Let $z_1,\dots,z_n$ be $n$ real numbers with $n \ge z_i \ge 1$ for all $i$. I can show that $$ \left(\sum_{i =1}^n (z_i^2-1)\right)^k \ge \sum_{i =1}^n (z_i^{2k}-1), $$ for all $k \ge 1$ by induction on $k$. But I wonder if there is a way to prove this directly without induction.

Note that if we remove the $-1$ this is just the lp-norm inequality.

NOTE: The inequality is not true as it's stated above. See the accepted answer below. But I do believe it should be true if we further assume $z_i$ are integers.


The inequality does not hold if $n \ge 2$ and $k > 1$. As a counter-example, choose $z_1 = \ldots = z_n = \sqrt x > 1$. Then your inequality becomes $$ n^k(x-1)^k \ge n(x^k - 1) \, . $$ But the difference $$ f(x) = n^k(x-1)^k - n(x^k - 1) $$ satisfies $f(1) = 0$ and $f'(1) < 0$, so that $f(x) < 0$ at least in some interval $(1, 1 + \varepsilon)$.

  • $\begingroup$ But I assumed that $z_i \ge 1$ for all $i$. $\endgroup$ – ablmf Jun 28 '17 at 4:50
  • $\begingroup$ @ablmf: You are right, my fault. – But then your claim is wrong at least for $n=1$. $\endgroup$ – Martin R Jun 28 '17 at 4:56
  • $\begingroup$ Yeah, sorry, I forgot to add $z_i \le n$. That's what I meant. $\endgroup$ – ablmf Jun 28 '17 at 5:32
  • $\begingroup$ @ablmf: It is still wrong, see update. $\endgroup$ – Martin R Jun 28 '17 at 5:58
  • $\begingroup$ you're right, this does not work for real numbers. I implicitly assumed that $z_i$'s are integers. Thanks for pointing out the mistake. $\endgroup$ – ablmf Jun 28 '17 at 6:46

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.