It is well-known that any positive rational number can be written as the sum of finitely many distinct unit fractions. This is easy since $$\frac1n=\frac1{n+1}+\frac1{n(n+1)}\quad\text{for all}\ n=1,2,3,\ldots.$$ In 2015 I thought that this easy fact should have a further refinement which is somewhat sophisticated. Note that the series $\sum_p\frac1{p-1}$ and $\sum_p\frac1{p+1}$ (with $p$ prime) diverge just like the harmonic series $\sum_{n=1}^\infty\frac1n$. Also, for any positive integer m there are infinitely many primes $p$ congruent to $1$ (or $-1$) modulo $m$ (by Dirichlet's theorem). Motivated by this, I made the following conjecture in Sept. 2015.

Conjecture. For any rational number $r>0$, there are finite sets $P_r^-$ and $P_r^+$ of primes such that $$r=\sum_{p\in P_r^-}\frac1{p-1}=\sum_{p\in P_r^+}\frac1{p+1}.$$

This appeared as Conjecture 4.1 of this published paper of mine. For example, $$2=\frac1{2-1}+\frac1{3-1}+\frac1{5-1}+\frac1{7-1}+\frac1{13-1}$$ with $2,3,5,7,13$ all prime, and $$1=\frac1{2+1}+\frac1{3+1}+\frac1{5+1}+\frac1{7+1}+\frac1{11+1}+\frac1{23+1}$$ with $2,3,5,7,11,23$ all prime. Also, \begin{align*}\frac{10}{11}=&\frac1{3-1}+\frac1{5-1}+\frac1{13-1}+\frac1{19-1}+\frac1{67-1}+\frac1{199-1} \\=&\frac1{2+1}+\frac1{3+1}+\frac1{5+1}+\frac1{7+1}+\frac1{43+1}+\frac1{131+1}+\frac1{263+1} \end{align*} with $2,3,5,7,13,19,43,67,131,199,263$ all prime. The reader may see more numerical data in my detailed introduction to this conjecture.

After learning this conjecture from me, Prof. Qing-Hu Hou and Guo-Niu Han checked my above conjecture seriously and their computational results support my conjecture. For example, in 2018 Prof. Han found 2065 distinct primes $p_1<\ldots<p_{2065}$ with $p_{2065}\approx 4.7\times10^{218}$ such that $$\frac1{p_1+1}+\ldots+\frac1{p_{2065}+1}=2.$$

My question is whether the above conjecture is true. I would like to offer 500 US dollars as the prize for the first correct solution.

Remark. Let $r$ be any positive rational number, and let $\varepsilon\in\{\pm1\}$. As the series $\sum_p\frac1{p+\varepsilon}$ (with $p$ prime) diverges, there is a unique prime $q$ such that $$\sum_{p<q}\frac{1}{p+\varepsilon}\le r<\sum_{p\le q}\frac1{p+\varepsilon}.$$ Thus $$0\le r_0:=r-\sum_{p<q}\frac1{p+\varepsilon}<\frac1{q+\varepsilon}\le1.$$ If $r_0=\sum_{j=1}^k\frac1{p_j+\varepsilon}$ with $p_1,\ldots,p_k$ distinct primes, then $p_1,\ldots,p_k$ are all greater than $q$, and $$r=\sum_{p<q}\frac1{p+\varepsilon}+\sum_{j=1}^k\frac1{p_j+\varepsilon}.$$ Therefore it suffices to consider the conjecture only for $r<1$.

  • 5
    $\begingroup$ Voting for this question because it rekindles in me the sheer joy of natural numbers. There will, I sincerely hope, always be people interested in 'strange' questions like this one, that provoke the creative mind. $\endgroup$ Nov 29, 2018 at 10:38
  • 7
    $\begingroup$ Interesting conjecture. One could consider defining a set $S$ of natural numbers to be Egyptian if every positive rational number admits an Egyptian fraction decomposition with all the denominators in $S$, and then ask for sufficient conditions for $S$ to be Egyptian. Perhaps it suffices for $S$ to have high enough density and to satisfy certain congruence conditions? $\endgroup$ Nov 29, 2018 at 21:37

1 Answer 1


This conjecture is true (as is the version for $p-h$ for any $h\neq 0$).

The proof is too long to reproduce here, but the preprint is at https://arxiv.org/abs/2305.02689


Quick summary as requested: My previous paper (https://arxiv.org/abs/2112.03726) essentially shows that for any set $A$ which is large enough and satisfies some mild number theoretic conditions we can write $1$ as the sum of distinct unit fractions with denominators from $A$. (The most important of these conditions is 'smoothness', that no $n\in A$ is divisible by a large prime power.)

Such a result was previously proved by Croot in 2003, but with a stronger smoothness condition. The proof is a strengthening of Croot's method, using the circle method and some combinatorial analysis.

What's new is observing that we can show that the set $\{p-h\}$ satisfies these mild number theoretic conditions - which is to be expected, since these conditions are all 'generic', and we expect $p-h$ to look like a typical integer of the same size from a multiplicative point of view. In fact very classical techniques suffice here (e.g. the Siegel-Walfisz theorem, Selberg's sieve, and an Erdos-Kac type theorem due to Halberstam from the 1950s).

This means we can write $1$ as the sum of $\frac{1}{p-h}$. But this is robust, in that we can remove any finite set of primes and still find such a sum. Using the identity $n=1+\cdots+1$ we can therefore write any integer $n$ as the sum of distinct $\frac{1}{p-h}$. Now to write any $\frac{n}{m}$ do the above with primes $p\equiv h\pmod{m}$ and the set $A=\{\frac{p-h}{m}\}$.

  • 1
    $\begingroup$ What about a summary? $\endgroup$
    – Wlod AA
    May 5 at 10:47
  • 2
    $\begingroup$ Added a summary $\endgroup$ May 5 at 10:54
  • $\begingroup$ Many thanks -- it's so nice! $\endgroup$
    – Wlod AA
    May 5 at 11:09
  • $\begingroup$ Dr. Thomas, thank you very much for your surprising proof! $\endgroup$ May 6 at 7:32

Your Answer

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

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