I am sure I saw a similar question like this one before but I can't find it now. I tried using "order" but failed. It is quite obvious when $y$ is an even number. The real problem is when $y$ is odd. Is there any easy way to solve this? Thanks.
-
4It's a special case of Mihailescu's theorem (formerly Catalan's conjecture). A bit of a sledgehammer.... – Angina Seng Aug 01 '18 at 06:24
-
1A solution using modular equivalences would not work, because $7^2 \equiv -1 \pmod {5^2}$ – Kenny Lau Aug 01 '18 at 06:31
-
Yes. I wonder if there is a shorter proof rather than a general one. It seems that this is only an exercise for number theory. It shouldn't be too hard. @Lord Shark the Unknown – Zhenyuan Lu Aug 01 '18 at 06:46
-
Well to prove Mihailescu's theorem, you have to prove special cases like the one above anyway $\ddot\smile$ @ZhenyuanLu – Angina Seng Aug 01 '18 at 06:47
-
Alright. That's a good suggestion. I'll read through the proof of Mihailescu's theorem. Thanks! @ Lord Shark the Unknown – Zhenyuan Lu Aug 01 '18 at 06:50
-
It is easy to show that $y$ is not even because $1=(5^r+x)(5^r-x)$ has no solution. I think that chapter 3 of https://web.archive.org/web/20060221125555/http://www.math.leidenuniv.nl/~jdaems/scriptie/Catalan.pdf has a method which will work – Mark Bennet Aug 01 '18 at 07:00
-
1@MarkBennet … apart from the trivial solution of $r=0, x=0$ :) – PM 2Ring Aug 01 '18 at 07:31
-
Yes. It is quite obvious that $y$ should be odd. Thank you for the suggestion. @ Mark Bennet – Zhenyuan Lu Aug 01 '18 at 07:39
-
Yes. Using Legendre symbol would not work. That is why I tried using "order". @ Kenny Lau – Zhenyuan Lu Aug 01 '18 at 07:41
-
See also: On equations $m^2+1=5^n$. Found using Approach0. – Martin Sleziak Aug 01 '18 at 08:43
-
FWIW, I just posted an answer using elementary methods on the linked question. – PM 2Ring Aug 02 '18 at 11:40
1 Answers
Not a full answer, but way too long for a comment and might have some useful ideas:
We see that, in complex numbers, we need
$$(x+i)(x-i)=(2+i)^y(2-i)^y.$$
So, we have
$$x+i=\pm (2+i)^m(2-i)^n, x-i=\pm (2+i)^{y-m}(2-i)^{y-n}.$$
So,
$$\gcd(x+i,x-i)=\gcd(x+i,x-i)|2i,$$
but a divisor of it is also
$$(2+i)^{\min(m,y-m)}(2-i)^{\min(n,y-n)};$$
as each of $2+i$ and $2-i$ are coprime to $2i$, we must thus have that $m,n\in\{0,y\}$. Clearly, this implies $(m,n)=(0,y)$ or $(m,n)=(y,0)$. Either way,
$$\Im((2+i)^y)=\pm 1.$$
Let $2+i=z=re^{i\theta}$. Then we have that
$$z^y\pm z\in\mathbb{R}$$
$$z^y\pm z = \bar{z}^y\pm\bar{z}$$
$$z^y-\bar{z}^y = \pm(z-\bar{z})$$
$$\frac{z^y-\bar{z}^y}{z-\bar{z}}=\pm 1.$$
Define
$$a_n=\frac{z^n-\bar{z}^n}{z-\bar{z}}.$$
We have that, as $a_n=Az^n+B\bar{z}^n$ for some complex numbers $A$ and $B$,
$$a_{n+2}-4a_{n+1}+5a_n=0,$$
and $a_0=0$, $a_1=1$. It suffices to show that no element of this sequence with index $>1$ has magnitude $1$.
Another idea is that
$$\text{Im}\left(z^n\right)=r^n\sin(n\theta),$$
so we need
$$\sin(n\theta)=\pm\frac{1}{r^n},$$
which implies the existence of a positive integer $m$ so that
$$|n\theta-m\pi|\leq \frac{2}{r^n}$$
(as $x/2<\sin(x)$ for all $0<x<\pi/4$, say). This implies that
$$\left|\frac{\theta}{\pi}-\frac{m}{n}\right|\leq \frac{2}{\pi(nr^n)}.$$
In other words, $\theta/\pi$ is very well-approximated by rational numbers.
- 32,931
-
Using complex numbers is a really interesting idea. Is it true that we can imply $m+n=y$ from the second step? @Carl Schildkraut – Zhenyuan Lu Aug 01 '18 at 07:37
-
@ZhenyuanLu Yes it is, because $|x+i|=|x-i|$ - that's a simplification I didn't think of. Then you can simply say $5\nmid \pm 1$, so ${m,n}={0,y}$. Ah well, I got there anyway. – Carl Schildkraut Aug 01 '18 at 12:24