Let $a_1,...,a_n$ be positive real numbers and let $0 < p < 1$. Then $$(a_1 + \cdots + a_n)^p \leq a_1^p + \cdots + a_n^p. $$ Now take $ 1 \leq p < \infty$. Can we get a similar inequality, like $$ (a_1 + \cdots + a_n)^p \leq C_p(a_1^p + \cdots + a_n^p),$$ where $C_p > 0$ is a constant only depending on $p$? If not, what is the better approach we can get?

  • 1
    $\begingroup$ Take $a_1=\cdots=a_n=1$… $\endgroup$ – Saad Jun 9 '18 at 9:31
  • 2
    $\begingroup$ @AlexFrancisco $C_p=n^{p-1}$ works and is tight. Proof could be by using convexity or mean inequalities like Power Means. $\endgroup$ – Macavity Jun 9 '18 at 9:32
  • $\begingroup$ @Macavity $C_p$ should be independent from $n$. $\endgroup$ – Saad Jun 9 '18 at 9:33
  • 1
    $\begingroup$ @AlexFrancisco If that’s intended there is no $C_p$. $\endgroup$ – Macavity Jun 9 '18 at 9:34
  • $\begingroup$ Once you have the educated guess that $C_p=n^{p-1}$, Jensen's Inequality fairly rapidly proves that this is indeed the case by convexity of $x^p$. $\endgroup$ – Robert Wolfe Jun 9 '18 at 14:20

Let $f(x)=x^p$, $0<p<1$ and $a_1\geq a_2\geq...\geq a_n$.

Thus, $f$ is a concave function and $(a_1+a_2+...+a_n,0,...,0)\succ(a_1,a_2,...,a_n)$.

Thus, by Karamata we obtain: $$f(a_1+a_2+...+a_n,0,...,0)\leq f(a_1)+f(a_2)+...f(a_n)$$ or $$(a_1+a_2+...+a_n)^p\leq a_1^p+a_2^p+...+a_n^p$$ and we are done!

Also, we see that we got a best estimation: $C_p=1$.

For $p\geq1$ and $a_1=a_2=...=a_n=1$ we get $C_p=n^{p-1}.$

We'll prove that $$(a_1+a_2+...+a_n)^p\leq n^{p-1}(a_1^p+a_2^p+...+a_n^p)$$ or $$\left(\frac{a_1+a_2+...+a_n}{n}\right)^p\leq\frac{a_1^p+a_2^p+...+a_n^p}{n},$$ which is Jensen for $f(x)=x^p$.

  • 1
    $\begingroup$ If you take $a_1 = \cdots = a_n = 1$ you would have $ n^p \leq n$, and this is not true for $p > 1$. $\endgroup$ – Javier González Jun 9 '18 at 12:58
  • $\begingroup$ Yes, I fixed my post. Thank you! I thought that we need to prove that $(a_1+...+a_n)^p\leq a_1^p+...+a_n^p$ for $0<p<1$. $\endgroup$ – Michael Rozenberg Jun 9 '18 at 14:09

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.