2
$\begingroup$

I'm working on some inequality proofs involving cyclic and/or symmetric sums involving three positive variables $(a,b,c)$, and I found myself trying to use a plausible combination of Muirhead and AM-GM. But I have no proof that that combination is valid.

I will present the question for three variables, but the generalization to more than three variables is obvious.

Some notation: $\sigma$ will be used to represent all the permutations of $(a,b,c)$ so that $$ \sum_\sigma a^p b^q c^r \equiv a^p b^q c^r + a^p c^q b^r + b^p c^q a^r +b^p a^q c^r + c^p a^q b^r + c^p b^q a^r $$ and we always take $p\geq q\geq r$, which we are free to do because of the symmetry of the sum.

Muirhead's inequality says (remember that $a,b,c$ are all positive)

If $[p_1 \geq p_2] \,\, \land \,\,[(p_1+q_1)\geq (p_2+q_2])\,\, \land \,\,[(p_1+q_1+r_1)\geq (p_2+q_2+r_2)]$ then $$ \sum_\sigma a^{p_1} b^{q_1} c^{r_1} \geq \sum_\sigma a^{p_2} b^{q_2} c^{r_2} $$ I will word that if condition as "If $(p_1,q_1,r_1)$ majorizes $(p_2,q_2,r_2)$

In the applications I have, there are combinations of $\sum_\sigma a^p b^q c^r$ with various combinations of $p,q,r$ that always add up to the same number.

My conjecture is:

If $(p_1+q_1+r_1)=(p_2+q_2+r_2)=(p_3+q_3+r_3)$ and $(p_{12},q_{12},r_{12})$ majorizes $(p_3,q_3,r_3)$ where $p_{12}$ is the mean of $p_1$ and $p_2$ and similarly for $q_{12}, r_{12}$, then $$ \frac{\sum_\sigma a^{p_1} b^{q_1} c^{r_1}+\sum_\sigma a^{p_2} b^{q_2} c^{r_2}}{2}\geq \sum_\sigma a^{p_3} b^{q_3} c^{r_3} $$

Note that by AM-GM $$ \frac{\sum_\sigma a^{p_1} b^{q_1} c^{r_1}+\sum_\sigma a^{p_1} b^{q_1} c^{r_1}}{2}\geq \sqrt{\left( \sum_\sigma a^{p_1} b^{q_1} c^{r_1}\right) \left( \sum_\sigma a^{p_2} b^{q_2} c^{r_2}\right) } $$ but that is not good enough to prove the conjecture since it is not obvious that $\left( \sum_\sigma a^{p_1} b^{q_1} c^{r_1}\right) \left( \sum_\sigma a^{p_2} b^{q_2} c^{r_2}\right)$ is the same as $$ \left( \sum_\sigma a^{p_{12}} b^{q_{12}} c^{r_{12}}\right)^2 $$

Does anybody know how to prove this conjecture, or show that it is untrue?

$\endgroup$
  • 1
    $\begingroup$ But isn't the last step (which you mention as non-obvious) just Cauchy Schwartz inequality? $\endgroup$ – Macavity May 9 '17 at 17:35
  • $\begingroup$ Yes, @Macavity you are right. If you let the index $i$ in the C-S inequality range over the same order of permutations in both $\vec{u} = \sum a^{p_1/2} b^{q_1/2} c^{r_1/2} $ and $\vec{v} = \sum a^{p_2/2} b^{q_2/2} c^{r_2/2} $ then $$\left( \sum_\sigma a^{p_1} b^{q_1} c^{r_1}\right) \left( \sum_\sigma a^{p_2} b^{q_2} c^{r_2}\right) = (u\cdot v)^2 \geq |u|^2 |v|^2 = \left( \sum_\sigma a^{p_{12}} b^{q_{12}} c^{r_{12}}\right)^2 $$ If you transfer this to be an answer, the question can be closed. $\endgroup$ – Mark Fischler May 9 '17 at 18:34
  • $\begingroup$ OK did that. Though how far this generalises seems an interesting question. $\endgroup$ – Macavity May 9 '17 at 18:46
1
$\begingroup$

As mentioned in comments, your argument can be completed using CS inequality as: $$\frac{\sum_\sigma a^{p_1} b^{q_1} c^{r_1}+\sum_\sigma a^{p_1} b^{q_1} c^{r_1}}{2}\geq \sqrt{\left( \sum_\sigma a^{p_1} b^{q_1} c^{r_1}\right) \left( \sum_\sigma a^{p_2} b^{q_2} c^{r_2}\right) } \geqslant \sum_{\sigma} a^{p_{12}}b^{q_{12}}c^{r_{12}} $$ which Muirhead assures is not lower than $\displaystyle \sum_\sigma a^{p_3} b^{q_3} c^{r_3}$.

More generally, using the same approach, if $(p_{12}, q_{12}, r_{12})$ are weighted averages of $(p_1, q_1, r_1)$ and $p_2, q_2, r_2)$, then using weighted AM-GM and Holder, it appears one can reach the same conclusion.

$\endgroup$

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.