Suppose that $u\in\mathbb{R}^{n\times 1}$ is such that $\sum_{i=1}^n u_i=0$.

Prove that $|u^\top v|\leq||u||_ 1(\frac{v_{max}-v_{min}}2)$, where $v_{max}=\max |v_i|$ and $v_{min}=\min|v_i|$ for any $v \in \mathbb{R}^{n\times 1}$.

I tried to prove this similarly to how my professor did it, using the fact that $|u^\top(v-\alpha e)|=|u^\top v|$ if $e=(1,\cdots,1)$ and $\alpha$ is any real number because the sum of the components of $u$ is $0$. Then I used Hölder's inequality to get $|u^\top v|\leq||u||_1||v-\alpha e||_\infty$, and this is where I got stuck.

My professor does not justify why $$\inf ||v-\alpha e||_\infty =\frac{v_{max}-v_{min}}2.$$ I guess that the average of the maximum and minimum values of $v$ could minimize $||\cdot||_\infty$, but I'm not really convinced without a proper explanation and I can't prove it myself.

I guess what we really need to find is $$\lim_{p\to\infty} \inf_\alpha (\sum_{i=1}^n |v_i - \alpha|^p)^\frac1p $$ (because the limit of the p-norms is $||\cdot||_\infty$), which does not sound so easy and straightforward, though maybe it actually is and I'm not seeing how.


Note that $$ v_{\min}-\alpha \leq v_i-\alpha \leq v_\max-\alpha, $$and thus $$ \max_i |v_i -\alpha|= \max\{|v_\min -\alpha|, |v_\max - \alpha|\}. $$ The right-hand side is the maximum of $d(v_\min, \alpha)$ and $d(v_\max, \alpha)$. One can see that the minimizer of it is $$\alpha =\frac{v_\min + v_\max}{2}$$ and $$ \inf_{\alpha\in \mathbb{R}} \|v - \alpha e\|_\infty = \frac{v_\max - v_\min}{2} $$ follows.

  • $\begingroup$ Oh, I get it now. We want to pick an $\alpha$ that makes both terms we want to find the maximum of be the same will minimize the maximum. Thanks for the explanation! $\endgroup$ – AstlyDichrar Jan 2 at 22:19

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.