11
$\begingroup$

It is well known that $\frac{1}{2}+\frac{1}{3}+\frac{1}{6}=1$ and this is the only solution to $\frac{1}{x_1}+\frac{1}{x_2}+\frac{1}{x_3}=1$
with $2\leq x_1<x_2<x_3$.
My question is:

Let $n\in \mathbb{N}$ and $x_i\in \mathbb{N} ,1\leq i\leq n$.

How many $n$-tuples $(x_1,x_2,...,x_n)$ exist such that $\frac{1}{x_1}+\frac{1}{x_2}+\cdots+\frac{1}{x_n}=1$ with $2\leq x_1<x_2<...<x_n$ ?

Any reference would be appreciated!

$\endgroup$
6
  • 3
    $\begingroup$ See this paper with upper and lower bounds: link.springer.com/article/10.1134%2FS0001434614010295 $\endgroup$
    – Ian Agol
    Jan 28, 2016 at 15:33
  • 1
    $\begingroup$ See also: oeis.org/A006585 $\endgroup$
    – Ian Agol
    Jan 28, 2016 at 16:24
  • $\begingroup$ math.stackexchange.com/questions/290435/… started out as a question about the case $n=5$, but several of the answers turned to the more general case. $\endgroup$ Jan 28, 2016 at 21:44
  • 1
    $\begingroup$ The paper by Konyagin (mentioned above by Ian) gives the best known lower bound. The best known upper bound is in a paper by Browning and Elsholtz, The number of representations of rationals as a sum of unit fractions, Illinois J. Math. 55, Number 2 (2011), 685-696. maths.bris.ac.uk/~matdb/preprints/IJM343.pdf $\endgroup$ Jan 29, 2016 at 11:17
  • $\begingroup$ @ChristianElsholtz, do you think the asymptotics with distinct denominators is comparable to allowing repeats like in your and Sandor's papers? $\endgroup$
    – Ian Agol
    Jan 29, 2016 at 14:15

3 Answers 3

7
$\begingroup$

N. Burshtein has published several papers on this problem, On distinct unit fractions whose sum equals 1, and a more recent paper, making the problem more challenging by adding the restriction that the $n$ integers $x_i$ must be odd. There are no solutions for even $n$. The smallest odd $n=2k+1$ for which a solution exists is $k=4$, and the number of solutions is five.

For larger $n=2k+1$ the number of solutions (with the odd denominator restriction) is not known, but a lower bound of $(\sqrt 2)^{(k+1)(k-4)}$ was derived in Egyptian fractions with restrictions.

$\endgroup$
9
$\begingroup$

It might be worth remarking that if $G$ is a finite group with $n$ conjugacy classes and representatives $g_{1},g_{2}, \ldots ,g_{n}$, then the class equation for $G$ (divided through by $|G|$) gives $\sum_{i=1}^{n} \frac{1}{|C_{G}(g_{i})|} = 1$, though the $|C_{G}(g_{i})|$ are not usually distinct. Long ago (around 1895, I think) , E. Landau gave a bound on the number of solutions of $\sum_{i=1}^{n} \frac{1}{x_{i}} = 1$ with the $x_{i}$ (not necessarily distinct) positive integers. This allows a crude bound on the size of a finite group with $n$ conjugacy classes though this is quite far from the presently known bounds (eg of L. Pyber) which use much more group-theoretic information.

$\endgroup$
2
  • 2
    $\begingroup$ an upvote because of new (group theoretic) view. $\endgroup$
    – Shahrooz
    Jan 28, 2016 at 19:25
  • 1
    $\begingroup$ I wonder if any other kinds of finite algebraic structure might also generate solutions? I found this paper which seems relevant: math.ucr.edu/~jbergner/GpdEFrac.pdf $\endgroup$
    – Flounderer
    Jan 28, 2016 at 22:01
8
$\begingroup$

Any perfect number is an answer for your question. Also, Prof. Graham showed that $a(m)>0$ for $m>77$, where $a(m)$ denotes the total number of solution for this equation such that the sums of the denominators is $m$. You can see this page for the number of such solutions:

http://oeis.org/A051907

Also the below papers studied somehow the question:

[1] "Representation of One as the Sum of Unit Fractions" by Yuya Dan

[2] "UNIT FRACTIONS THAT SUM TO 1" by Yutaka Nishiyama

$\endgroup$
1
  • $\begingroup$ For clarity, since the OP used the variable $n$ in $\frac{1}{x_n}$, it might be better to use the variable $m$ as the sum of the denominators. $\endgroup$ Jan 28, 2016 at 14:36

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.