $\| A\|_{C^{n}_{\log n}} > \| A \|$.

Am I doing something wrong without realizing it? I'm confused here. I'd like to show the equivalence of two norms but I'm stuck...

share|improve this question
add comment

1 Answer

up vote 1 down vote accepted

In finite dimensional vector space all norms are equivalent, and you can use a general theorem to prove your assertion, otherwise $$ \lVert A \rVert_{C^n_p} := \left(\sum_1^n s_i(A)^p\right)^{1/p} \leq \\ \left(n \cdot\sup_i\{s_i(A)^p\}\right)^{1/p} = n^{1/p} \cdot \sup_i\{s_i(A)\} = n^{1/p} \cdot \lVert A \rVert $$ and $$ \lVert A \rVert := \sup_i \{s_i(A)\} = \left(\sup_i\{s_i(A)^p\}\right)^{1/p} \leq \left(\sum_1^n s_i(A)^p\right)^{1/p} = \lVert A \rVert_{C^n_p} $$

share|improve this answer
    
Thanks for the answer, it cleared some things up. However, could you confirm that the inequality stated in the paper I refered to, i.e. $\| A\|_{C^{n}_{\log n}} \leq \| A \| \leq e\|A\|_{C^{n}_{\log n}}$ is actually false? It certainly seems so given my "counter-example". Adapting it from the proof, the correct one should look like $$\|A\|_{C^{n}_{\log n}} \leq e\|A\| \leq n\|A\|_{C^{n}_{\log n}}$$ which is obtained setting $p = \log n$. –  johnny Oct 8 '12 at 7:56
    
To add to my comment, I now guess that the inequalities can be sharpened. It seems that in all cases $\|A\| \leq \| A \|_{C^{n}_{p}}$ for $p \geq 1$ so that we have $\| A \|_{C^{n}_{p}} \leq e\|A\| \leq e\| A \|_{C^{n}_{p}}$. I guess that the paper missed an $e$ somewhere. –  johnny Oct 8 '12 at 8:23
    
@johnny I confirm that the inequality stated in the paper is false and, as you suggested, I sharpened the second inequality in my answer. –  AlbertH Oct 8 '12 at 14:51
    
Thanks. Your help is greatly appreciated. –  johnny Oct 8 '12 at 15:11
    
The article "Random Vectors in the Isotropic Position" by M. Rudelson also has the inequality that has the same 'wrong' form as that you mentioned. –  chl Nov 20 '13 at 14:10
add comment

Your Answer

 
discard

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.