4

prove $$\sum_{i=1}^{n}\sqrt{a_i}\ge ({n-1})\sum_{i=1}^{n}\frac{1}{\sqrt{a_i}}$$ if $$\sum_{i=1}^{n}\frac{1}{1+a_i}=1$$

My try: i tried substituiting $y_i=\frac{1}{1+a_i}$ thus $\sum y_i=1$

also rearranging inequality we have to prove $$\sum_{i=1}^{n}(\sqrt{a_i}+\frac{1}{\sqrt{a_i}})\ge n\sum_{i=1}^{n}\frac{1}{\sqrt{a_i}}$$

putting value in terms of $y_i$ $$\sum_{i=1}^{n}\frac{1}{\sqrt{y_i(1-y_i)}}\ge n\sum_{i=1}^{n}\frac{\sqrt{y_i}}{\sqrt{1-y_i}}$$.i am stuck now i dont know which inequality to use.i tried using A-M$\ge$ H-M inequality.

source ' inequalities A mathematical Olympiad approach'

  • Is the index $i=0$ correct for the RHS? – user Aug 26 '20 at 06:36
  • @user Most likely not. Summation should go from $1$ to $n$ on both sides and it should involve a family of strictly positive reals indexed over the natural interval $[1, n]$. – ΑΘΩ Aug 26 '20 at 06:42
  • @user typo corrected! – Hari Ramakrishnan Sudhakar Aug 26 '20 at 06:45
  • Are you sure about that re-arrangement? – Gary Aug 26 '20 at 06:47
  • 1
    @Gary not only that, but we could do with more detailed references: where does this inequality come from? what is its background? – ΑΘΩ Aug 26 '20 at 06:52
  • your rearrangement is wrong. What you have rearranged, that is easy to prove but the rearrangement is not the same as original question. – Math Lover Aug 26 '20 at 06:54
  • it comes from " inequalities A mathematical Olympiad approach – Hari Ramakrishnan Sudhakar Aug 26 '20 at 06:54
  • @Math Lover i have fixed it – Hari Ramakrishnan Sudhakar Aug 26 '20 at 06:56
  • 1
    There are several solutions on AoPS: https://approach0.xyz/search/?q=%24%5Csum_%7Bi%3D1%7D%5E%7Bn%7D%5Csqrt%7Ba_i%7D%5Cge%20(%7Bn-1%7D)%5Csum_%7Bi%3D1%7D%5E%7Bn%7D%5Cfrac%7B1%7D%7B%5Csqrt%7Ba_i%7D%7D%24&p=1. – E.g. here: https://artofproblemsolving.com/community/c6h33608p208903 and here: https://artofproblemsolving.com/community/c6h5592 – Martin R Aug 26 '20 at 07:06
  • @Quantum Thank you for the reference. Not wanting to spoil the fun of the challenge, but the problem book in question is accessible online and also seems to include a full list of solutions to the exercises proposed... – ΑΘΩ Aug 26 '20 at 07:07
  • yes i know but the solution is very complicated .i am looking for a simpler solution – Hari Ramakrishnan Sudhakar Aug 26 '20 at 07:10
  • it involves double summing and i am not very familiar with that – Hari Ramakrishnan Sudhakar Aug 26 '20 at 07:11
  • @Quantum: as a word of pedagogic advice, it is then a clear understanding of general summation that you should seek for, rather than shortcut solutions to this or that particular contest inequality. The distinction you seem to make between ''simple'' summation and ''double'' summation is ultimately artificial, to the effect that not understading the latter very well brings into question to what extent you understand the former to begin with. – ΑΘΩ Aug 26 '20 at 07:21

4 Answers4

2

Rearrangement works again!

Let $a_i=\frac{\sum\limits_{i=1}^nx_i-x_i}{x_i},$ where $i\in\{1,2,...,n-1\}$ and $x_1$, $x_2$,...,$x_n$ are positives.

Thus, $a_n=\frac{\sum\limits_{i=1}^nx_i-x_n}{x_n}$ and we need to prove that: $$\sum_{i=1}^n\sqrt{\frac{\sum\limits_{i=1}^nx_i-x_i}{x_i}}\geq(n-1)\sum_{i=1}^n\sqrt{\frac{x_i}{\sum\limits_{i=1}^nx_i-x_i}}$$ and since the last inequality is homogeneous, we can assume that $\sum\limits_{i=1}^nx_i=n$ and we need to prove that: $$\sum_{i=1}^n\sqrt{\frac{n-x_i}{x_i}}\geq(n-1)\sum_{i=1}^n\sqrt{\frac{x_i}{n-x_i}}$$ or $$\sum_{i=1}^n\left(\sqrt{\frac{n-x_i}{x_i}}-(n-1)\sqrt{\frac{x_i}{n-x_i}}\right)\geq0$$ or $$\sum_{i=1}^n\frac{1-x_i}{\sqrt{x_i(n-x_i)}}\geq0.$$ Now, let $x_1\leq x_2\leq...\leq x_n.$

Thus, for $i>j$ we have $$1-x_i\leq1-x_j$$ and $$\frac{1}{\sqrt{x_i(n-x_i)}}\leq\frac{1}{\sqrt{x_j(n-x_j)}}$$ because the last it's $$x_j(n-x_j)\geq x_i(n-x_i)$$ or $$(x_i-x_j)(n-x_i-x_j)\geq0,$$ which is obvious.

Thus, $(1-x_1,1-x_2,...,1-x_n)$ and $\left(\frac{1}{\sqrt{x_1(n-x_1)}},\frac{1}{\sqrt{x_2(n-x_2)}},...,\frac{1}{\sqrt{x_n(n-x_n)}}\right)$ have the same ordering and by Chebyshov we obtain: $$\sum_{i=1}^n\frac{1-x_i}{\sqrt{x_i(n-x_i)}}\geq\frac{1}{n}\sum_{i=1}^n(1-x_i)\sum_{i=1}^n\frac{1}{\sqrt{x_i(n-x_i)}}=0$$ and we are done!

  • 2
    @Explicitly what is the family $x$ in relation to the given family $a$? how do you introduce it, how do you justify its existence? From the point of view of full rigour, that would be the only observation I have to make regarding your argument. – ΑΘΩ Aug 26 '20 at 07:24
2

We have that

$$\sum_{i=1}^{n}\sqrt{a_i}\ge ({n-1})\sum_{i=1}^{n}\frac{1}{\sqrt{a_i}} \iff \sum_{i=1}^{n}\left(\sqrt{a_i}+\frac{1}{\sqrt{a_i}}\right)\ge n\sum_{i=1}^{n}\frac{1}{\sqrt{a_i}}$$

and since

$$\sum_{i=1}^{n}\frac{1}{1+a_i}=\sum_{i=1}^{n}\frac{\frac{1}{\sqrt{a_i}}}{\frac{1}{\sqrt{a_i}}+\sqrt{a_i}}=1$$

by Chebyshev's Inequality we obtain

$$\sum_{i=1}^{n}\left(\sqrt{a_i}+\frac{1}{\sqrt{a_i}}\right)\cdot\sum_{i=1}^{n}\frac{\frac{1}{\sqrt{a_i}}}{\frac{1}{\sqrt{a_i}}+\sqrt{a_i}}\ge n\sum_{i=1}^{n}\frac{1}{\sqrt{a_i}}$$

indeed assuming wlog $a_i$ not decreasing by $x=\sqrt{a_i}$ we have that

  • $f(x)=f\left(\frac1{x}\right)=x+\frac1x \implies f'(x)=1-\frac1{x^2}$ is not decreasing for $x\ge 1$
  • $g(x)=\frac{\frac1x}{x+\frac1x}=\frac1{x^2+1}$ is not increasing

moreover we have that

$$\frac{1}{1+a_1}+\frac{1}{1+a_2}\le 1 \iff a_1a_2\ge 1 \iff \sqrt{a_1a_2}\ge 1\iff \sqrt{a_2}\ge\frac1{\sqrt{a_1}}$$

therefore if $a_1\le1$ we have that

$$f(\sqrt{a_1})= f\left(\frac1{\sqrt{a_1}}\right)\le f(\sqrt{a_2})$$

and the condition for the application of the inequality is preserved.

user
  • 162,563
2

$\sum \limits_{i=i}^{n}\frac{1}{1+a_i}=1$ ...(i)

So $a_i \gt 0$ for all $i \in (1, n)$ and for $n \gt 1$.

We have to prove $\sum \limits_{i=1}^{n}\sqrt{a_i}\ge ({n-1})\sum \limits_{i=1}^{n}\frac{1}{\sqrt{a_i}}$

or prove, $\sum _\limits{i=1}^{n} \dfrac{1 + a_i}{\sqrt{a_i}} \ge n\sum \limits_{i=1}^{n}{\frac{1}{\sqrt a_i}}$

or prove, $(\sum _\limits{i=1}^{n} \dfrac{1 + a_i}{\sqrt{a_i}}) \, (\sum \limits_{i=1}^{n} \dfrac{1}{1+a_i}) \ge n\sum \limits_{i=1}^{n}\frac{1+a_i}{\sqrt a_i}{\frac{1}{1+a_i}}$ ...(ii)

WLOG, assume, $a_i \le a_{i+1}$ for $1 \le i \le (n-1)$.

Say, $f(a_i) = \dfrac {1+a_i}{\sqrt a_i} = \sqrt a_i + \dfrac{1}{\sqrt a_i}$. So, for all $a_i \gt 0$, $f(a_i) = f(\dfrac{1}{a_i})$

Say, $g(a_i) = \dfrac {1}{1+a_i}$. We can see $g(a_{i+1}) \le g(a_i)$.

We can see $g(a_i)$ is non-increasing. We now need to prove that $f(a_i)$ is non-decreasing. With that (ii) holds good as per Chebyshev's inequality.

It is easy to see that the function is non-decreasing for $a_i \gt 1$. It is also easy to see that $a_i$ can be $\lt 1$ only for one value of $i$, at max, due to given condition (i).

Say, $a_1 \lt 1$. We also know that $\dfrac {1}{1+a_2} \le 1 - \dfrac {1}{1+a_1} = \dfrac{1}{1+\dfrac{1}{a_1}}$.

So, $a_2 \ge \dfrac{1}{a_1}$ and $f(a_1) = f(\dfrac{1}{a_1}) \le f(a_2) \le ... \le f(a_n)$

With this, we prove that (ii) holds good as per Chebyshev's inequality.

Math Lover
  • 52,200
  • 1
    I lost at first that important detail assuming $f$ not decreasing as a simple condition. I've now rielaborated my answer including that. Thanks – user Aug 26 '20 at 10:26
0

For this solution, I will only use this inequality, AM-GM Inequality and C-S. We need to prove that: $$\sum_{i=1}^{n}\sqrt{a_i}\ge ({n-1})\sum_{i=1}^{n}\frac{1}{\sqrt{a_i}}$$ if $$\sum_{i=1}^{n}\frac{1}{1+a_i}=1$$

Case I: Assume $a_i\geq1$ for all $i$ till $n$

$$1=\sum_{i=1}^{n}\frac{1}{1+a_i}\geq {n^2\over n+\sum_{i=0}^n{a_i}}\Rightarrow \sum_{i=0}^n{a_i}\geq n(n-1)\Rightarrow n+\sum_{i=0}^n{a_i}\geq n(n-1)+n$$

$$ \Rightarrow 2\sum_{i=0}^n{\sqrt{a_i}}\geq n(n-1)+n $$

As $\sum_{i=0}^n{\sqrt{a_i}}\geq n$,

$$ \sum_{i=0}^n{\sqrt{a_i}}\geq n(n-1) $$

Now it's enough to prove that

$$ n(n-1)\geq (n-1)\sum_{i=1}^{n}\frac{1}{\sqrt{a_i}} \Rightarrow n \geq \sum_{i=1}^{n}\frac{1}{\sqrt{a_i}}$$ which is obvious for this case.

Case II:
Assume that some $a_i \ge 1$ and others $\le 1$.
Note that, as mentioned in @MathLover's solution, only one $a_i$ can be $\leq 1$. Let $a_1$ be less than $1$ and rename others $a_{2},...,a_n$ where $1\leq k\leq n$.
Just to reduce this case hence the solution, the cases that arise in this case are left, as resolving them is very easy, in the sense that they are similar to resolving case I as before. (and the other is very easy) Just assume $$\frac{1}{1+a_1}=p,0\leq p\leq 1$$ for one case in this case, and $$\sum_{i=2}^{n}\frac{1}{1+a_i}=1-p,0\leq 1-p\leq 1$$ for the other. Just prove the original inequality for these two cases, and sum up the hypotheses and the inequalities to get what was desired. Done!