4

Problem Statement:Does there exist a subset of rational numbers "S" such that for each and every integer( positive,negative,0 ) "n" there is a unique nonempty finite subset of S such that sum of its elements is n?

i tried to disprove it using the fact that we could have a sum subset of an integer "n" and add zero ( or the integers used to form zero in the set "S" ) to it and the sum would be same , but the 2 subsets so formed wont be unique( not disjoint ) and hence my approach fails .

we didnt use the "finite" subset part , would that be used?

Asaf Karagila
  • 405,794
Shaktik
  • 41
  • Do you target positive and negative integers? – Christophe Boilley Feb 14 '25 at 16:07
  • Please clarify your specific problem or provide additional details to highlight exactly what you need. As it's currently written, it's hard to tell exactly what you're asking. – Community Feb 14 '25 at 16:08
  • yes , positive and negative integers – Shaktik Feb 14 '25 at 16:11
  • 1
    If it was only for positive integers, then the set $S={1,2,4,8,16,32,\dots}$ would suffice, as each integer has a unique binary representation. So your argument has to use that you require this for negative numbers and $0$ as well. – Daniel P Feb 14 '25 at 16:17
  • 1
    The problem statement should appear in the main body of the post, not just the title. – Greg Martin Feb 14 '25 at 17:36
  • 7
    The set $S={1,-2,4,-8,16,\dots}$ works for all nonzero integers. – Greg Martin Feb 14 '25 at 17:46
  • yes if we have a sum subset of zero call it "X" and if we have a sum subset for "n" which is disjoint from "X" then we could get 2 ways of making "n" namely 1- the sum subset of "n" 2- sum subset of "n" + X. hence uniqueness fails – Shaktik Feb 14 '25 at 17:52

2 Answers2

3

For all integer $k>0$, note $n_k=(-1)^k\left\lfloor\frac k2\right\rfloor$, so that $k\mapsto n_k$ is a bijection from $\mathbb N^*$ to $\mathbb Z$, and $$r_k=n_k-\sum_{i=1}^k\frac{1}{3^i}=n_k-\frac 12\left(1-\frac{1}{3^k}\right)$$ Then every integer can be written as a sum of a finite non-empty subset of $S=\left\{\frac{1}{3^k},k\in\mathbb N^*\right\}\cup\{r_k,k\in\mathbb N^*\}$.

Since $\sum_{k>0}\frac{1}{3^k}=\frac 12$, any such decomposition must involve some terms of the sequence $(r_k)$.

Let $A$ and $B$ two finite sets of $\mathbb N^*$ such that $\displaystyle\sum_{i\in A}r_i+\sum_{j\in B}\frac{1}{3^j}\in\mathbb Z$. Then we get $$\sum_{i\in A}\frac{1}{3^i}+2\sum_{j\in B}\frac{1}{3^j}\in\mathbb Z$$ By decomposition of a number in base 3, you get that $A$ cannot have other element than its maximum $k$, and $B$ must be equal to $[[1,k]]$. Then this is the canonical decomposition of $n_k$

  • Neat solution , what were your motivations for making that form – Shaktik Feb 15 '25 at 13:25
  • @Shaktik You must cumulate fractional parts to avoid non-overlapping sets. I tried harmonic series, inverse of factorials, inverse of primorials, then geometric sequences but ratio $1/2$ did not work. – Christophe Boilley Feb 15 '25 at 13:52
1

After a lot of fiddling around, I have found a set $S$ that doesn't work, and maybe we should try to disprove this instead (read the comments...):

$$S = S_0 \cup S_1 \cup S_2 \cup S_3 $$

where

$$\begin{align} S_0 &= \left\{\frac{1}{3},\frac{1}{5},-\frac{8}{15},\frac{1}{21},\frac{13}{21}\right\} \\ S_1 &= \bigcup_{k=5}^\infty \left\{\frac{1}{5p_{k}}\right\} \\ S_2 &= \bigcup_{k=2}^\infty \left\{k-\frac{1}{5}-\frac{1}{5p_{2k+2}}\right\} \\ S_3 &= \bigcup_{k=2}^\infty \left\{-k-\frac{1}{5}-\frac{1}{5p_{2k+3}}\right\} \\ \end{align}$$

where $p_k$ is the $k$'th prime number ($p_1 = 2, p_2 = 3, p_3 = 5, \dots$).


Constructing every integer

$$\begin{align} 0 &= \left(\frac{1}{3}\right)+\left(\frac{1}{5}\right)+\left(-\frac{8}{15}\right) \\ 1 &= \left(\frac{1}{3}\right)+\left(\frac{1}{21}\right)+\left(\frac{13}{21}\right) \\ \forall k \in \mathbb{N}^+ \setminus \{1\} \quad k &= \left(\frac{1}{5}\right) + \left(\frac{1}{5p_{2k+2}}\right) + \left(k - \frac{1}{5} - \frac{1}{5p_{2k+2}}\right)\\ \forall k \in \mathbb{N}^+ \quad -k &= \left(\frac{1}{5}\right)+\left(\frac{1}{5p_{2k+3}}\right) + \left(-k-\frac{1}{5}-\frac{1}{5p_{2k+3}}\right) \end{align}$$


Examples

$$\begin{align} 0 &= \frac{1}{3}+\frac{1}{5}+\left(-\frac{8}{15}\right) \quad &(\text{using }p_2=3, p_3=5)\\ 1 &= \frac{1}{3}+\frac{1}{21}+\frac{13}{21} \quad &(\text{using }p_4=7) \\ -1 &= \frac{1}{5}+\frac{1}{55}+\left(-\frac{67}{55}\right) \quad &(\text{using }p_5=11) \\ 2 &= \frac{1}{5}+\frac{1}{65}+\frac{116}{65} \quad &(\text{using }p_6=13) \\ -2 &= \frac{1}{5}+\frac{1}{85}+\left(-\frac{188}{85}\right) \quad &(\text{using }p_7=17) \\ 3 &= \frac{1}{5}+\frac{1}{95}+\frac{53}{19} \quad &(\text{using }p_8=19) \\ -3 &= \frac{1}{5}+\frac{1}{115}+\left(-\frac{369}{115}\right) \quad &(\text{using }p_9=23) \\ 4 &= \frac{1}{5}+\frac{1}{145}+\frac{110}{29} \quad &(\text{using }p_{10}=29) \\ &\ \ \vdots \end{align}$$

Daniel P
  • 2,832
  • Neat solution Daniel P , would you mind explaining your motivations behind how you found the forms for the numbers ? ( i need to know how i couldve come up with the solution myself ) once again , thank you. – Shaktik Feb 14 '25 at 19:04
  • First I noticed the $n=n+0$ problem. This means that every number has to have at least 1 term from the expansion of $0$. First I tried with whole numbers, but I quickly ran into problems. Namely that if the largest number in the expansion of $0$ is $x$, then if the expansion of $\lceil x \rceil+1$ uses a whole number $t$, then the expansion of $t$ can be $t$ or $t+0$, which isn't unique. I convinced myself with other examples that whole numbers just wouldn't work, so I tried the first non-trivial fraction in every expansion, $\frac{1}{3}$. Then I added unique fractions to each number, $1/p$. – Daniel P Feb 14 '25 at 19:09
  • @Pranay Edited, hopefully it works now. – Daniel P Feb 14 '25 at 19:48
  • does it still not work? – Shaktik Feb 14 '25 at 19:54
  • Corrected, thank you. – Daniel P Feb 14 '25 at 19:55
  • I am positive we'll eventually get a correct answer...! – Daniel P Feb 14 '25 at 19:56
  • Maybe one of the Egyptian fractions can add to one of the non-Egyptian fractions perhaps...? I just don't see how you could get a denominator with a completely different prime decomposition. – Daniel P Feb 14 '25 at 20:20
  • The $5$ instances of $x$ and $y$ used in your example have to have at least $5$ unique denominators with at least $5$ distinct prime factors, and with all of those factors being greater than $5$. So as far as I can see their sum cannot possibly equal something with denominator $5$, but correct me if I'm wrong. – Daniel P Feb 14 '25 at 20:33
  • Nevermind, I found the issue. For example, $$\frac{1}{55}-\frac{67}{55}+\frac{1}{65}+\frac{116}{65}+\frac{1}{85}-\frac{188}{85}+\frac{1}{95}+\frac{53}{19}+\frac{1}{115}-\frac{369}{115}=-2$$ – Daniel P Feb 14 '25 at 20:43
  • should we trying disproving instead? – Shaktik Feb 14 '25 at 20:44
  • I think we should try disproving instead :D – Daniel P Feb 14 '25 at 20:45
  • 1
    Yes, I get it now. – Daniel P Feb 14 '25 at 20:53