35

My math teacher left two questions last week, prove (1) $6^9>10^7$ and (2) $7^{71}>75^{32}.$

I did the first question: \begin{align}\frac{6^9}{10^7}&=\frac{4}{5}\times\frac{27^3}{25^3}\\&=0.8\times1.08^3\\&>0.8\times(1+3\times0.08+3\times0.08^2)\\&>0.8\times(1+3\times0.086)\\&>0.8\times1.25=1.\end{align}

But I can't work out the second, I calculated it out on my computer, $\frac{7^{71}}{75^{32}}=1.000000949\cdots$

lsr314
  • 15,806
  • Maybe use $7^4=2401>2400$? – Gerry Myerson Mar 12 '21 at 06:52
  • 14
    There seems to be too little room for "elementary" approximation without logarithm. – Tesla Daybreak Mar 12 '21 at 07:54
  • For the first one we have $ 6^9=6(1300^2)(1-4/1300)^2=$ $(1.014\times 10^7)(1-4/1300)^2>$ $>(1.014\times 10^7)(1-8/1300)>10^7.$ – DanielWainfleet Mar 12 '21 at 09:52
  • 1
    @GerryMyerson . The computer calculation shows the LHS larger by less than 1 part per million so approximating 2401 by 2400 will give too small an estimate for the LHS. – DanielWainfleet Mar 12 '21 at 10:02
  • Doing the arithmetic by hand using exponentiation by squaring is laborious, but not preposterously so. Of course it's an error-prone and inelegant solution, and I doubt anybody will actually spend a few hours (?) doing it. – Joshua P. Swanson Mar 12 '21 at 10:54
  • 1
    @lonestudent No, even a 11 year old pupil (in gifted schools) can has this kind of exercises and normally, these questions can be indeed solved in 5-7 lines. No series, large numbers or logarithm are allowed here. – NN2 Mar 12 '21 at 12:00
  • @NN2 Gifted schools in which country? – Toby Mak Mar 12 '21 at 12:06
  • 2
    @TobyMak At least, the countries in Asia. – NN2 Mar 12 '21 at 12:08
  • 2
    @NN2 This is not true in my experience: in Hong Kong, the number theory for junior secondary is comparable to the problems on AoPS such as: what is the sum of factors of $899899$? You speak of Asia as if it were a single country, when in fact, the mathematics instruction varies widely between different places. – Toby Mak Mar 12 '21 at 12:40
  • $\frac{7^{71}}{75^{32}}$ can be reduced to $\Big(1-\frac{49}{50}\Big)^{32} \Big(1-\frac{1}{3}\Big)^{32} 7^{7} $. Approximation for first Binomial sum is $e^{\frac{-32}{50}}$ But for second Binomial sum such approximation requires very high precision $\Bigg(e^{-32\bigg(1-\Big(\frac{2}{3}\Big)^{\frac{4294967296}{1853020188851841}}\bigg)}\Bigg)$ otherwise on multiplying approximations we get high percentage error. So Using approximations this result can't be proved so easily. – user6262 Mar 12 '21 at 13:17
  • 5
    It is very easy to generate these near-equalities. For example, I just tried a log-$13$ base and found $13^7>89^4$ where $13^7/89^4=1.000100028$. – ə̷̶̸͇̘̜́̍͗̂̄︣͟ Mar 12 '21 at 14:54
  • 11
    @Isr314 Now, if your teacher asked you this question, please share with us the correct solution you received from your teacher. Presumably, during this time your teacher has told you the correct and effective solution to the problem... – lone student Mar 15 '21 at 14:38
  • 5
    Look at the other questions and answers this user has posted. Clearly, this user is well beyond mere secondary school algebra or even undergraduate number theory. Consequently I find it implausible that a teacher would ask such a question and expect an answer using only pre-calculus methods. – heropup Mar 18 '21 at 07:58
  • 3
    The inequality is quite tight. To see how tight, let's do the following: $$7^{71}>75^{32} \iff 49\cdot 7^{\frac{7}{32}}>75\iff 7^{\frac{7}{32}}>1+\frac{26}{49}\iff\ e^{\frac{7}{32}\log{7}}-1>\frac{26}{49}$$

    Given $$e^x=\sum\limits_{k=0}\frac{x^k}{k!}$$

    for $x=\frac{7}{32}\log{7}$, we need at least $7$ iterations to reach the inequality.

    – rtybase Mar 20 '21 at 19:33
  • 1
    I know long answers usually don't get read, but I posted a solution that might interest you. There is basically no calculatipn involved beside elementary ones. It seems a long answer but it's because I wanted to motivate. If you guys prefer, I can put the short version that seems completely out of the blue :) – Andrea Marino Mar 21 '21 at 19:05
  • Related $7^{31}>8^{29}$, there wasn't a quick method to prove that inequality so I don't expect differently to this one. – ə̷̶̸͇̘̜́̍͗̂̄︣͟ Mar 23 '21 at 11:12
  • Commenting regarding my decision for the bounty: This is a hard inequality with a tight gap ($<1$ part in $10^6$) and large numbers $(\approx 10^{60})$, where "nice" (@AndreaMarino ) and "elementary" (@Yuri ) approaches might not be more desirable than direct "floating point" calculation (@Joshua ). Andrea's approach seemed promising and applicable to other situations, but has an unfortunate mistake in the numerics. Yuri's 1st answer used a lot of "calculator power", and the 2nd improved on this to cut down on calculations. – Benjamin Wang Mar 26 '21 at 22:18
  • All these (even Joshua's) suffer from the fact that "if you are too sloppy, you have to tweak your approximations, and redo most of your previous calculations". Yuri's improved answer, Andrea's answer, and Joshua's answer all contain calculations which are, in principle, doable by hand. It's very hard for me to decide between these answers. Since Yuri has had two answers with $11$ and $9$ upvotes (to date), I don't feel too bad for awarding the $+200$ rep bounty to Andrea. Imperfect (dare I say "sloppy") as it is, it somewhat resembles a "recipe" one might be able to apply in similar problems. – Benjamin Wang Mar 26 '21 at 22:24
  • 1
    Thank you! I'll redo the calculations when I can to adjust the small error. Note that the section "out of the blue" uses a different inequality in that passage and it does not rely on the previous part, so it is actually possible that the error you spotted has been adjusted. In writing the new part, indeed, I corrected some things which were sloppy or slightly wrong. Anyway I feel like I should adjust what is wrong in the first part. Thank you for the bounty :) – Andrea Marino Mar 27 '21 at 09:44
  • @BenjaminWang New proof "Evident approach" is ready. :) – Yuri Negometyanov Mar 28 '21 at 23:32
  • 1
    Just for fun: $7^{71}=1.004.525.211.269.079.039.999.221.534.496.697.502.180.541.686.174.722.466.474.743>75^{32}= 1.004.524.257.206.332.858.195.774.182.519.244.277.500.547.468.662.261.962.890.625$ – calculatormathematical Feb 15 '24 at 15:16

7 Answers7

12

$\color{green}{\textbf{Improved version.}}$

There is an alternative approach.

As it was shown in the my first answer, $$R=\dfrac{75^{32}}{7^{71}}=\dfrac{5^2}{3^3}\,\dfrac{5^6}{3^7\cdot7}\,\left(\dfrac{3^3\cdot5^4}{7^5}\right)^{14} = \dfrac{625^2}{7\cdot 243^2}\,\left(\dfrac{16875}{16807}\right)^{14}.\tag1$$

Identity $(1)$ can be presented in the form of $$R= \dfrac{625^2}{7\cdot 243^2}\,\left(\dfrac{225}{224}\right)^{14} \,\left(\dfrac{2400}{2401}\right)^{14} =\dfrac{625^2}{7\cdot 243^2}\,\left(\dfrac{225}{224}\right)^{14} \,\left(1-\dfrac{1}{2401}\right)^{14}\tag{1.1}$$

$$<\dfrac{625^2}{7\cdot 243^2}\,\left(\dfrac{225}{224}\right)^{14} \,\left(1-\dfrac{14}{2401}+\dfrac{91}{2401^2}\right),\tag{1.2}$$

$$R < \dfrac{35557}{37845}\,\left(\dfrac{225}{224}\right)^{14}.\tag2$$

Since $$\dfrac{225^2}{224^2} = 1+\dfrac t7,\quad\text{where}\quad t=\dfrac{449}{7168},\tag3$$ one can get $$\left(1+\dfrac t7\right)^7 = 1 + \dfrac 77t + \dfrac{21}{7^2}t^2+\dfrac{35}{7^3}t^3 +\dfrac{35}{7^4}t^4+\dfrac{21}{7^5}t^5+\dfrac7{7^6}t^6+\dfrac1{7^7}t^7$$ $$< 1+t+\dfrac37t^2\left(1+\dfrac5{21}t+\dfrac {5^2}{21^2}t^2+\dots\right) =1+t+\dfrac{9t^2}{21-5t} = \dfrac{21+16t+4t^2}{21-5t},$$ $$\dfrac{35557}{37845}\left(1+\dfrac t7\right)^7-1 < \dfrac{35557}{37845}\cdot\dfrac{21+16t+4t^2}{21-5t}-1 $$ $$=\dfrac{142228 t^2 + 758137 t - 48048}{37845(21-5t)},$$ wherein $$(142228 t^2 + 758137 t - 48048)\bigg|_{\large\frac{449}{7168}} = -\dfrac{7\,828\,635}{12\,845\,056} < 0.$$ Therefore, $$\dfrac{35557}{37845}\left(1+\dfrac t7\right)^7 < 1$$ and $\;\color{brown}{\mathbf{R<1}}.\;$

Proved!

11

I will give a solution that involve only calculations with 2 digit numbers; all the other contributions will be estimated. I will call my approach the sloppy continued fractions method. Unluckily, if you want to cut down computations you have to be smarter, and in particular lengthy. Sorry for that, but you will enjoy! After the theoretical framework is set up, the problem is easy to understand, but if you prefer you can skip to the "out of the blue" section at the bottom.

Theory

Firstly, notice that the problem can be reformulated as $\log_7(75) < \frac{71}{32}$. The instance of finding good rational approximants to irrational numbers is a classical one, and it is addressed by the continued fractions method. In case of logarithms and inequalities, it has the following simpler form. Say we want to understand whether $\log_a(b) < \frac{p}{q}$. Then:

  1. Take the maximum integer $n_0$ less than $\log_a(b)$, i.e. solve $a^{n_0} < b < a^{n_0+1} $, and let $m_0$ be the maximum integer less than $p/q$.
  2. If $m_0 \ge n_0 +1$ the inequality is true, because $$\frac{p}{q} \ge m_0 \ge n_0 +1 \ge \log_a(b) $$ Else, if $n_0 \ge m_0 +1$ the inequality is false, because $$ \frac{p}{q} \le m_0 +1 \le n_0 \le \log_a(b) $$ Otherwise, go to step 3 supposing $n_0 = m_0$.
  3. Since $\log_a(b) - n_0 = \log_a(b/a^{n_0})$ is less than one, we can write $$ \log_a(b) = n_0+ \frac{1}{\log_{a_1}(b_1) } $$ where $a_1 = b/a^{n_0}, b_1 = a$. Analogously we can write $$ \frac{p}{q} = n_0 + \frac{1}{ \frac{p_1}{q_1} } $$ where $p_1 = q, q_1 = p-n_0 q$.
  4. Now we reformulate the inequality and we get $$ n_0 + \frac{1}{\log_{a_1}(b_1) } < n_0 + \frac{p-n_0 q}{q} $$ $$ \log_{a_1}(b_1) > \frac{p_1}{q_1} $$ And we can repeat from step 1 with a similar problem (but sign reversed).

There is a little problem: while the fraction gets simpler at step (5), the numbers involved in logarithms becomes harder. Let me state a little lemma.

Logarithm approximation. Let $\alpha, \beta \in \mathbb{N}$ and $k,j \in \mathbb{Z}$ small with respect to $\alpha, \beta$. Then if $x = k/\alpha, y = j/\beta$:

$$ 1+\frac{1}{\ln(\alpha/\beta)} (x -y -(x^2+y^2)) < \log_{\frac{\alpha}{\beta}} (\frac{\alpha+k}{\beta+j} ) < 1+\frac{1}{\ln(\alpha/\beta)} (x-y +(x^2+y^2))$$

Proof. This is a simple consequence of $ x-x^2 \le \ln(1+x) \le x+x^2$. Also, depending on the sign of $k,j$ and the inequality that must be used, sometimes the quadratic term is not necessary (indeed $\log(1+x) \le x$ for positive $x$ and $\log(1+x) \ge x$ for negative $x$).

Now let me restate the continued fraction method with a sloppiness parameter. We want to solve inequalities of the form $\log_a (b) < \frac{p}{q} \cdot \alpha$, where $\alpha \simeq 1 $ is a rational number (the approximation parameter). The bad thing is that it can stop at some time because of our sloppiness, but if it doesn't stop it works with few digits!! Also, don't worry if you can't follow the exact formulas, because they will get explicit thereafter.

  1. Take the maximum integer $n_0$ less than $\log_a(b)$, i.e. solve $a^{n_0} < b < a^{n_0+1} $, and let $m_0$ be the maximum integer less than $p/q$.
  2. If If $\alpha m_0 \ge n_0 +1$ the inequality is true, because $$\alpha\frac{p}{q} \ge \alpha m_0 \ge n_0 +1 \ge \log_a(b) $$ Else, if $n_0 \ge \alpha (m_0 +1)$ the inequality is false, because $$ \alpha \frac{p}{q} \le \alpha(m_0 +1) \le n_0 \le \log_a(b) $$ Else, if $n_0 = m_0$ go to step 3. In any other case, return "YOU HAVE BEEN TOO SLOPPY! TRY AGAIN!".
  3. Since $\log_a(b) - n_0 = \log_a(b/a^{n_0})$ is less than one, we can write $$ \log_a(b) = n_0+ \frac{1}{\log_{a_1}(b_1) } $$ where $a_1 = b/a^{n_0}, b_1 = a$. Analogously we can write $$ \alpha \frac{p}{q} - n_0 = \alpha(n_0 + \frac{q_1}{p_1} ) - n_0 = (\alpha -1)n_0 + \alpha \frac{q_1}{p_1} = \frac{q_1}{p_1} ( \alpha + \frac{p_1}{q_1} n_0 (\alpha-1) )= \frac{1}{ \alpha_1 \frac{p_1}{q_1} } $$ where $p_1 = q, q_1 = p-n_0 q, \alpha_1 = \left ( \alpha + \frac{p_1}{q_1} n_0 (\alpha-1) \right )^{-1} $.
  4. Now we reformulate the inequality and we get $$ n_0 + \frac{1}{\log_{a_1}(b_1) } < n_0 + \frac{1}{\alpha_1 \frac{p_1}{q_1} } $$ $$ \log_{a_1}(b_1) > \alpha_1 \frac{p_1}{q_1} $$ If we want to, we can approximate a bit the logarithm on the left using the logarithm approximation and move this contribution in $\alpha_1$. Then we can repeat from step 1 with a similar problem (but reversed inequality).

Calculations

Let's start! At the beginning $a=7, b= 75, p=71, q=32, \alpha=1$. Also, let me notice that the continued fraction of $71/32$ is $[2,4,1,1,3]$, that is: $$ \frac{71}{32} = 2+\frac{1}{4+\frac{1}{1+ \frac{1}{1+\frac{1}{3}}}} $$

Since we suspect this approximations is really close, probably the continued fraction of the logarithm will be the same for a lot of time, so that we have a guess of what the "$n_0$" should be.

STEP 1

1.1 Note that $7^2 < 75 < 7^3$, so that $n_0 = 2= m_0$ and we go to point 3.

1.3 We have $a_1 = 75/49, b_1 = 7, p_1 = 32, q_1 = 71 - 64 = 7$. The inequality is now $$ \log_{75/49} (7) > \frac{32}{7}$$ here it is a trick we will use repeatedly: $$ \log_{75/49}(7) = \color{red}{\log_{75/49}(75/50)}\log_{75/50}(7) > \color{red}{\left(1- \frac{1}{50} \right)} \log_{3/2}(7) = \frac{49}{50} \log_{3/2}(7) $$

1.5 It is enough to verify that something smaller than LHS is greater than RHS, and we get the sloppy inequality $$ \log_{3/2}(7) > \frac{50}{49} \frac{32}{7} $$

STEP 2

2.1 The suggestion from the continued fraction of 32/7 makes us try $n_1 = 4$, and indeed $$ 3^4 = 81 < 112 = 7 \cdot 16 = 7 \cdot 2^4, \ \ \ 3^5 = 343 > 7 \cdot 32 $$ 2.3 Formulas yield $a_2 = 112/81 , b_2 = 3/2, p_2 = 7, q_2 = 4$ and $$\alpha_2 = \left ( \frac{50}{49} + \frac{7}{4} \cdot 4 \cdot \frac{1}{49}\right)^{-1} = \left ( \frac{57}{49} \right)^{-1} = \frac{49}{57} $$ The inequality is now $$ \log_{112/81} (3/2) < \frac{7}{4} \frac{49}{57}$$ and the log trick yields $$ \log_{112/81}(3/2) = \log_{112/81}(110/80) \log_{11/8}(3/2) < \left ( 1-\frac{2}{112} + \frac{1}{81} \right ) \log_{11/8}(3/2) < \log_{11/8}(3/2)$$

2.5 The sloppy inequality remains $$ \log_{11/8} (3/2) < \frac{7}{4} \frac{49}{57} $$

STEP 3

3.1 Under the usual suggestion we try $n_2 = 1$ and it works: $$ \frac{11}{8} = 1+ \frac{3}{11} < 1+ \frac{1}{2} = \frac{3}{2}, \frac{11^2}{8^2} = \left( 1+ \frac{3}{11} \right)^2 > 1+ \frac{6}{11} > \frac{3}{2} $$ 3.3 We get $p_3 =4, q_3 = 3, a_3 = 24/22 = 12/11, b_3 = 11/8$ and

$$ \alpha_3 = \left ( \frac{49}{57} -\frac{4}{3} \frac{9}{57} \right )^{-1} = \frac{57}{37} $$

3.5 The fraction $12/11$ is nice, so we don't do the trick at this step. The sloppy inequality is just

$$ \log_{12/11}( 11/8) > \frac{4}{3} \frac{57}{37} $$

STEP 4.

4.1 Here it is the surprise: using the rule for the cube of a sum we have that

$$ \left ( 1 + \frac{1}{11} \right) ^3 = 1 + \frac{3}{11} + \frac{3}{11^2} + \frac{1}{11^3} < 1 + \frac{3}{11} + \frac{10}{121} + \frac{11}{11^3} = \frac{15}{11} < \frac{11}{8}$$

because $15 \cdot 8 = 120 < 121 = 11^2$. This translates into $\log_{12/11}(11/8) \ge 3$. On the other side

$$ \frac{4}{3} \frac{57}{37} < \frac{4}{3} \frac{60}{36} = \frac{20}{9} < 3 \le \log_{12/11}(11/8) $$

And we are done. Victory! Yey!

Out of the blue

Let's start from the observation that

$$ (\#1) \ \ \ \ \ \ \left ( \frac{12}{11} \right ) ^3 = \left (1+ \frac{1}{11} \right ) ^3 = $$ $$ =1+ \frac{3}{11} +\frac{3}{11^2} +\frac{1}{11^3} \le 1+ \frac{3}{11} + \frac{10}{11^2} + \frac{11}{11^3} $$ $$ = \frac{15}{11} < \frac{11}{8} $$

So that $\log_{12/11}(11/8) > 3$. Also, note that

$$ (\#2) \ \ \ \ \ \ 3 > \frac{20}{9} = \frac{60}{36} \frac{4}{3} > \frac{57}{37} \frac{4}{3} $$ If we combine (1) and (2) we get $ \log_{12/11}(11/8) > \frac{57}{37} \frac{4}{3}$. Taking reciprocal (that swaps base and argument in the logarithm) and adding 1 we get

$$ (\#3) \ \ \ \ \ \ \log_{11/8} \left ( \frac{12}{11} \frac{11}{8} \right ) = \log_{11/8}(3/2) < $$ $$ <1+\frac{37}{57} \frac{3}{4} < \frac{37}{57}+ \frac{21}{57} + \frac{37}{57} \frac{3}{4} = $$ $$ = \frac{37}{57} \cdot 1 + \frac{12}{57} \frac{7}{4} + \frac{37}{57} \frac{3}{4} = $$ $$ =\frac{37}{57} \frac{7}{4} + \frac{12}{57} \frac{7}{4} = \frac{49}{57} \frac{7}{4}$$

Using the inequality

$$ (\#4) \ \ \ \ \ \ \log_{112/81} (11/8) = 1+ \log_{112/81}\left (\frac{110 \cdot 81}{112 \cdot 80} \right ) = $$ $$ = 1+ \log_{112/81} \left (1- \frac{2}{122} \right ) - \log_{112/81}\left ( 1-\frac{1}{81} \right ) < $$ $$ < 1+ \frac{1}{\ln(112/81) } \left ( -\frac{2}{122} + \frac{1}{81} \right ) < 1 $$

We can multiply inequality (3) by the (4), and using $\log_x y \log_y z = \log_x z$:

$$ (\#5) \ \ \ \ \ \ \log_{11/8}(3/2) < \frac{49}{57} \frac{7}{4} \Rightarrow \log_{112/81}(3/2) = \log_{112/81}(11/8) \log_{11/8}(3/2) < \frac{49}{57} \frac{7}{4} $$

Taking reciprocals and adding 4 we get

$$ (\#6) \ \ \ \ \ \ \log_{3/2} \left ( \frac{112}{81} \frac{81}{16} \right) = \log_{3/2}(7) > $$ $$ > 4+ \frac{57}{49} \frac{4}{7} = \frac{57}{49} \cdot 4 - \frac{8}{49} \cdot 4 + \frac{57}{49} \frac{4}{7} = $$ $$ = \frac{57}{49} \frac{32}{7} - \frac{32}{7} \frac{7}{49} = \frac{50}{49} \frac{32}{7} $$

Consider the following inequality $$ (\#7) \ \ \ \ \ \ \log_{75/49}(3/2) = 1+ \log_{75/49} \left ( \frac{49 \cdot 75}{75 \cdot 50} \right ) = $$ $$ = 1+ \log_{75/49} \left ( 1- \frac{1}{50} \right ) > 1+ \frac{1}{\ln(75/49) } (- \frac{1}{50} + \frac{1}{50^2}) > $$ $$ > 1- 2 \cdot \frac{1}{51} > 1- \frac{1}{50} = \frac{49}{50} $$ To show (7) we used, between the second and the third row, that $$ (75/49)^2 < (75/48)^2 = (25/16)^2 < e \Rightarrow \frac{1}{\ln(75/49) } = \log_{75/49}(e) > 2 $$ Indeed, $625 < 256 \cdot e$. We are almost finished: combining the inequalities

$$\log_{3/2}(7) > \frac{50}{49} \frac{32}{7}, \ \ \color{red}{\log_{75/49}(3/2) > \frac{49}{50}} $$

we get $ \log_{75/49}(7) > \frac{32}{7}$. Taking inverses and adding 2 we get

$$ 2+ \log_7(75/49) = \log_7(75) < 2+ \frac{7}{32} = \frac{71}{32} $$

Putting this to the exponent of $7$ we get $$ 75 < 7^{71/32} \Rightarrow 75^{32} < 7^{71} $$

as desired!

zhuli
  • 2,561
  • Thanks for the detailed answer which seems to not use more than elementary computations. If it is not too much work, would you please write down a chain of (forward) implications that lead to the final answer? It will help me read it tomorrow. – Benjamin Wang Mar 21 '21 at 22:53
  • That seems like a good idea to implement a short version in the first lines! I'll do it when I can :) – Andrea Marino Mar 22 '21 at 07:23
  • Added an out of the blue section. I don't know if it really helps clarifying, but for sure it's more convincing. I noticed that I got an estimate or two wrong in the theoretical part; I mean, they were true but with a wrong intermediate passage. This is because of the quadratic term in the estimate $\ln(1-x) > -x+x^2$ and because of $\log_a(1-x) = \ln(1-x) / \ln(a)$ – Andrea Marino Mar 22 '21 at 09:35
  • What an unexpected method! How incredibly clever. However, I do think the presentation could definitely be cleaned up some. – zhuli Mar 22 '21 at 19:12
  • Yeah, I think so... Unfortunately I spent three days to write this down at least correctly, so if someone wants to review the exposition, I would be definitely grateful!! – Andrea Marino Mar 22 '21 at 19:38
  • @Zhuli: Can I ask you how would you suggest to improve the presentation? – Andrea Marino Mar 22 '21 at 19:55
  • 1
    I noticed that it takes a really long time to process what's happening when you "combine inequalities", so I think it would help if you explicitly laid out which inequalities you're combining and how. In addition, formatting by splitting into lines may help. Some lines are really long and difficult to read because it's hard to see where, say, the equal signs are. – zhuli Mar 22 '21 at 21:03
  • Hi @Zhuli, thanks for your feedback. I line spaced and numbered long equations in the out of the blue section. Also, I explained which inequalities and how they are combined, putting some emphasis on the log properties used if the passage is not explicit enough. Do you think now readability is improved? P.S. I moved the short section at the bottom, since I think it is the less instructive one – Andrea Marino Mar 23 '21 at 18:31
  • 1
    Much better! :) – zhuli Mar 24 '21 at 17:06
  • At the end of Step 1.3 (explicit numbers calculations), where did it come from? Because link – Benjamin Wang Mar 26 '21 at 21:10
  • I have highlighted in red a wrong inequality. – Benjamin Wang Mar 26 '21 at 21:42
  • 1
    It's a nice method. I have been able to follow the theory, and prove the logarithmic approximation for myself. However, the wrong inequality in one of the first steps is a little distracting. I imagine fixing this will require changing all the other steps(?) – Benjamin Wang Mar 26 '21 at 22:00
9

$\color{green}{\textbf{Just now!}}$

$$\ln\dfrac{75^{32}}{7^{71}}=\ln\left(\dfrac{20}{21}\left(\dfrac{2400}{2401}\cdot\dfrac{225}{224}\right)^{14}\dfrac{78125}{78732}\right)$$ $$=\ln\left(\dfrac{41-1}{41+1}\left(\dfrac{4801-1}{4801+1}\cdot\dfrac{449+1}{449-1}\right)^{14}\dfrac{156857-607}{156827+607}\right)$$ $$= 2\sum\limits_{k=1}^\infty\dfrac1{2k-1}\left(-\dfrac1{41^{2k-1}}-\dfrac{14}{4801^{2k-1}}+\dfrac{14}{449^{2k-1}}-\dfrac{607^{2k-1}}{156827^{2k-1}}\right)$$ $$< 2\left(-\dfrac1{41}-\dfrac1{4\cdot41^3}-\cdot\dfrac{14}{4801}+\dfrac{14}{449}-\dfrac{607}{156827}\right)$$ $$-\dfrac1{6\cdot41^3}+ \dfrac1{48\cdot449^2}\left(1+\dfrac{5}{3\cdot 449^2}+\dfrac{5^2}{3^2\cdot 449^4}+\dots\right) <0.$$ Proved!


$\color{brown}{\textbf{Previous version.}}$

Equivalent inequality is $$71\ln 7 > 32\ln3+64\ln 5.$$

Taking in account numeric equalities $$7^5=16807,\quad 3^3\cdot 5^4=16875,\quad 5^6 = 15625,\quad 3^7\cdot7 = 15309,$$ one can denote $$2\ln5-3\ln3=\ln\dfrac{25}{27} = \ln\dfrac{1-x}{1+x},\quad x = \dfrac1{26},$$ $$3\ln3+4\ln5-5\ln 7= \ln \dfrac{3^3\cdot5^4}{7^5} = \ln\dfrac{1+y}{1-y},\quad y = \dfrac{34}{16841},$$ $$7\ln3+\ln7-6\ln5 = \ln \dfrac{3^7\cdot7}{5^6} = \ln\dfrac{1-z}{1+z},\quad z = \dfrac{158}{15467}.$$

Easily to check expression for the goal sum: $$S=32\ln3+64\ln 5-71\ln7 = \ln\dfrac{1-x}{1+x}+14\ln\dfrac{1+y}{1-y}-\ln\dfrac{1-z}{1+z}.$$ If $\;x,y,z,t>0\;$ then $$S = -\ln\dfrac{1+x}{1-x}+14\ln\dfrac{1+y}{1-y}+\ln\dfrac{1+z}{1-z},$$ $$\ln\dfrac{1+t}{1-t} = \ln(1+t)-\ln(1-t) = 2t +\dfrac23t^3+\dfrac25t^5+\dfrac27t^7+\dots,$$ $$S_1=-\ln\dfrac{1+x}{1-x} < -2x-\dfrac23x^3 = -\dfrac{2x(3+x^2)}{3}$$ $$S_1<-\dfrac{2029}{26364}<-0.076\,961\,007,$$ $$S_2=14\ln\dfrac{1+y}{1-y}<28y(1+y^2+y^4+\dots)=\dfrac{28y}{1-y^2},$$ $$S_2<\dfrac{28\cdot34\cdot16841}{16841^2-34^2}<0.056\,528\,941,$$ $$S_3=\ln\dfrac{1+z}{1-z}\le2z+\dfrac23z^3(1+z^2+z^4+\dots) =2z+\dfrac23\,\dfrac{z^2}{1-z^2} =\dfrac{2z}{3}\dfrac{3-2z^2}{1-z^2},$$ $$S_3 <\dfrac{316}{46401}\dfrac{3\cdot15467^2-2\cdot158^2}{15467^2-158^2}<0.020\,431\,305$$ $$S=S_1+S_2+S_3 < -0.000\,000\,761 < 0.$$

Proved!

  • 1
    Interesting approach of breaking the logarithm into smaller chunks. Could you please explain: 1) how you came up with the splitting of logs, and 2) how one might compare the fractions with 9-digit denominators efficiently? Thanks. Downvote not mine by the way. – Benjamin Wang Mar 19 '21 at 21:56
  • Also, I think an explanation of how the error margins of the natural logs don't change the conclusion would be nice to have. – zhuli Mar 20 '21 at 00:03
  • @BenjaminWang Thank you for advices! Elaborated. – Yuri Negometyanov Mar 20 '21 at 10:38
  • @Zhuli Thanks a lot! Done. – Yuri Negometyanov Mar 20 '21 at 10:40
  • 5
    I appreciate your effort. But this is not the answer I was looking for.. People who are probably here can do this using at least 15-20 lines of inequality, some theorems. Your answer is good, but the answer I am looking for is the best one and cleverly. Involves excessive computation.What I'm looking for is more than an answer. A more striking answer, definitely less calculating, without logarithm. If someone gives such a striking answer, I'll definitely take off a hat.. – lone student Mar 20 '21 at 11:08
  • 1
    @lonestudent Thank you for the comment! Read my second answer. – Yuri Negometyanov Mar 20 '21 at 17:00
  • 1
    In my opinion, logarithms are fine and still in the spirit of the problem, as long as it doesn't involve something like Taylor series and error margins, which is usually a calculus thing and also just dry. – zhuli Jun 16 '21 at 07:57
6

Here's a basic approach where the overwhelming majority of the work is squaring three 9-digit numbers. Make of that what you will!

We find that $75$ is $135$ in base 7. I'll write $\ldots$ to mean some sequence of zeros. Successively square using arithmetic in base 7 to get

\begin{align*} 135^1 &= 135 \\ 135^2 &= 22254 \\ 135^4 &= 22254^2 = 532640502 \\ 135^8 &= 532640502^2 = 420605123423162604 < 420605124\ldots \\ 135^{16} &< 420605124\ldots^2 < 243433061\ldots \\ 135^{32} &< 243433061\ldots^2 < 66666663\ldots \end{align*}

If you count the number of $0$'s in the final $\ldots$, you'll see that $135^{32}$ (in base 7) is at most 71 digits, which is less than $7^{71}$ which is 72 digits.

The $22254^2 = 532640502$ step would be tedious by hand, but really not bad. The obnoxious bit would be $532640502^2 = 420605123423162604$ and the two virtually identical calculations after it. Strictly speaking you only have to do enough calculations to assure yourself the claimed upper bounds hold, though that probably doesn't help much.


You could use Karatsuba's algorithm to make the big squaring operations less painful. For instance,

\begin{align*} 532640502^2 &= (5326\ldots + 40502)^2 \\ &= 5326\ldots^2 + 2 \cdot 5326\ldots \cdot 40502 + 40502^2 \\ &= 42051211\ldots + 630111343\ldots + 2255562604 \\ &= 4206051233(\cdots) \\ &< 420605124\ldots. \end{align*}

In all you'd need to do 4 5-digit squares, 3 4-digit squares, 3 4-digit and 5-digit products, and some easy additions and other little things. That's actually getting pretty reasonable.

  • 1
    Basically this is doing floating point arithmetic by hand, but in base $7$. I wonder whether there are people who are familiar with base $7$ arithmetic ... A multiplication table would definitely be handy. Instead of writing $\dots$, you may also write $\times 7^*$ so that the exponent is more explicit. – WhatsUp Mar 20 '21 at 12:02
  • Sure, scientific notation would be best. I just didn't want to actually do the bookkeeping--I certainly didn't do any of the multiplications myself either! – Joshua P. Swanson Mar 20 '21 at 12:33
  • Thank you for the arguably natural way to go about it, using the base provided by the question. For completeness, here's a multiplication table in base 7 https://math.tools/table/multiplication/base/7 – Benjamin Wang Mar 20 '21 at 15:34
  • At the same time, my optimized proof does not use big numbers at all. – Yuri Negometyanov Mar 21 '21 at 23:07
  • @YuriNegometyanov Hmm, I suppose that's a matter of opinion. Your lovely "digitally simplified" argument does for instance use 1791015625 (in base 10). Using Karatsuba's algorithm, mine never actually requires tracking more than 10 or so base 7 digits at once. In particular you never actually have to compute 420605123423162604, though it's also very easy to read it off from the work that is required.... – Joshua P. Swanson Mar 22 '21 at 06:25
  • Okay, The terrible number is eliminated. Are 8 digits so hard too? :-) – Yuri Negometyanov Mar 22 '21 at 07:35
  • @YuriNegometyanov Haha, thanks for the chuckle! – Joshua P. Swanson Mar 22 '21 at 08:40
  • So this works out to just over 200 base-7 multiplications and around the same number of additions (depending how you count additions)? I guess it's "in principle" doable by hand if one is determined. Of course, if you are too sloppy, a lot of the work needs to be redone. – Benjamin Wang Mar 26 '21 at 22:25
0

Just for your curiosity :

Using Young's inequality for product we get for $x>1$ :

$$f(x)=\left(\frac{(\frac{75}{7})^{x}}{x}+\frac{x-1}{x}7^{\frac{x}{x-1}}\right)>75$$

Now we have to show for some $x>1$ :

$$g(x)=\frac{f(x)}{49*7^{\frac{7}{32}}}<1$$

And now the miracle of the inequality is :

$$g\Big(\frac{71}{39}\Big)<1$$

And as we now $71=39+32$

So the inequality can be rewritten :

$$\frac{\ln(7)}{\ln(75)-\ln(7)}+1>\frac{71}{39}$$

The LHS is the equality case for the Young's inequality .

So now I suspect we can use Reductio ad absurdum to solve it .

Well some explanations :

If we suppose :

$$7^{71}<75^{32}$$

It's easy to remark that it's equivalent to :

$$\frac{75*39}{71}+\frac{32}{71}7^{\frac{71}{32}}>75$$

By the hypothesis wich make the reductio ad absurdum plausible we have :

$$\Big(\frac{75}{7}\Big)^{\frac{71}{39}}>75$$

So we have :

$$\frac{\Big(\frac{75}{7}\Big)^{\frac{71}{39}}39}{71}+\frac{32}{71}7^{\frac{71}{32}}>\frac{75*39}{71}+\frac{32}{71}7^{\frac{71}{32}}>75$$

Wich is the Young's inequality !

Now to get the end we need to show :

$$7^{\frac{71}{32}}>\frac{\Big(\frac{75}{7}\Big)^{\frac{71}{39}}39}{71}+\frac{32}{71}7^{\frac{71}{32}}$$

So we need to find an upper bound instead of a lower bound .

Any idea ?

0

$\color{green}{\textbf{Evident approach.}}$

Let us prove the chain of inequalities: $$\dfrac{1709}{9380}\,\dfrac{75^4}{49^4} =\dfrac{1709}{9380}\,\dfrac{31\,640\,625}{5\,764\,901} ==\dfrac{1709}{9380}\,\dfrac{3373\cdot9380+1885}{3373\cdot1709+344}$$ $$=\dfrac{3373\cdot1709\cdot9380+3\,221\,465}{3373\cdot1709\cdot9380+3\,326\,720} < 1,$$ $$\dfrac{75^4}{49^4} <\dfrac{9380}{1709},\tag1$$ $$\dfrac{5256}{655}\,\dfrac{363\,970}{2\,920\,681} =\dfrac{5256}{655}\,\dfrac{555\cdot655+445}{555\cdot5256+3601} =\dfrac{555\cdot1709\cdot9380+2\,338\,9200}{555\cdot1709\cdot9380+2\,358\,655} < 1,$$ $$\dfrac{9380^2}{1709^2}=30+\dfrac{363\,970}{2\,920\,681} < 30+\dfrac{655}{5256},\tag2$$ $$\dfrac{3493}{1721}\,\dfrac{13\,611\,073}{27\,625\,536} =\dfrac{3493}{1721}\,\dfrac{7908\cdot1721+1405}{7908\cdot3493+2892} =\dfrac{7908\cdot3793\cdot1721+4\,907\,665}{7908\cdot3793\cdot1721+4\,977\,132} < 1,$$ $$\left(30+\dfrac{655}{5256}\right)^2=907+\dfrac{1721}{3493},\tag3$$ $$\left(907+\dfrac{1721}{3493}\right)^2=822649+\dfrac{1814\cdot1721}{3493}+\dfrac{1721^2}{3493^2},$$ $$ = 822649+893+\dfrac{(2645+1721)1721}{3493^2}= 823542+\dfrac{12\,200\,826}{12201049} < 823543,$$ $$\left(907+\dfrac{1721}{3493}\right)^2 < 823543 = 7^7.\tag4$$ From $(1)-(4)$ should $$\left(\dfrac{75^4}{49^4}\right)^{2\times 2\times2}< 7^7,$$ $$\color{brown}{\mathbf{7^{71} > 75^{32}.}}$$

0

Using methods similar to those in the earlier version of Yuri Negometyanov's first answer (this won't please @lone-student!), we prove the inequality by (implicitly - see the comment) calculating precise error bounds for the approximations $7^{41} \bumpeq 2^{115},$ $75^{41} \bumpeq 2^{255}.$

With the exception of the brute force decimal calculation at the end (I carried it out on paper to see how bad it is - it's fairly bad, but doable and checkable), not much use of pencil and paper should be needed to follow the calculations, because all but the most routine steps are spelled out in detail.

For brevity, write: $$ s(x) = \tfrac12\ln\frac{1 + x}{1 - x} = x + \frac{x^3}3 + \frac{x^5}5 + \frac{x^7}7 + \frac{x^9}9 + \cdots \quad (|x| < 1), $$ and define $$ a = \frac1{4801}, \ b = \frac{37}{65573}, \ c = \frac5{2053}, \ d = \frac13, $$ so that \begin{gather*} \frac{1 + a}{1 - a} = \frac{4802}{4800} = \frac{2401}{2400} = \frac{7^4}{2^5\cdot75}, \\ \frac{1 + b}{1 - b} = \frac{65610}{65536} = \frac{32805}{32768} = \frac{81^2\cdot5}{2^{15}} = \frac{3^8\cdot5}{2^{15}}, \\ \frac{1 + c}{1 - c} = \frac{2058}{2048} = \frac{1029}{1024} = \frac{3\cdot7^3}{2^{10}}, \\ \frac{1 + d}{1 - d} = 2. \end{gather*} We then have \begin{align*} 2s(a) & = 4\ln7 - 5\ln2 - \ln75, \\ 4s(b) & = 15\ln3 + \ln75 - 30\ln2, \\ 2s(c) & = \ln3 + 3\ln7 - 10\ln2, \\ 2s(d) & = \ln2, \end{align*} whence: \begin{gather*} 2[1369s(a) + 114s(b) - 855s(c) + 5s(d)] = 1369(4\ln7 - 5\ln2 - \ln75) + {} \\ 57(15\ln3 + \ln75 - 30\ln2) - 855(\ln3 + 3\ln7 - 10\ln2) + 5\ln2 \\ = (5476 - 2565)\ln7 + (57 - 1369)\ln75 - (6845 + 1710 - 8550 -5)\ln2 \\ = 2911\ln7 - 1312\ln75 \\ = 41(71\ln7 - 32\ln75). \end{gather*} We are thus reduced to proving: \begin{equation} \label{4058725:eq:1}\tag{1} 1369s(a) + 114s(b) + 5s(d) > 855s(c). \end{equation} To begin with, \begin{gather*} 5s(d) > 5\left(\frac13 + \frac1{3\cdot3^3} + \frac1{5\cdot3^5} + \frac1{7\cdot3^7} + \frac1{9\cdot3^9}\right) \\ = \frac{5\cdot7 + 3^4(5 + 3^2\cdot7(1 + 5\cdot3(1 + 3^3)))} {3^{11}\cdot7} = \frac{5\cdot7 + 3^4(5 + 3^2\cdot7\cdot421)}{1701\cdot3^6} \\ = \frac{5\cdot7 + 3^4(5 + 9\cdot2947)}{15309\cdot81} = \frac{5\cdot7 + 81\cdot26528}{1240029} = \frac{35 + 2148768}{1240029} = \frac{2148803}{1240029}. \end{gather*} Next, \begin{gather*} 855s(c) < 855c\left(1 + \frac{c^2}3\left( 1 + c^2 + c^4 + \cdots\right)\right) = \frac{4275}{2053}\left(1 + \frac{c^2}{3(1 - c^2)}\right). \end{gather*} For the bracketed factor, a crude estimate will do. We have $c < 1/400,$ therefore $c^2/3 < 1/480000$ and $1 - c^2 > 5/6,$ therefore: $$ \frac{c^2}{3(1 - c^2)} < \frac1{400000}. $$ The remaining two terms in \eqref{4058725:eq:1} are mercifully easy to handle: $$ 1369s(a) > \frac{1369}{4801}, \quad 114s(b) > \frac{4218}{65573}. $$ I see nothing for it now but to resort to long division, but it's not too bad - at least not if you cheat, as I did, by doing the long divisions on a calculator, and verifying the results by long multiplication on paper! Rounding down the terms on the left of \eqref{4058725:eq:1}, and rounding up the term on the right, we get: \begin{align*} \frac{1369}{4801} & > 0.285148 \\ \frac{4218}{65573} & > 0.064325 \\ \frac{2148803}{1240029} & > 1.732865 \\ \therefore\ \text{lhs}\,\eqref{4058725:eq:1} & > 2.082338, \end{align*} while on the other hand: \begin{align*} \frac{4275}{2053} & < 2.082319 \\ \frac{4275}{2053\times400000} & < 0.000006 \\ \therefore\ \text{rhs}\,\eqref{4058725:eq:1} & < 2.082325 < \text{lhs}\,\eqref{4058725:eq:1}. \quad\square \end{align*}

  • Although the formulae \begin{gather} 41\ln7 = 115\ln2 - 2s(a) - 4s(b) + 30s(c), \ 41\ln75 = 255\ln2 - 90s(a) - 16s(b) + 120s(c) \end{gather} are easily obtainable, their inclusion only lengthens the argument unnecessarily. – Calum Gilhooley Dec 22 '21 at 00:45