23
$\begingroup$

First I will explain why a weaker form is needed. And then I formulate the conjecture (more precisely, the formulation will be clear).

It is related to the question https://math.stackexchange.com/questions/40945/triangular-factorials and several Mathoverflow questions from the comments to that question. A number $m$ is called a triangular factorial if $m=\frac{n(n+1)}{2}=k !$ for some $n,k$. It is an open problem whether the set of triangle factorials is finite. Moreover the only known such numbers are $1, 6, 120$.

But (somewhat surprisingly for me) it can be shown that the ABC conjecture implies that there are only finitely many triangular factorials. Indeed, suppose that for arbitrary large $k,m$ we have $ \frac{n(n+1)}{2}=k!$. Then $n+1=\frac {2k!}{n}$. Let $a=n, b=1, c= \frac {2k!}{n}$. Then by the ABC conjecture $\frac {2k!}{n}<rad(2k!)^2$ where $rad(x)$ is the product of primes dividing $x$. Note that $n\sim \sqrt{2k!}$ and $rad(2k!)=rad(k!)$ is the product of all primes $\le k$ which, by Erdos theorem $\sim e^{k}$. Thus we have $\sqrt{2k!}< e^{2k}$ which is impossible for big enough $k$. Recall that $2k!\sim 2\sqrt{2\pi k}\, e^{k\log k-k}$.

Question: In the proof above what seems to be a very weak version of the ABC conjecture is used (instead of $rad(abc)^{1+\epsilon}$ one can take a much bigger function in $rad(abc)$). Perhaps that version can be proved easier than the original ABC conjecture?

Edit: It is easy to see that in the version of ABC conjecture used here, $b=1$. Perhaps that makes the conjecture easier? So we can formulate

A conjecture For every constant $d<\frac 12$ there are only finite number of natural $a$ such that $$a>rad(a(a+1))^{d\log\log a}.$$ Note that the exponent in the right hand side may have to be a little different.

$\endgroup$
  • 4
    $\begingroup$ Yes, I do. Also, I need to write ten characters. $\endgroup$ – Felipe Voloch Jun 15 '17 at 2:53
  • 2
    $\begingroup$ I thought there is a dichotomy: a written proof can be either correct or not correct. It seems that there exists the third option. It is not related to the question, though. $\endgroup$ – Mark Sapir Jun 15 '17 at 3:03
  • 3
    $\begingroup$ The product of the primes up to $k$ is asymptotically $e^k$, not $e^{2k}$, and I think this is a simple consequence of the Prime Number Theorem and known before Erdos came along. $\endgroup$ – Gerry Myerson Jun 15 '17 at 4:08
  • 3
    $\begingroup$ So he left out two words: which, by Erdos theorem has square $\sim e^{2k}.$ $\endgroup$ – Aaron Meyerowitz Jun 15 '17 at 5:10
  • 2
    $\begingroup$ Your conjecture simply means that rad$(a(a+1))$ being bounded has finitely many solutions. This is true by the finiteness of solutions to the $S$-unit equation. But this doesn't help with the question you started with -- your reformulated conjecture is not strong enough for that application. $\endgroup$ – Lucia Jun 20 '17 at 4:46
5
$\begingroup$

For $a> e$, we have $\log \log a > 0$ and hence $x^{\log \log a}$ is an increasing function, so $$a>\mathrm{rad}(a(a+1))^{\frac {\log a}{\log\log a}}$$

then implies that $a^{\log \log a} > \mathrm{rad}(a(a+1))^{\log a}$. Note that $a^{\log \log a} = (e^{\log a})^{\log \log a} = (e^{\log \log a})^{\log a} = (\log a)^{\log a}$.

So we have $(\log a)^{\log a} > \mathrm{rad}(a(a+1))^{\log a}$,
so $(\log a) > \mathrm{rad}(a(a+1))$, so with $b=1$ and $c=a+1$ we have $$c > a > \exp(\mathrm{rad}(abc))$$

but for every $\varepsilon > 0$ there is a $K$ such that $c < \exp(K\,\mathrm{rad}(abc)^{\frac13+\varepsilon})$ (Stewart & Yu 2001) , which gives a contradiction for sufficiently large $\mathrm{rad}(abc)$. So we can just show that for every constant $L$, we have that $L > \mathrm{rad}(abc) = \mathrm{rad}(a(a+1))$ for only finitely many $a$.

This again follows form this result by Stewart & Yu (2001). If there were an infinitude of $a$ such that $L > \mathrm{rad}(a(a+1))$, since every such triple has a different value of $c$, there are arbitrarily large values of $c$ such that $L > \mathrm{rad}(abc)$, so $c < \exp(K\,\mathrm{rad}(abc)^{\frac13+\varepsilon}) < \exp(K\cdot L^{\frac13+\varepsilon})$, for arbitrarily large values of $c$, but the last is a constant, contradiction.

$\endgroup$
  • $\begingroup$ Thank you for the answer. It does resolve the wrong A conjecture. The conjecture has been improved. $\endgroup$ – Mark Sapir Jun 20 '17 at 13:41

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.