I want to rigorously prove that given any $x\geq 0$, we can write $x=\sum_{k\in J} \frac{1}{k}$ for some $J\subseteq \mathbb{N}$ eg. $2=\sum \frac{1}{2^{k}}$.

Also, for what other $\{a_{k}\geq 0\}$ is that true (any theorems?)? For example, it fails for $a_{k}=\frac{1}{k!}$: Writing number as sum of reciprocals of factorial

Another possible $\{a_{k}\}$ collection is $\mathbb{N}\cup \{\frac{1}{2^{n}}\}$.

For conditionally convergent we have the Riemann series theorem but in our case $\sum a_{k}$ is not even conditionally convergent.

Anyhow back to proving it for reciprocals.

So there is a bunch of algorithms https://en.wikipedia.org/wiki/Greedy_algorithm_for_Egyptian_fractions

and a proof of what I want in "The Approximation of Numbers as Sums of Reciprocals". But the proof there is a bit too wordy for me. I bet if I sit down I can flesh it out completely.


1)First for rationals $\frac{x}{y}$, we split into $\frac{x}{y}=\frac{1}{\lceil y/x\rceil}+\frac{(-y)\bmod x}{y\lceil y/x\rceil}$, then repeat for $\frac{(-y)\bmod x}{y\lceil y/x\rceil}$ and the result follows by induction on x ($x=1$ and then $x>1$):

$\frac{x+1}{y}-\frac{1}{\lceil y/x+1\rceil}=\frac{(x+1)\lceil y/x+1\rceil-y}{y\lceil y/x+1\rceil}\leq \frac{(x+1)( \frac{y-1}{x+1}+1)-y}{y\lceil y/x+1\rceil}=\frac{x}{y\lceil y/x+1\rceil}$.

Thus, $(x+1)\lceil y/x+1\rceil-y<x$, for which the result is already true by IH.

Then because any irrational (eg. decimal expansion) can be written as sum of rationals we are done. A possible problem here is repetition of reciprocals.

  • $\begingroup$ If $J \subset \mathbb{N}$ then you cannot write any irrational number in the sum of $\frac{1}{k}$ because sum of rational numbers will give you a rational number. $\endgroup$
    – gamma
    Sep 28 '15 at 17:05
  • 1
    $\begingroup$ J can be infinite eg. $e=0!+\frac{1}{1!}+\frac{1}{2!}+...$. $\endgroup$
    – user133100
    Sep 28 '15 at 17:07
  • $\begingroup$ How does your proof for rationals guarantee distinct denominators? $\endgroup$
    – Aravind
    Sep 28 '15 at 18:44

If $0<x<1$, then the reciprocals of the powers of 2 suffice. Otherwise, find $N$ such that $0<|x-\sum_{i=1}^{N}\frac{1}{(2i-1)}|<1$. So $J$ is the union of odd numbers and powers of 2.

  • $\begingroup$ what is that N explicitly or why does it exist? $\endgroup$
    – user133100
    Sep 28 '15 at 19:01
  • $\begingroup$ It exists because the sum of reciprocals of the odd numbers diverges. $\endgroup$
    – Aravind
    Sep 28 '15 at 19:02
  • $\begingroup$ Yeah, but why does it follow from that? Sum of reciprocal factorials go to infinity too. $\endgroup$
    – user133100
    Sep 28 '15 at 19:03
  • $\begingroup$ There's a $M$ such that $\sum_{i=1}^{M}\frac{1}{2i-1}>x$; hence there's a minimum such $M$; so let $N=M-1$. $\endgroup$
    – Aravind
    Sep 28 '15 at 19:14

I think this may work:

Let $x>0.$ Let $n_0$ be the largest integer $t$ for which $t\leqslant x.$ Having chosen $n_0,\ldots,n_m,$ let $n_{m+1}$ be the largest integer $t$ for which $\dfrac{n_0}{10^0}+\cdots+\dfrac{n_{m+1}}{10^{m+1}}\leqslant x.$ Then $x=\sum\limits_{m=0}^\infty\dfrac{n_m}{10^m}.$ Since every rational number is an egyptian fraction, then each $\dfrac{n_m}{10^m}$ is an egyptian fraction and the result follows.

  • $\begingroup$ They might be repetitions of egyptian fractions and so you might get integers. $\endgroup$
    – user133100
    Sep 28 '15 at 19:49
  • $\begingroup$ Then we express one of the repeated numbers as an egyptian fraction and so on until we have something "nice" (I'm being too "chatty" here of course) $\endgroup$
    – CIJ
    Sep 28 '15 at 19:56

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.