## ParthKohli one year ago Help.

1. ParthKohli
2. ParthKohli

UUse the weighted means inequality.

3. dan815

solve this one first |dw:1437428390898:dw|

4. dan815

u can math it pure, like an optimization problem with a restriction, lagrnge multipled stuff DelF =k-DelG

5. dan815

6. dan815

im trying to think about it like perimeter problems with max area but iwth a twist

7. anonymous

AM-GM gives: $$\frac13=\frac13 \left(\frac{\lambda^2}{d_1^2}+\frac{\mu^2}{d_2^2}+\frac{\omega^2}{d_3^2}\right)\ge \sqrt[3]{\frac{\lambda^2\mu^2\omega^2}{d_1^2 d_2^2 d_3^2}}\\\implies (\lambda\mu\omega)^2\le \frac{(d_1 d_2 d_3)^2}{27}\\\implies\lambda \mu \omega\le\frac{d_1 d_2 d_3}{3\sqrt3}$$

8. anonymous

AM-GM and its derivatives (incl. weighted AM-GM) are consequence of the fact that $$\exp$$ is convex, so by Jensen's inequality consider points $$x_i$$ and weights $$w_i$$ such that $$\sum w_i=1$$: $$\exp\left(\sum_i w_i \log x_i\right)\le\sum_i w_i\exp(\log x_i)\\\prod_i \exp(w_i\log x_i)\le\sum_i w_i x_i\\\prod_i x_i^{w_i}\le\sum_i w_i x_i$$

9. anonymous

for the AM-GM inequality with $$n$$ such points choose $$w_i=\frac1n$$: $$\prod_i x_i^{1/n}\le\sum_i\frac{x_i}n\\\left(\prod_i x_i\right)^{1/n}\le\frac1n\sum_i x_i\\\sqrt[n]{\prod_i x_i}\le\frac1n\sum x_i$$

10. ParthKohli

@oldrin.bataku The last term is not $$\omega^2$$; it's just $$\omega$$. Hence the weights.

11. dan815

ya i considered AM -GM stuff too

12. ParthKohli

Y'know, sometimes, you can see the proportion of the weights. Like in this question: If $$a+b + c= 10$$, what is the maximum value of $$a^3 b^2 c^5$$? It's obvious that $$a=3,~ b=2, ~c=5$$.

13. ganeshie8
14. dan815

herse the one with your domain http://www.wolframalpha.com/input/?i=max+xyz%2C+x^2%2Fpi^2%2By^2%2Fe^2%2Bz%2FGoldenRatio^2+%3D+1

15. ParthKohli

OK, sorta got it using pure AM-GM (but weighted in disguise):$\dfrac{\lambda^2/3d_1^2 + \lambda^2/3d_1^2 + \lambda^2/3d_1^2 + \mu^2/3d_2^2 + \mu^2/3d_2^2 + \mu^2/3d_2^2 + \nu/6d_3^2 + \cdots +\nu/6d_3^2}{12}$$\ge \sqrt[12]{\frac{\lambda^6\mu^6 \nu^6}{3^6 6^6 d_1^6 d_2^6 d_3^6}}$$\Rightarrow (1/12) \cdot \sqrt{18} \cdot \sqrt{d_1d_2d_3 }\ge \sqrt{\lambda \mu \nu}$$\Rightarrow \lambda\mu\nu \le \frac{d_1d_2d_3}{8}$Of course the answer is wrong. :P

16. anonymous

are you sure it's not a typo?

17. anonymous

the point of 'weighted' here is simply because the factors $$1/d_1^2,1/d_2^2,1/d_3^2$$ are the weights for $$\lambda^2,\mu^2,\omega^2$$, but we can just absorb them into the variables for AM-GM so it's actually unnecessary. you are overanalyzing a typo, it's supposed to be $$\omega^2$$

18. anonymous

also your work is incorrect since you should actually get $$d_3^{12}$$, and it's very much overcomplicated (you don't need to expand each in multiples of 3 terms, only the last in twice as many as the others): $$\frac14=\frac14\left(\lambda^2/d_1^2+\mu^2/d_2^2+\omega/(2d_3^2)+\omega/(2d_3^2)\right)\ge \sqrt[4]{\frac{\lambda^2\mu^2\omega^2}{4d_1^2d_2^2d_3^4}}\\\implies \lambda^2\mu^2\omega^2\le\frac{d_1^2 d_2^2 d_3^4}{64}\\\implies |\lambda\mu\omega|\le \frac{d_1 d_2 d_3^2}8$$

19. anonymous

also note that upper bound only works for $$d_1, d_2, \omega\ge 0$$, which is where AM-GM holds; if you allow negatives, then you can make $$x,y$$ as large as possible and then $$-z$$ sufficiently big to satisfy the inequality, whichc learly suggests no upper bound

20. ParthKohli

Of course I know that we can apply pure AM-GM if the question says $$\omega^2$$. What would the answer have been had it been $$\omega$$?

21. anonymous

oops, i meant $$\lambda^2,\mu^2$$ big as possible and then $$-\omega$$ sufficiently big to maintain the inequality; you would need the condition $$\omega\ge 0$$ for there to be a maximum for the reason i gave above, but if you had that, then the maximum would be $$|\lambda\mu|\omega\le\frac{|d_1 d_2| d_3^2}8$$

22. anonymous

if you have everything nonnegative then it gives $$\lambda \mu\omega\le\frac{d_1 d_2 d_3^2}8$$ by the argument i gave

23. ParthKohli

Oh, ew, of course. I found that on my own too (almost).

Find more explanations on OpenStudy