19
$\begingroup$

Let $S = \{1, 1/2,1/3,\dots\}$

Can we find a partition $P$ of $S$ so that each part sums to 1, e.g. $$P_1 = {1}$$

$$P_2 = { 1/2,1/5,1/7,1/10,1/14,1/70}$$

$$P_3 = {1/3,1/4,1/6,1/9,1/12,1/18}$$

$$P_4 = \{... \}$$

I think I could go on, technically, but it got increasingly complex.

$\endgroup$
3
  • 2
    $\begingroup$ Unfortunately, $1/2+1/5+1/7+1/10+1/14+1/70=1+1/35$. $\endgroup$
    – Rosie F
    Commented Nov 8, 2023 at 6:58
  • 2
    $\begingroup$ You only asked for a partition of $S$, which could be a partition into finite and infinite sets, but I think you meant to ask for a partition into finite sets. It would be easy to partition $S$ into infinite sets, each of which sums to $\pi$. $\endgroup$
    – bof
    Commented Nov 8, 2023 at 11:28
  • 1
    $\begingroup$ Why don't you take $P_2=\{1/2,1/3,1/6\}$? $\endgroup$
    – user
    Commented Nov 8, 2023 at 17:35

1 Answer 1

24
$\begingroup$

The answer is yes!

Given already chosen disjoint subsets $P_1,\dots,P_k$ of $S$, here’s how to choose the next one $P_{k+1}$:

  • Let $m$ be the smallest positive integer whose reciprocal is not already in $P_1\cup\cdots\cup P_k$.
  • Let $\dfrac1N$ be the minimum element of $P_1\cup\cdots\cup P_k$ (the one with the largest denominator), and let $n$ be the largest integer such that $\displaystyle \sum_{k=N+1}^{N+n} \frac1k < 1 - \frac1m$.
  • The standard greedy algorithm for Egyptial fractions will produce a representation of the form $\displaystyle 1 - \frac1m - \sum_{k=N+1}^{N+n} \frac1k$ as $\dfrac1{d_1} + \cdots + \dfrac1{d_\ell}$ where the integers $d_1,\dots,d_\ell$ are distinct and (necessarily) larger than $N+n$.
  • Then set $P_{k+1} = \biggl\{\dfrac1m, \dfrac1{N+1},\dfrac1{N+2}, \dots, \dfrac1{N+n}, \dfrac1{d_1}, \dfrac1{d_2}, \dots, \dfrac1{d_\ell}\biggr\}$.

Always including the smallest integer $m$ available ensures that the union of the $P_j$ will equal all of $S$; choosing the other denominators of $P_{k+1}$ to be larger than all the denominators in $P_1\cup\cdots\cup P_k$ ensures that the $P_j$ are pairwise disjoint, so that this is a true partition.

$\endgroup$
5
  • $\begingroup$ I'm confused when you say "Let $N$ be the maximum element of $P_1 \cup \cdots \cup P_k$. These are reciprocals of integers, but it seems like you intend $N$ to be an integer. $\endgroup$
    – heropup
    Commented Nov 7, 2023 at 22:39
  • $\begingroup$ @heropup I assume Greg meant the smallest element, i.e., the reciprocal with the largest integer. $\endgroup$ Commented Nov 8, 2023 at 0:00
  • $\begingroup$ @EricSnyder I see. but if $1/N$ is the minimum element of $P_1 \cup \cdots \cup P_k$, but also $1/N \in P_{k+1}$ as stated in the last bullet, then this is not a partition of $S$. $\endgroup$
    – heropup
    Commented Nov 8, 2023 at 0:10
  • $\begingroup$ In that case, I think we simply remove $1/N$ from $P_{k+1}$ and have $\{1/m, 1/(N+1), 1/(N+2), \cdots \}$ and so forth. $\endgroup$ Commented Nov 8, 2023 at 1:39
  • 1
    $\begingroup$ Thanks for catching typos; hopefully this updated version no longer has any. $\endgroup$ Commented Nov 8, 2023 at 2:10

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .