Let $x_1, \dots, x_n$ be positive real numbers. Arithmetic-geometric mean inequality tells us that:

$GM = \sqrt[n]{x_1 \dots x_n} \leq \frac{x_1 + \dots + x_n}{n} = AM$

and that equality occurs iff $x_1 = \dots = x_n$. This condition can be restated as $\sum_{k=1}^n (x_k - AM)^2 = 0$, i.e. sample variance of $x_1, \dots , x_n$ is zero. I'm curious: are there any inequalities connecting arithmetic mean, geometric mean and sample variance?

This is my point: Difference between $AM$ and $GM$ gets larger as $x_1, \dots, x_n$ get further away from each other, i.e. when their sample variance is big. Is there a way to account for the error in arithmetic-geometric mean inequality using sample variance?

  • $\begingroup$ What you ask for might be related to the concept of majorization. Hardy, Littlewood and Polya had already developed the theory to a very advanced level. If you are willing to pursue the study, you should probably check the book "Inequalities: Theory of Majorization and Its Applications". $\endgroup$ – Troy Woo Aug 18 '14 at 10:33
  • $\begingroup$ Maybe these can help: arxiv.org/pdf/1210.4417.pdf and arxiv.org/abs/1203.4454 and references therein. $\endgroup$ – pisoir Aug 18 '14 at 10:41

Power mean inequality can be used to get bounds the difference between various means. Not exactly using variances, but you have $$AM - GM \le \max_i x_i - \min_i x_i$$ Or if that's too large a bound, note $$AM - GM \le AM - HM$$

If you want variances involved, try manipulating $$\sqrt{\frac1n \sum x_i^2} - GM \ge AM-GM$$


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.