25

let $a,b>0.$ Show that $$\sqrt{\dfrac{a^b}{b}}+\sqrt{\dfrac{b^a}{a}}\ge 2\tag{1}$$

I known How to prove $a^b+b^a>1$,where $a,b>0.$ See $x^y+y^x>1$ for all $(x, y)\in \mathbb{R_+^2}$

to prove $(1)$, I want use AM-GM inequality $$\sqrt{\dfrac{a^b}{b}}+\sqrt{\dfrac{b^a}{a}}\ge 2\left(\dfrac{a^b}{b}\cdot\dfrac{b^a}{a}\right)^{1/4}=2\left(a^{b-1}b^{a-1}\right)^{1/4}$$

But $a^{b-1}b^{a-1}$ is not always $>1$

Thomas Andrews
  • 177,126
math110
  • 93,304
  • 3
    We can rewrite as $A^{B^2+1}+B^{A^2+1}\ge2AB$ which is similar to this problem. – ə̷̶̸͇̘̜́̍͗̂̄︣͟ Oct 09 '21 at 10:42
  • could we consider different cases? One possibility would be to go if they are both > 1, both < 1 or one >1 and the other <1. If you can prove those cases, you have proved everything. Another possibility would be to assume a > b and try to prove it for that – Marin Althuis Oct 12 '21 at 12:48
  • I am kind of surprised that this question doesn't show much effort from the OP. But it's getting so many upvotes. – Rounak Sarkar Oct 13 '21 at 13:29
  • 4
    @RounakSarkar I believe there's sufficient context. They have linked to a relevant inequality, shown an attempt and identified the particular issue in their attempt. – ə̷̶̸͇̘̜́̍͗̂̄︣͟ Oct 13 '21 at 15:24
  • yes and inequalities like these are just entertaining on their own when it's not the usual three-variable, symmetric/cyclic and homogeneous ones that are more or less considered well understood. – dezdichado Oct 13 '21 at 16:19
  • 2
    @RounakSarkar Interest isn't enough to keep a question up on its own for sure, but you'll slowly get a hang of how context checking works in the contest/inequality setting. Since an attempt is quite difficult to provide, "source" and "linked-questions" are the two preferred ways of adding context, and there is a (somewhat) useful linked question here, and there is an attempt (although I'd put a source above that, but the attempt shows that the question isn't too trivial to admit AM-GM immediately). So this is ok for a contest/inequality type problem. – Sarvesh Ravichandran Iyer Oct 14 '21 at 05:18
  • 1
    This seems to be a hard nut. Some remarks: If $a, b \ge 1$ or $a, b \le 1$ then $a^{b-1}b^{a-1} \ge 1$ and the conclusion follows (that are the easy cases). If $0 < a \le 1/4$ and $b > 1$ then $\sqrt{b^a/a} \ge \sqrt{1/a} \ge 2$. If $1/4 < a < 1$ and $b \ge 256$ then $\sqrt{b^a/a} \ge \sqrt{b^a} \ge 2$. – It remains “only” the rectangle $1/4 < a < 1$, $1 < b < 256$ to investigate. – Martin R Oct 14 '21 at 06:56
  • 3
    @MartinR We can do better :Note that if either $\sqrt{\frac{a^b}{b}} \geq 2$ or $\sqrt{\frac{b^a}{a}} \geq 2$ then we are done. The former case translates to $a \geq (4b)^{\frac 1b}$ and the latter to $b$ switched with $a$. The function $(4b)^{\frac 1b}$ attains a maximum of $e^{\frac 4e} \approx 4.35$ over $b>0$, therefore if either $b$ or $a$ is above $4.35$ then the inequality holds because that particular term on the LHS is above $2$. So this further restricts the domain of investigation. – Sarvesh Ravichandran Iyer Oct 14 '21 at 07:03
  • 1
    Yes, I am stuck at where @Teresa Lisbon is at for some time now. – dezdichado Oct 14 '21 at 12:56
  • 1
    @dezdichado I know you might be following this question so you will already know this, but there's an answer down below by River and I honestly don't know what is wrong with it and why it's on $-2$, but you might want to see it if you're trying to follow a calculus-type argument on the part of the domain that's left out after we exclude everything we discussed above. On a perusal I see nothing wrong with the answer, but still wondering why the downvotes. Thanks. – Sarvesh Ravichandran Iyer Oct 14 '21 at 18:04
  • 1
    @TeresaLisbon yes I just saw that and upvoted. I am not going to check the details; however, RiverLi is perhaps the only one on this site capable of doing this type of inequalities. I think the downvotes come from people who want to see pen and paper solutions but non-homogeneous and tight inequalities like this hardly ever such solutions.

    I think the linked question with 84 upvotes is basically where this type of questions started getting attentions.

    – dezdichado Oct 14 '21 at 18:08
  • @dezdichado Thanks for the feedback, I'll check the answer once. – Sarvesh Ravichandran Iyer Oct 14 '21 at 18:23
  • 1
    @TeresaLisbon Thanks. My proof is not nice. As @ dezdichado pointed out, a nice proof (even by hand) is expected. – River Li Oct 15 '21 at 02:03

6 Answers6

10

Remarks (2022/04/01): @Erik Satie's proof is simpler than mine.

Sketch of @Erik Satie's proof:

Fact 3: Let $0 < y \le 1 \le x$. Then $$x^{y^2} \ge 1 + \left(x \cdot \frac{1}{1 - (x - 1)(y - 1)} - 1\right) y.$$

Fact 4: Let $0 < y \le 1 \le x$. Then $$y^{x^2} \ge 1 + \left(y \cdot \frac{1}{1 - (x - 1)(y - 1)} - 1\right) x.$$

By Facts 3-4, it suffices to prove that $$\frac{1}{x}\left[1 + \left(\frac{x}{1 - (x - 1)(y - 1)} - 1\right) y\right] + \frac{1}{y}\left[1 + \left(\frac{y}{1 - (x - 1)(y - 1)} - 1\right) x\right] \ge 2$$ or $$\frac{(2xy - x - y)^2}{(x + y - xy)xy} \ge 0$$ which is true.



$\phantom{2}$

Update: I found a simpler proof.


WLOG, assume that $b \le a$.

If $a, b > 1$ or $a, b < 1$, then $a^{b - 1}\ge 1$ and $b^{a - 1}\ge 1$ and thus $$\sqrt{\frac{a^b}{b}} + \sqrt{\frac{b^a}{a}}\ge 2\sqrt[4]{a^{b-1}b^{a-1}} \ge 2.$$

It remains to prove the case when $0 < b \le 1 \le a$.

Let $a = x^2, b = y^2$. It suffices to prove that, for all $0 < y \le 1 \le x $, $$\frac{x^{y^2}}{y} + \frac{y^{x^2}}{x} \ge 2.$$

$\phantom{2}$

Fact 1: If $x \ge 1$ and $0 < y \le 1$, then $$x^{y^2} \ge \frac{1 + x + (x - 1)y^2}{1 + x - (x - 1)y^2}.$$ (Proof: Let $f(x) = y^2\ln x - \ln \frac{1 + x + (x - 1)y^2}{1 + x - (x - 1)y^2} $. We have $f'(x) = \frac{(1 - y^4)y^2(x - 1)^2}{x[(1 + x)^2 - (x - 1)^2y^4]}\ge 0$. Also, $f(1) = 0$. Thus, $f(x) \ge 0$ for all $x\ge 1$.)

Fact 2: If $x \ge 1$ and $0 < y \le 1$, then $$y^{x^2} \ge \frac{1 + y + (y - 1)x^2}{1 + y - (y - 1)x^2}.$$ (The proof is given at the end.)

$\phantom{2}$

Now, using Facts 1-2, it suffices to prove that $$\frac{1}{y}\cdot \frac{1 + x + (x - 1)y^2}{1 + x - (x - 1)y^2} + \frac{1}{x}\cdot \frac{1 + y + (y - 1)x^2}{1 + y - (y - 1)x^2} \ge 2.$$

Let $x = 1 + s$ for $s \ge 0$. After clearing the denominators, it suffices to prove that $$q_4 s^4 + q_3 s^3 + q_2 s^2 + q_1s + q_0 \ge 0 \tag{1}$$ where \begin{align*} q_4 &= (1 - y)(2y^3 + y^2 - 2y + 1), \\ q_3 &= (1 - y)(7y^3 + 3y^2 - 11y + 5), \\ q_2 &= - 6y^4 + 8y^3 + 24y^2 - 32y + 10, \\ q_1 &= -2y^4 + 4y^3 + 16y^2 - 28y + 10, \\ q_0 &= 4(1 - y)^2. \end{align*} It is easy to prove that $q_4, q_3, q_2, q_0 \ge 0$. Also, we have \begin{align*} 4q_2q_0 - q_1^2 = 4(y^3 + y^2 + 7y + 15)(1 - y)^5 \ge 0. \end{align*} Thus, (1) is true.

We are done.


Proof of Fact 2: We only need to prove the case when $\frac{1 + y + (y - 1)x^2}{1 + y - (y - 1)x^2} > 0$, i.e. $y > \frac{x^2 - 1}{x^2 + 1}$. In other words, we only need to prove the case when $\frac{x^2 - 1}{x^2 + 1} < y \le 1$. Let $$g(y) = x^2\ln y - \ln \frac{1 + y + (y - 1)x^2}{1 + y - (y - 1)x^2}.$$ We have $$g'(y) = - \frac{x^2(x^4 - 1)(1 - y)^2}{y[1 + y + (y - 1)x^2]^2}\cdot \frac{1 + y + (y - 1)x^2}{1 + y - (y - 1)x^2} \le 0.$$ Also, $g(1) = 0$. Thus, $g(y) \ge 0$ for all $y \in (0, 1]$.

We are done.

River Li
  • 37,323
  • Good solution! It is very useful – Sickness Oct 14 '21 at 22:44
  • 1
    @3017 Thanks. Honestly, it is not nice. :) – River Li Oct 15 '21 at 02:00
  • Nice!+1,can you explain How to found the fact 1 and fact 2? – math110 Oct 15 '21 at 06:34
  • @functionsug It is just (1,1)-Pade approximant of $x^{y^2}$ at $x = 1$. – River Li Oct 15 '21 at 06:57
  • @River Li.-Good your answer to this defiant question. Congratulations. It would be even easier if we consider for all point $(a,b)$ in the first quadrant the following function of $[a,b]$ (or $[b,a]$ to have a sens) in $\mathbb R$ $$F(x)=\sqrt{\frac{x^{s-x}}{s-x}}+\sqrt{\frac{(s-x)^x}{x}}$$ and determine its minimum. This minimum is always greater than 2 and is the smallest possible value for all point $(a,b)$ in the line $x+y=s$

    Unfortunately this minimum should be calculated using numerical methods, which is contrary to the local maximun equal to $F(\frac s2)$. (suite)

    – Piquito Oct 16 '21 at 16:44
  • @River Li.- All preliminary is not necessary in my second answer. Enough with the final part. The value of $s$ is $a+b$ in the definition of $F$ (I forget to say this). – Piquito Oct 16 '21 at 17:15
  • @Piquito Do you mean to use numerical method? I think a rigorous non-numerical proof is preferred. – River Li Oct 16 '21 at 17:21
  • River Li: I agree of course. – Piquito Oct 18 '21 at 19:33
1

My try (please point out the errors if any!):

Multiplying both sides by $\sqrt{ab}$, we get that $$\sqrt{a^{b+1}}+\sqrt{b^{a+1}}\geq 2\sqrt{ab}\tag{1}$$ $$\frac{\sqrt{a^{b+1}}+\sqrt{b^{a+1}}}{2}\geq \sqrt{ab}$$ Using AM-GM inequality,

$$\frac{\sqrt{a}+\sqrt{b}}{2}\geq \sqrt{\sqrt{ab}}$$

For $0\leq a\leq b\leq 1$, we have that $\sqrt{\sqrt{ab}}\geq\sqrt{ab}$, $\sqrt{a^{b+1}}\geq \sqrt{a}$ and $\sqrt{b^{a+1}}\geq \sqrt{b}$. Therefore, $$\frac{\sqrt{a^{b+1}}+\sqrt{b^{a+1}}}{2}\geq\frac{\sqrt{a}+\sqrt{b}}{2}\geq \sqrt{\sqrt{ab}}\geq \sqrt{ab}$$

Thus, the original inequality holds.

For $1\leq a\leq b$, we can transform $(1)$ into $$a^{b+1}+b^{a+1}+2{a^{\frac{b+1}{2}}b^{\frac{a+1}{2}}}\ge 4ab$$

As ${a^{\frac{b+1}{2}}b^{\frac{a+1}{2}}}\geq ab$, it follows that $4ab-2{a^{\frac{b+1}{2}}b^{\frac{a+1}{2}}}\leq 2$. As $a^{b+1}+b^{a+1}\geq 2$, the original inequality holds in this case too.

QED

Juan Moreno
  • 1,110
1

Alternative sketch of proof for $0<x\leq 1\leq a$

Using derivative we have the inequalities :

$$\left(1+\left(a\cdot\frac{1}{1-\left(a-1\right)\left(x-1\right)}-1\right)x\right)\leq a^{x^2}$$

$$\left(1+\left(x\cdot\frac{1}{1-\left(a-1\right)\left(x-1\right)}-1\right)a\right)-x^{a^{2}}\leq 0$$

Then we need to show :

$$\frac{\left(1+\left(a\cdot\frac{1}{1-\left(a-1\right)\left(x-1\right)}-1\right)x\right)}{x}+\frac{\left(1+\left(x\cdot\frac{1}{1-\left(x-1\right)\left(a-1\right)}-1\right)a\right)}{a}\geq 2$$

Wich is easy !

1

My second proof:

We use isolated fudging.

It suffices to prove that, for all $a, b > 0$, $$\sqrt{\frac{a^b}{b}} \ge \frac{3\sqrt{a} - \sqrt{b}}{\sqrt{a } + \sqrt{b}}. \tag{1}$$ (Summing cyclically on (1), the desired result follows. The proof of (1) is given at the end.)

We are done.

$\phantom{2}$


Proof of (1):

Letting $x = \sqrt{a/b}$, the desired inequality is written as $$b^{(b-1)/2}x^b \ge \frac{3x-1}{x+1}.$$

We only need to prove the case that $x > 1/3$.

Taking logarithm, it suffices to prove that, for all $b > 0$ and $x > 1/3$, $$f(b, x) := \frac{b-1}{2}\ln b + b\ln x - \ln \frac{3x-1}{x+1} \ge 0. $$

It is not difficult to prove that $f(b, x) \ge 0$ on the boundary of $b> 0, x > 1/3$.

It remains to prove that $f(b, x) \ge 0$ for all stationary points in the interior of the region.

We have \begin{align*} \frac{\partial f}{\partial b} &= \frac12\ln b + \frac{b-1}{2b} + \ln x, \\[6pt] \frac{\partial f}{\partial x} &= \frac{(x+1)(3x-1)b - 4x}{(3x-1)(x+1)x}. \end{align*}

We claim that $\frac{\partial f}{\partial b} = \frac{\partial f}{\partial x} = 0$ has exactly one solution $(b, x) = (1, 1)$ on $b > 0, x > 1/3$. Indeed, from $\frac{\partial f}{\partial x} = 0$, we have $b = \frac{4x}{(x+1)(3x-1)}$. Inserting $b = \frac{4x}{(x+1)(3x-1)}$ into $\frac{\partial f}{\partial b} = 0$, we have $$h(x) := \ln\frac{4x^3}{(3x-1)(x+1)} - \frac{(3x+1)(x-1)}{4x} = 0.$$ We have $$h'(x) = -\frac{(9x^2+12x-1)(x-1)^2}{4x^2(3x-1)(x+1)}.$$ Thus, we have $h'(x) < 0$ on $(1/3, 1) \cup (1, \infty)$. Also, we have $h(1) = 0$. Thus, $h(x) = 0$ has exactly one solution $x = 1$. Then $b = 1$. The claim is proved.

Note that $f(1, 1) = 0$. Thus, we have $f(b, x) \ge 0$ for all stationary points in the interior of the region.

We are done.

River Li
  • 37,323
-1

Using again Generalized Young inequality we have :

$$\left(\frac{1}{a^{b}}+\frac{1}{x^{b}}\right)\left(a^{\sqrt{2^{-1}}x^{\left(1-b\right)}}\cdot x^{\left(\sqrt{2^{-1}}\left(a\right)^{\left(1-b\right)}\right)}\cdot a^{\frac{\left(b-0.5\right)\sqrt{2}}{a^{b}}}\cdot x^{\frac{\left(b-0.5\right)\sqrt{2}}{x^{b}}}\right)^{\frac{1}{\sqrt{2}a^{-b}+\frac{\sqrt{2}}{x^{b}}}}\leq \left(\sqrt{\frac{x^{a}}{a}}+\sqrt{\frac{a^{x}}{x}}\right)$$

where $b\to1$ and $a,x>0$

Final conjecture :

Let $a,x>0$ then it seems we have :

$$\left(\frac{1}{a^{b}}+\frac{1}{x^{b}}\right)\left(a^{\sqrt{2^{-1}}x^{\left(1-b\right)}}\cdot x^{\left(\sqrt{2^{-1}}\left(a\right)^{\left(1-b\right)}\right)}\cdot a^{\frac{\left(b-0.5\right)\sqrt{2}}{a^{b}}}\cdot x^{\frac{\left(b-0.5\right)\sqrt{2}}{x^{b}}}\right)^{\frac{1}{\sqrt{2}a^{-b}+\frac{\sqrt{2}}{x^{b}}}}\geq 2$$

Where $b\to 1$

I think we can settle $b=1$ it works also .

Using a bit of algebra (introducing log and invert the variables)

We need to show :

$x,a > 0$, we need to prove that $$f\left(x\right)=(x+a)\ln\frac{a+x}{2}-\frac{\left(a+1\right)\ln a+\left(x+1\right)\ln x}{2} \ge 0.$$

A proof of this fact can be found here :

Prove $2(x + y)\ln \frac{x + y}{2} - (x + 1)\ln x - (y + 1)\ln y \ge 0$

-1

COMMENT.- I have trouble writing in English so this schematic presentation.

$f(x,y)=\sqrt{\dfrac{x^y}{y}}$

► One has to prove for all point $(a,b)$ of the first quadrant $$f(a,b)+f(b,a)\ge 2$$ ► For all positive value $k$ the curves $f(x,y)=k$ and $f(y,x)=k$ are symmetric respect to the line $y=x$.

enter image description here

► One can use the black curve above to reduce the problem to prove for the complement of the region defined by $$f(x,y)\ge 2$$ because for any point in the shadow region the proposed inequality is trivially verified (one term of the sum, $f(x,y)$, is already greater than $2$). Note that a neighborhood of $0$ is discarded which is “natural” because a denominator very small is involved.

enter image description here

► Now we can reduce the white region with the symmetric function $f(y,x)=\sqrt{\dfrac{y^x}{x}}=2$ (note for all point $(b,a)$ “above” this curve its symmetric point $(a,b)$ satisfies $f(a,b)\gt2$ so $f(a,b)+f(b,a)\ge 2$).

enter image description here

► So it remains to prove the inequality for the points inside the white region.

HINT.- Let the line $L: y = -x + a$ where $a$ runs through the interval $[0.36,5.02]$ (these numbers correspond to the values of $a$ for the points of intersection of the two considered symmetric curves). We are going to prove that each point $(a,b)$ of the segment $\overline{PQ}$, where $P$ and $Q$ are the points of intersection of the line with said curves, satisfies the inequality.

enter image description here

For this, let $F$ be the function $$F(x)=\sqrt{\frac{x^{a-x}}{a-x}}+\sqrt{\frac{(a-x)^x}{x}}$$ and study its variation. For example, for $a=4$, if the point $P=(x_0,y_0)$ then $x_0\approx 0.426$ and $F(0.426)\approx2.1247\gt2$ (ideally obviously must be equal to $2$). At the point where $x=y$, when $x=2$ (because of $x=-x+4$) one has $F(2)\approx2.8284\gt2$ and the minimum of $F(x)$is equal to $2.087\gt2$ and the study of the curve can be stop here (by symmetry). Similarly with all other segment with $a\in[0.36,5.02]$ , the problem with $a\ne4$ is analogue. You can verify that for values of $a$ in $[1.77,2.3]$ the minimum of the function $F(x)$ is very near of $2$ but always greater that $2$.

It is clear that you can also calculate the minimum of $F(x)$ for a literal $a$ belonging to the interval above mentioned.

enter image description here

Piquito
  • 29,594
  • 4
    “You can verify that... the minimum of the function $F(x)$ is very near of $2$ but always greater that $2$.” – I miss a rigorous proof here. – Martin R Oct 16 '21 at 01:00
  • @Martin R.- I agree with you (I gave this for anyone who wanted to verify it). Read my comment to River Li's anwer, please. It is an answer more concise to the question then the mine here. – Piquito Oct 16 '21 at 16:54