3
$\begingroup$

We know about $L_p$ spaces that average of the $p$-th powers of a vector is a norm:

$$ |x|_p = \sqrt[p]{ \frac{x_1^p + \dots + x_n^p}{p}} $$

This means for example, this norm satisfies the triangle inequality (in this case, it's the Minkowski inequality):

$$ |x+y|_p = \sqrt[p]{ \frac{(x_1+y_1)^p + \dots + (x_n+y_n)^p}{p}} \leq \sqrt[p]{ \frac{x_1^p + \dots + x_n^p}{p}} + \sqrt[p]{ \frac{y_1^p + \dots + y_n^p}{p}} = |x|_p + |y|_p $$

I'd like to know if we could use other symmetric polynomials as norms. Let's define a function:

$$ f(x) = \sum_{1 \leq i < j < k \leq n} |x_i x_j x_k|^3 $$

Does this satisfy triangle inequality? E.g. is $f(x+y) < f(x) + f(y)$, where $x_i, y_i > 0$ ? And if it is a norm we typically have constants such that:

$$ A ||\cdot ||_1 < ||\cdot||_2 < B ||\cdot||_1 $$

and I'd like to know possible constants $A$ and $B$, not even optimal.


Originally my question was if $|abc|+|bcd|+|abd|+|acd|$ is a norm on vectors $(a,b,c,d) \in \mathbb{R}^4$ and to compare it to the $L^3$ norm.

Please put absolute value signs as needed. At least, what I've written s hould be true for $p$ even.

$\endgroup$
4
  • $\begingroup$ Did you mean $(x_n + y_n)^p$ in your version of Minkowski instead of $(x_n + n_1)^p$? Also, your expression for the $p$-norm isn't correct. $\endgroup$ – B. Mehta Jul 30 '18 at 20:28
  • 1
    $\begingroup$ It's not the $p$-norm (it is up to a factor $p^{1/p}$), but it is a norm (for $p\geq 1$). $\endgroup$ – Clement C. Jul 30 '18 at 20:31
  • 4
    $\begingroup$ your expressions with $x_ix_jx_k$ $i<j<k$ are 1) not $1$-homogeneous 2) not convex 3) are 0 for the canonical basis so they satisfy precisely 0 conditions out of 3 for being a norm. $\endgroup$ – Surb Jul 30 '18 at 20:42
  • $\begingroup$ What is your question? The question in the title is different from the (rather unclear) question in the body of your post. One very simple criterion for a norm is that it must restrict to a norm on any subspace. Your proposed norms on $\Bbb{R}^4$ do not restrict to a norm on the $x_1$-axis. $\endgroup$ – Rob Arthan Jul 30 '18 at 20:48
5
$\begingroup$

Let $p\geq 1$ and $$ f(x) = \sum_{1 \leq i < j < k \leq n} |x_i x_j x_k|^p $$ Then $f(1,0,\ldots,0)=0$ so it can not be a norm, let us fix that by redefining $f$ as $$ \tilde f(x) = \sum_{1 \leq i \leq j \leq k \leq n} |x_i x_j x_k|^p $$ Now, $\tilde f(\alpha x)= \alpha^{3p} f(x)$ for $\alpha>0$ so it is still not norm, let us fix this by redefining $\tilde f$ as $$ \hat f(x) = \Big(\sum_{1 \leq i \leq j \leq k \leq n} |x_i x_j x_k|^p\Big)^{\frac{1}{3p}}.$$ Now, by computing the Hessian of $\hat f$, you will see that it is not always positive semi-definite. Hence $\hat f$ is not convex and therefore not a norm (and I don't think this can be fixed).

Conclusion: $f$ is not a norm and not even close from being so.

$\endgroup$

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.