4

I have already met this type of problem with $e^\pi$ $>$ $\pi^e$. I also wondered if someone would please explain when $a^b$ $>$ $b^a$ if $b>a$ please. It is clear that the "powers dominate over bases rule" is not a rigorous rule at all.

Thank you.

  • I am guessing $b>a>0$. Take the logarithm with base $b$ of both sides. You might also want to keep in mind that logarithm function grows slower than linear function. Also keep in mind that $b=a+\epsilon$ might be useful. – Predrag Punosevac Nov 08 '16 at 22:25
  • 1
    Please use parentheses to indicate that you mean the usual interpretation $$\sqrt{2}^{\left(\sqrt{3}^\sqrt{3} \right)}$$ rather than $$\left( \sqrt{2}^\sqrt{3}\right)^\sqrt{3}$$ There is a difference, and if the latter interpretation is used, then of course the right hand side is greater, not the left. – Mark Fischler Nov 08 '16 at 22:30
  • Apologies, thank you for clarifying that. It is now edited :) – John Smith Nov 08 '16 at 22:35
  • There ought also be a ground rule that it is not acceptable to just say that "this one is 2.45326 and that one is 2.45164 so this one is greater." – Mark Fischler Nov 08 '16 at 23:18
  • This would be non-calculator, yes ;) – John Smith Nov 08 '16 at 23:23
  • @JohnSmith: Your title is not of the form $a^b\gt b^a$. So, your title is different from the body. So, do you mean that you are interested in two distinct type of inequalities? – mathlove Nov 09 '16 at 14:54
  • You used the tag "algebra-precalculus", but both solutions currently appearing use calculus methods. Do you really want a precalculus solution, or are we allowed to the use the first derivative to show certain functions are increasing? – Dave L. Renfro Nov 09 '16 at 20:37

2 Answers2

3

I can answer your $a^b$ versus $b^a$ question. Take $a,b$ positive:

$a^b > b^a$ if $\frac{a^b}{b^a} >1$.

We can work with that by turning all exponentials into base $e$: $$ \frac{a^b}{b^a} = \frac{e^{b\ln a}}{e^{a\ln b}}=e^{b\ln a - a\ln b}>1\\ \implies b\ln a - a \ln b >0 \implies b\ln a > a \ln b \\ \implies \frac{\ln a}{a} > \frac{\ln b}{b} $$ Now the function $f(x) = \frac{\ln x}{x}$ is an increasing function for $x<e$, and a decreasing function for $x>e$. $f(x) = 0$ at $x=1$ and $\lim_{x \to \infty}f(x) = 0$. That said, we have:

  • If $b>a>e$ then $\frac{\ln a}{a} > \frac{\ln b}{b}$ and $a^b > b^a$. For example, $3^4 > 4^3$.

  • If $a<b<e$ then $\frac{\ln a}{a} < \frac{\ln b}{b}$ and $a^b < b^a$. For example, $2^{2.5} < (2.5)^2$.

  • If $a<1<b$ then $\frac{\ln a}{a} < \frac{\ln b}{b}$ and $a^b < b^a$. For example, $\left(\frac12\right)^2 < 2^\frac12$.

  • If $1 < a < e < b$ we cannot say much about the relative values of $a^b$ and $b^a$.


Your title question is much more interesting; I have not found a good reason why $$ \sqrt{2}^\left( \sqrt{3}^\sqrt{3}\right) > \sqrt{3}^\left( \sqrt{2}^\sqrt{2}\right) $$ although I have squeezed that into $g(3) > g(2)$ with $$g(x) = \frac12\sqrt{x}\ln x - \ln (\ln x)$$ The difficulty is that while $g(x)$ forms a concave upward curve with a minimum at about 2.4, the values $2$ and $3$ are on opposite sides of the minimum, kind of analogous to the $1 < a < e < b$ case in the above answer. So I don;t see why $g(x)$ is greater at $2$ than at $3$.

I hope this question is left open and active until somebody comes up with a good reason or a reference in the literature.

Mark Fischler
  • 41,743
  • Thank you so much! This was all very interesting. I know the second question was more basic, but I really appreciate the explanation. And I do hope someone finishes the first question. It was apparently an old Cambridge interview question but seems quite a bit harder than any others I've tried (not least because you haven't solved it yet). But again, thank you so much for your response :) – John Smith Nov 09 '16 at 18:50
3

EDIT: As pointed out by Mark Fischler in the comments, I made a miscalculation and the answer below is not correct.

This is an answer to the title question of why $\sqrt{2}^{({\sqrt{3}}^{\sqrt{3}})}>\sqrt{3}^{({\sqrt{2}}^{\sqrt{2}})}$. Consider the function

$$f(x,y)=\ln\left(x^{\displaystyle\left(y^y\right)}\right)=y^y\ln(x)$$

Let $p=(\sqrt{2},\sqrt{3})$ and $q=(\sqrt{3},\sqrt{2})$. We wish to determine whether or not $f(p)>f(q)$. To that end, consider the line segment $L$ joining them, and $g:[0,1]\longrightarrow \mathbb{R}$ given by

$$g(t)=f(t\cdot p + (1-t)\cdot q)$$

that is, consider the function $f$ along $L$. Notice that $g(0)=f(q)$ and $g(1)=f(p)$. We will show that $g'(t)$ is always positive, which implies that $g$ is strictly increasing, proving the result.

Indeed, let

\begin{align} a(t)=t\sqrt{2}+(1-t)\sqrt{3}\\ b(t)=t\sqrt{3}+(1-t)\sqrt{2} \end{align}

so that $a'(t)=-b'(t)=\sqrt{2}-\sqrt{3}<0$ for all $t \in [0,1]$. Then we have that

$$g(t)={b(t)}^{b(t)}\cdot\ln(a(t))$$

and hence

$$g'(t)=\frac{b'}{a(t)}\cdot\left({b(t)}^{b(t)}\cdot\Big(a'+a(t)\cdot\ln(a(t))\Big)\cdot (\ln(b(t)+1)\right)$$

Since $\frac{b'}{a(t)}>0$ for all $t \in [0,1]$, we need only prove the other term is always positive. Now, ${b(t)}^{b(t)}$ and $(\ln(b(t)+1)$ are also always positive, so it suffices to prove that

$$h(t)=a'+a(t)\cdot\ln(a(t))$$

is always positive. The derivative of $h$ is

$$h'(t)=a'\cdot(\ln(a(t))+1)$$

which is always negative (because $a'<0$), so $h$ is strictly decreasing. On the other hand, we have that:

\begin{align} h(1)&=(\sqrt{2}-\sqrt{3})+\sqrt{2}\cdot(\ln(\sqrt{2})+1)\\ &=\underbrace{(2\sqrt{2}-\sqrt{3})}_{>0}+\underbrace{\sqrt{2}\cdot\ln(\sqrt{2})}_{>0} \end{align}

So that $h(t) \geq h(1)>0$ for all $t$, which concludes the proof.

Fimpellizzeri
  • 23,126
  • You're welcome! – Fimpellizzeri Nov 09 '16 at 20:52
  • Actually, you have the wrong expression for $g'(t)$, which is $$\frac1{a(t)}b(t)^{b(t)}\left[a'(t) + b'(t)a(t)\ln(a(t)) (\ln b(t)+1)\right]$$. Thos makes the next step a bit harder, I think. – Mark Fischler Dec 14 '16 at 20:34
  • Particularly since this $g'(t)$ is negative when $t>0.55$. – Mark Fischler Dec 14 '16 at 20:46
  • In fact, your proof applies exactly in the same way to show that $$\sqrt{2}^{\left(\sqrt{2.8}^{\sqrt{2.8}}\right)} > \sqrt{2.8}^{\left( \sqrt{2}^{\sqrt{2}}\right)}$$ But that turns out to be false. – Mark Fischler Dec 14 '16 at 21:05
  • You are right that I miscalculated the derivative, and this does make things harder. I don't see the connection to your last comment though. – Fimpellizzeri Dec 14 '16 at 21:59
  • My point was that the same proof, using the same erroneous derivative, would apply if you substituted 2.8 for 3 in your proof. And that then it would be proving something which is untrue. – Mark Fischler Dec 15 '16 at 20:03