If $x_1,x_2,\ldots,x_n>0$ and $x_1+x_2+\ldots+x_n=1$, prove that $$\frac{x_1}{1+x_2+x_3+\ldots+x_n} + \frac{x_2}{1+x_1+x_3+\ldots+x_n} +\ldots + \frac{x_n}{1+x_1+x_2+\ldots +x_{n-1}} \ge \frac {n}{2n-1}$$ This can easily be simplified: $$\frac{x_1}{2-x_1} + \frac{x_2}{2-x_2} +\ldots + \frac{x_n}{2-x_n} \ge \frac {n}{2n-1}$$

I could try using the Cauchy-Schwarz inequality: $$\frac{x_1}{2-x_1} + \frac{x_2}{2-x_2} +\ldots + \frac{x_n}{2-x_n} = \frac{x_1^2}{2x_1-x_1^2} + \frac{x_2^2}{2x_2-x_2^2} +\ldots + \frac{x_n^2}{2x_n-x_n^2} \ge \frac{(x_1+x_2+\ldots+x_n)^2}{2(x_1+x_2+\ldots+x_n)-(x_1^2+x_2^2+\ldots+x_n^2)}=\frac{1}{2-(x_1^2+x_2^2+\ldots+x_n^2)}=\frac{n}{2n-(x_1^2+x_2^2+\ldots+x_n^2)n}$$

It's left to prove that $$(x_1^2+x_2^2+\ldots+x_n^2)n \ge 1$$ or $$(x_1^2+x_2^2+\ldots+x_n^2)n=1$$I can continue using the Cauchy-Schwarz inequality again: $$(x_1^2+x_2^2+\ldots+x_n^2)n=(x_1^2+x_2^2+\ldots+x_n^2)(1+1+\ldots+1)\ge(x_1+x_2+\ldots+x_n)^2=1$$ So I've proved it myself. (I've shown this proof after editing, I didn't post the question with the solution in the details. I found the proof a while after putting the question here).

  • $\begingroup$ disregard my response its incorrect $\endgroup$ – frogeyedpeas Dec 22 '13 at 19:14

For any $x,y \in (0,2)$, we have $$\frac{x}{2-x} - \frac{y}{2-y} = \frac{2(x-y)}{(2-x)(2-y)} = 2\left(\frac{x-y}{2-y}\right)\left[\left(\frac{1}{2-x} - \frac{1}{2-y}\right) + \frac{1}{2-y}\right] = 2\left(\frac{x-y}{2-y}\right)\left[\frac{x-y}{(2-x)(2-y)} + \frac{1}{2-y}\right] = \frac{2(x-y)^2}{(2-x)(2-y)^2} + \frac{2(x-y)}{(2-y)^2} \ge \frac{2(x-y)}{(2-y)^2} $$ Substitute $x$ by $x_i$ and $y$ by $\frac{1}{n}$ and then sum over $i$, we get

$$ \sum_{i=1}^n \frac{x_i}{2-x_i} - \frac{n}{2n-1} = \sum_{i=1}^n \left(\frac{x_i}{2-x_i} -\frac{\frac{1}{n}}{2-\frac{1}{n}}\right) \ge \frac{2}{(2-\frac{1}{n})^2}\sum_{i=1}^n \left( x_i - \frac{1}{n}\right) = \frac{2}{(2-\frac{1}{n})^2} \left(\sum_{i=1}^n x_i - 1 \right) = 0$$ The steps look almost magical. How do I come up with that? The answer is we are using Jensen's inequality from behind.

Jensen's inequality is actually more generic and applicable beyond differentiable functions. If you have any function $f(x)$ whose graph curved up on both side at every point, then Jensen's inequality works. Such functions are called convex functions. For any convex function $f(x)$, Jensen's inequality tells us:

$$\frac{1}{n} \sum_{i=1}^n f(x_i)\;\;\ge\;\; f(\frac{1}{n} \sum_{i=1}^n x_i)$$

This means if you want to verify $\sum_{i=1}^n f(x_i) \ge$ some number $M$. You just need to check what happens when all $x_i$ are equal to each other.

In this case, we are told we cannot use Jensen's inequality. So we expand our target function $f(x_i)$ around $\frac{1}{n}$, the mean of the $x_i$'s, and we are sure after we cancel the linear parts, the rest of the expansion will be non-negative.

  • $\begingroup$ The supposed link to Jensen's inequality is a repeat of the link to convex functions. $\endgroup$ – John Bentin Dec 22 '13 at 22:31
  • $\begingroup$ @JohnBentin Oops, fixed that. thanks for pointing out the mistake. $\endgroup$ – achille hui Dec 23 '13 at 3:22
  • $\begingroup$ Is there a way of finding out whether a function is convex without using calculus? $\endgroup$ – user26486 Dec 26 '13 at 16:48
  • $\begingroup$ @mathh, When dealing with complicated function, if possible, the first thing I do is make a plot of it. It usually give you a rough idea what you are dealing with. In this case, a plot of $\frac{x}{2-x}$ tells us it is convex over $[0,1]$. $\endgroup$ – achille hui Dec 26 '13 at 17:24

Writing ${\displaystyle {x_i \over 2 - x_i} = -1 + {2 \over 2 - x_i}}$, you have $$\frac{x_1}{2-x_1} + \frac{x_2}{2-x_2} +\ldots + \frac{x_n}{2-x_n} = -n + \frac{2}{2-x_1} + \frac{2}{2-x_2} +\ldots + \frac{2}{2-x_n}$$ So it suffices to show $$ \frac{2}{2-x_1} + \frac{2}{2-x_2} +\ldots + \frac{2}{2-x_n} \geq n + \frac{n}{2n-1}$$ This is the same as $$\frac{1}{2-x_1} + \frac{1}{2-x_2} +\ldots + \frac{1}{2-x_n} \geq \frac{n^2}{2n - 1}$$ Letting $y_i = 2 - x_i$, this is the same as showing $$\frac{1}{y_1} + \frac{1}{y_2} +\ldots + \frac{1}{y_n} \geq \frac{n^2}{2n - 1}$$ The condition $\sum_i x_i = 1$ translates into $\sum_i y_i = 2n - 1$. By the arithmetic-harmonic mean inequality, $$n\bigg(\frac{1}{y_1} + \frac{1}{y_2} +\ldots + \frac{1}{y_n}\bigg)^{-1} \leq {1 \over n}\sum_i y_i $$ $$= {2n - 1 \over n}$$ This is equivalent to what you want to show.


Hint: Use Jensen's inequality for f(x) = x/(2 -x)

  • $\begingroup$ it's quite simple. Jensen is an elaborate name for second derivative positive. Basically f(ax + by) < af(x) + bf(y) for a + b = 1 and f''(x) > 0. $\endgroup$ – DeepSea Dec 22 '13 at 19:11
  • $\begingroup$ I'm very young and haven't even started learning about derivatives at school, so sadly this solution won't work for me. $\endgroup$ – user26486 Dec 22 '13 at 19:16

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.