4
$\begingroup$

Is there a way to show that (with integer $n>2$) $$2^{\lceil n \log_23\rceil}-3^n<3^n-2^n$$ I tried to figure a way with derivative or by looking how both side are growing, but I have some trouble with the ceiling function.

$\endgroup$
3
  • 1
    $\begingroup$ Based on the existence of this open problem, this could be hard. en.wikipedia.org/wiki/Waring%27s_problem $\endgroup$
    – pancini
    Feb 11, 2021 at 19:26
  • $\begingroup$ Also, playing with Mathematica, it seems like the constant $\log_23$ is sharp here. I.e., the inequality fails if you use $\lceil k n \rceil$ with $k=1.585$, but holds out to $n=100000$ with $k=1.5848$. $\endgroup$
    – pancini
    Feb 11, 2021 at 19:29
  • $\begingroup$ Also fails with $k=1.5849626$, before $n=100000$ (and $k-\log_23\approx 9.92\cdot 10^{-8}$ here) $\endgroup$
    – pancini
    Feb 11, 2021 at 19:33

4 Answers 4

3
$\begingroup$

With $n+k= {\lceil n \log_23\rceil}<n \log_23+1$ we can use the Rhin bound (page 160: $|µ_1\log 2+µ_2\log 3|\geq H^{-13.3}$ with $H=max(|µ_1|,|µ_2|)$)

$$|(n+k) \log2 - n \log3|>\frac 1{(n+k)^{13.3}}>\frac1{(n \log_23+1)^{13.3}}>\frac 1{n^a}$$ If we choose $a=15$ the above is true for $n>41$

Now (growth of $x^a$ vs $a^x$), $$\frac 1{n^a}>\frac{1}{(\frac{3}{2})^n}$$ is true for $n>\frac{-aW_{-1}(\frac{-\log\frac{3}{2}}{a})}{log(\frac{3}{2})}$ or $n>196$ with chosen $a=15$ ($W$ is the productlog), and we also have (using $x>log(1+x)$) $$\frac{1}{(\frac{3}{2})^n}>\log(1+\frac{1}{(\frac{3}{2})^n})=\log(1+\frac{2^n}{3^n})$$ So for $n>196$ we have $$|(n+k)\log2 - n \log3|=\log(\frac{2^{n+k}}{3^n})>\log(1+\frac{2^n}{3^n})$$ or $$\frac{2^{n+k}}{3^n}>1+\frac{2^n}{3^n}$$ $$2^{\lceil n \log_23\rceil}-3^n>2^n$$

Similarly we have $$|n \log3-(n+k-1) \log2|>\frac 1{(n+k-1)^{13.3}}>\frac1{(n \log_23)^{13.3}}>\frac 1{n^a}$$ which also holds for $a=15$, and additionally we have: $$\frac{1}{(\frac{3}{2})^n}=\frac{1}{2^{n\log_2\frac{3}{2}}}>\frac{1}{2^{\lceil n\log_2\frac{3}{2}\rceil}}>\log(1+\frac{1}{2^{\lceil n\log_2\frac{3}{2}\rceil}})$$ So for $n>196$ we have $$|n \log3-(n+k-1) \log2|=\log(\frac{2\cdot 3^n}{2^{n+k}})>\log(1+\frac{1}{2^{\lceil n\log_2\frac{3}{2}\rceil}})$$ or $$\frac{2\cdot 3^n}{2^{n+k}}>1+\frac{1}{2^{\lceil n\log_2\frac{3}{2}\rceil}}$$ $$2\cdot 3^n>2^{n+k}+2^n$$

$$2^{\lceil n \log_23\rceil}-3^n<3^n-2^n$$

This leads to ($n>196$):

$$\begin{array}{|c|}\hline 2^n<2^{\lceil n \log_23\rceil}-3^n<3^n-2^n\\\hline\end{array}$$ and with manual checking, the left inequality holds except for $n$ in $\{1,3,5\}$ and the right inequality holds except for $n$ in $\{1,2\}$

$\endgroup$
3
  • $\begingroup$ Updated. There are still shortcuts in the development but I hope the edit clarifies some of them (I get that some of them were not clear). $\endgroup$
    – Collag3n
    Jun 10, 2021 at 12:15
  • $\begingroup$ Hi Collag3n - I'm preparing a question with the intention to become a "biglist" - a list of usable estimates (G.Rhin (87), W.J.Ellison (71), ???-hopefully some more); the answers should explain one of the known estimation-methods in use. Perhaps your answer here would be a good starting point for one answer in that big list? If you are interested to contribute to this, please see in the "meta-sandbox" math.meta.stackexchange.com/a/4668/1714 $\endgroup$ Oct 30, 2021 at 12:34
  • 1
    $\begingroup$ @Gottfried Helms, feel free to use anything I wrote in MSE. I also put an answer in the sandbox which I think is mixing the answer above and Ellison's bound math.meta.stackexchange.com/a/4669/476591. In fact, you could adjust the bounds for some large $n$. This is probably what Tao meant by: "Thus we see that any proposed proof of the Collatz conjecture must either use transcendence theory, or introduce new techniques that are powerful enough to create exponential separation between powers of 2 and powers of 3" $\endgroup$
    – Collag3n
    Oct 30, 2021 at 20:18
3
$\begingroup$

If you calculate $n \cdot \log_2 (3)$ for $n = 1, 2, 3, ...$ and look at the digits after the decimal point, you will get a rather unpredictable looking sequence of numbers from 0 to 1. Using the ceil function will increase the argument by an amount between 0 and 1, so two raised to this power will be somewhere between $3^n$ and $2 \cdot 3^n$.

Your inequality will be wrong if $n \cdot\log_2(3)$ is just a tiny bit larger than an exact integer. You can calculate how close it has to be to make the inequality wrong. You can calculate how likely it would be, if those last digits were random. Since these digits are not random, but in fact predictable, it could be possible to prove this either way. Others have commented that it seems to be an unsolved problem.

$2^n$ gets small compared to $3^n$ very quickly, so I’d be willing to make a small bet that the inequality is indeed true, but would be absolutely unable to prove it.

$\endgroup$
2
  • $\begingroup$ To elaborate a bit: If instead of looking at the sequence of fractional parts of $n log_2(3)$ you were to pick a sequence of numbers all between 0 and 1 independently and uniformly at random, then the theorem would be true with very high probability especially if you condition on there being no small counterexamples. $\endgroup$
    – Nate
    Feb 11, 2021 at 21:01
  • $\begingroup$ Yes I noticed that the gap/difference LHS/RHS was growing as $n$ grows, but got nowhere with that. I already looked at the Waring problem some time ago, but I won't even try if this problem is equivalent. $\endgroup$
    – user489810
    Feb 11, 2021 at 21:18
2
$\begingroup$

Start with $$2^{\lceil n \log_23\rceil}-3^n<3^n-2^n$$

Bring things to the right side: $$2^{\lceil n \log_23\rceil}<2\cdot 3^n \cdot (1-2^n / (2\cdot 3^n))$$

Take base-2 logarithm: $${\lceil n \log_23\rceil}<1 + n \log_23 + \log_2 ( (1-2^n / (2\cdot 3^n)))$$

Since $n \log_23$ is irrational for n >= 1: $${\lfloor n \log_23\rfloor + 1}<1 + n \log_23 + \log_2 ( (1-2^n / (2\cdot 3^n)))$$

$$n \log_23 - {\lfloor n \log_23\rfloor}< - \log_2 ( (1-2^n / (2\cdot 3^n)))$$

The right hand side is about $$2^n / (2\cdot 3^n) / \ln 2$$ or $${(2/3)}^n / (2\cdot \ln 2)$$

The original inequality is true as long as $n \log_23$ is more than ${(2/3)}^n / (2\cdot \ln 2)$ above the next lower integer. This series shrinks very fast, so it’s most likely true for all n.

Using Euclid’s algorithm as if you were calculating $\gcd(\log_23, 1)$ you can find consecutive n where the distance is as small as possible, and using some software that can perform these calculations in very high precision you should be able to prove this for very large n, say $n < 10^{1,000,000}$. Which doesn’t prove it’s true for all n.

$\endgroup$
-2
$\begingroup$

Start by showing that $$2^{\lceil \log_23^n\rceil}=2^{\lceil n \log_23\rceil}<3^n$$ Then the rest follows!

$\endgroup$
1
  • $\begingroup$ You seem to have confused floor and ceil. $\endgroup$
    – gnasher729
    Feb 11, 2021 at 20:56

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .