2
$\begingroup$

Let $x_{1},\ldots,x_{n}$ be nonnegative numbers such that $m \leq x_{i} \leq M$. Let $$ S=\sum_{i=1}^{n}{x_{i}} $$ and $$ Q=\sum_{i=1}^{n}{x_{i}^{2}}. $$

Then $$ Q \leq S(M+m)-nMm. $$

This has been recently discovered by at least two different people (including myself) but I am sure that it has come up many times before. Does it have a standard name or reference?

$\endgroup$
5
  • 2
    $\begingroup$ I guess this is equivalent to the inequality $x^2 \leq x(m+M) - Mm,$ since it must be true term by term ($n=1$ case seems to imply the claim in general). So that inequality might have a name if such a thing exists. The fact that $M \geq x$ then immediately implies the claim, so it might not, because it could be seen as being too easy, since it doesn't seem like there's any trick. $\endgroup$ Sep 17, 2014 at 8:13
  • $\begingroup$ @J.E.Pascoe That's a nice proof - thanks! Still looking for the provenance, though... $\endgroup$ Sep 17, 2014 at 8:23
  • 2
    $\begingroup$ If $A$ is an $n \times n$ matrix, this inequality describes a relationship between ${\rm tr}\ A$ and ${\rm tr}\ A^2$. $\endgroup$ Sep 17, 2014 at 13:58
  • $\begingroup$ @DavidHandelman Can you please elaborate a bit? 10x! $\endgroup$ Sep 17, 2014 at 15:00
  • 2
    $\begingroup$ If $A$ has eigenvalues (counting multiple ones, including what are sometimes called algebraic ones, really bad terminology) $(x_i)$, then ${\rm tr}\ A = \sum x_i$ and ${\rm tr}\ A^2 = \sum x_i^2$. $\endgroup$ Sep 17, 2014 at 21:04

2 Answers 2

5
$\begingroup$

As @J.E.Pascoe notes the question reduces to the case $n=1$. The fact that $m\leq x$ and $x\leq M$ imply that $x^2+Mm \leq mx + xM$ is a special case of the rearrangement inequality.

$\endgroup$
1
  • $\begingroup$ Irinically, my original overwrought proof used the Gruss inequality - which is a complement to Chebyshev. So it all comes together. :) $\endgroup$ Sep 17, 2014 at 8:48
2
$\begingroup$

I first came across this inequality (n=1 case) from the short note http://www.ams.org/journals/proc/1965-016-05/S0002-9939-1965-0180561-X/S0002-9939-1965-0180561-X.pdf

which says that if $m\le A\le M$, then $(M-A)(A-m)$ is again positive semidefinite. This inequality has intimate connection with the Kantrovich inequality.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.