27
$\begingroup$

show that $$\sqrt{7}^{\sqrt{8}}>\sqrt{8}^{\sqrt{7}}$$

and I found $$LHs-RHS=0.017\cdots$$

I have post this interesting problem Prove $\left(\frac{2}{5}\right)^{\frac{2}{5}}<\ln{2}$

can someone suggest any other nice method? Thank you everyone.

$\endgroup$
  • 3
    $\begingroup$ It is true that $7^8>8^7$, since $(1+\frac{1}{7})^7<e<7$. Then we have, $\sqrt{7}^8>\sqrt{8}^7$. Does it imply that $\sqrt{7}^\sqrt{8}>\sqrt{8}^\sqrt{7}$. $\endgroup$ – mtm Jun 28 '13 at 9:26
  • $\begingroup$ No,How can have $\sqrt{7}^8>\sqrt{8}^7$? $\endgroup$ – math110 Jun 28 '13 at 9:31
  • $\begingroup$ $\sqrt{7}^{16}>\sqrt{8}^{14}$. So if we take $\frac{1}{2}$ power of both sides, we have $\sqrt{7}^{8}>\sqrt{8}^{7}$. Maybe i am wrong. $\endgroup$ – mtm Jun 28 '13 at 9:37
  • $\begingroup$ $\sqrt{7}^{8}>\sqrt{8}^{7}\Longrightarrow \sqrt{7}^{\sqrt{8}}>\sqrt{8}^{\sqrt{7}}$? I think is wrong. $\endgroup$ – math110 Jun 28 '13 at 10:58
  • $\begingroup$ @mtm Monotonicity argument like yours will not work here, no matter you take $(\ln t)/\sqrt{t}$ or $(\ln t)/t$. $\endgroup$ – Shuhao Cao Jun 28 '13 at 15:21
29
$\begingroup$

Note that $\sqrt{7}^{\sqrt{8}}\doteq15.673$ and $\sqrt{8}^{\sqrt{7}}\doteq15.656$, so these two are pretty close. Furthermore $\sqrt{7}<e<\sqrt{8}$, and the function $f(x):={\log x\over x}$ has a local maximum at $x:=e$. This excludes the use of monotonicity arguments. The following proof uses integer arithmetic instead.

To begin with we need a rational approximation to $\sqrt{7\over8}$ that is slightly larger than $\sqrt{7\over8}$. Using Mathematica (or continued fractions) one finds that $$7\cdot 31^2=6727<6728=8\cdot 29^2\ ,$$ which implies $$\sqrt{7}\cdot31<29\cdot\sqrt{8}\ .$$ Furthermore one computes $$8^{29}=15\>47425\>04910\>67253\>43623\>90528<15\>77753\>82034\>84580\>66150\>42743=7^{31}\ .$$ It follows that $$\bigl(8^{\sqrt{7}}\bigr)^{31}<\bigl(8^{29}\bigr)^{\sqrt{8}}<\bigl(7^{31}\bigr)^{\sqrt{8}}\ .$$ Now take the $62^{\rm th}$ root on both sides.

$\endgroup$
  • 1
    $\begingroup$ Very instructive! +1 $\endgroup$ – Markus Scheuer Nov 13 '14 at 7:58
11
$\begingroup$

The challenge here is to demonstrate the inequality in a way that can easily be checked by hand. Taking logs, the inequality to be proved is

$$\sqrt7\ln8 \lt \sqrt8\ln7$$

which can be written as

$$3\sqrt{1-{1\over8}}\ln2 \lt \ln(8-1) = 3\ln2+\ln(1-{1\over8})$$

so the inequality to prove is

$$-\ln\left(1-{1\over8}\right)\lt 3\left(1-\sqrt{1-{1\over8}}\right)\ln2$$

We'll use the Taylor series

$$-\ln(1-x)=x+{1\over2}x^2+{1\over3}x^3+{1\over4}x^4+\cdots$$

to obtain

$$\begin{align} -\ln\left(1-{1\over8}\right) &\lt {1\over8}+{1\over2}\left({1\over8}\right)^2+{1\over3}\left({1\over8}\right)^3+{1\over3}\left({1\over8}\right)^4\\ &={1\over8}+{1\over2}\left({1\over8}\right)^2+{1\over3}\left({1\over8}\right)^3{1\over1-{1\over8}}\\ &={1\over8}+{1\over128}+{1\over21}{1\over64}\\ &\lt{1\over8}+{1\over128}+{1\over16}{1\over64}\\ &={128+8+1\over2^{10}}\\ &={137\over2^{10}} \end{align}$$

We'll use the fact that all the coefficients in the Taylor series

$$1-\sqrt{1-x}={1\over2}x+{1\over8}x^2+{1\over16}x^3+\ldots$$

are positive to get

$$1-\sqrt{1-{1\over8}}\gt {1\over2}{1\over8}+{1\over8}\left({1\over8}\right)^2={33\over2^9}$$

Finally, we use the easily verified identity

$$\ln2=3\ln\left(1+{1\over4}\right)+\ln\left(1+{3\over125}\right)$$

and the alternating Taylor series

$$\ln(1+x)=x-{1\over2}x^2+{1\over3}x^3-\cdots$$

to get

$$\begin{align} \ln2&\gt 3\left({1\over4}-{1\over2}\left({1\over4}\right)^2+{1\over3}\left({1\over4}\right)^3-{1\over4}\left({1\over4}\right)^4\right)+{3\over125}\\ &={21\over32}+{1\over64}-{3\over1024}+{3\over125}\\ &\gt{21\over32}+{1\over64}-{3\over1024}+{3\over128}\\ &={21\cdot32+16-3+24\over1024}={709\over1024} \end{align}$$

hence

$$3\ln2\gt{2127\over1024}\gt{2126\over1024}={1063\over512}$$

so that

$$3\left(1-\sqrt{1-{1\over8}}\right)\ln2\gt{33\over2^9}{1063\over2^9}={33\cdot1063\over2^{18}}$$

Consequently, the inequality to verify is

$${137\over2^{10}}\lt{33\cdot1063\over2^{18}}$$

which is to say,

$$256\cdot137\lt33\cdot1063$$

Eschewing calculators, note that $1063=1024+32+8-1$, so

$$33\cdot1063=(2^5+1)(2^{10}+2^5+2^3-1)=2^{15}+2^{11}+2^8+2^3-1$$

whereas

$$256\cdot137=2^8(2^7+2^3+1)=2^{15}+2^{11}+2^8$$

so the desired inequality holds.

$\endgroup$
  • 1
    $\begingroup$ Reading along, I kept expecting the computation to end, but it never did! A rather intimidating verification. $\endgroup$ – Ryan Reich Aug 8 '13 at 5:29
  • 3
    $\begingroup$ @RyanReich, but it did end! ;-) $\endgroup$ – Barry Cipra Aug 8 '13 at 6:00
4
$\begingroup$

I just want to give you a bigger picture on this question. The fact you point out corresponds to $f(2)>0.$

$f(\text{x$\_$})\text{:=}\left(t=\lfloor \exp (x)\rfloor ;a=t^{1/x};b=(t+1)^{1/x};a^b-b^a\right);$

$\text{Plot}[f(x),\{x,1,3\}]$

enter image description here

$\endgroup$

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.