1
$\begingroup$

This Q. is an extension of Egyptian representations of $1$

Let $\ \emptyset\ne A\subseteq\mathbb N\ $ be a finite set. Then let

  • $\ \ E_n\ :=\ \{ A :\ \sum_{x\in A} \frac 1x \ =\ 1\quad\&\quad \min\,A\ =\ n\} $
  • $\ \ M_n\ :=\ \{A\in E_n :\ |A|\ =\ \min(|B|:\ B\in E_n)\} $

Questions:

  • What is the minimal $\ n\ $ such that $\ |M_n|>1\ ?$
  • Let $\ \mu_n:=\max( |M_k|:\ k=1\ldots n).\ $ What growth does function $\ (\mu_n)\ $ exhibit?
  • How often $\ |M_n|>|M_{n+1}|\ ?$

PS. It goes without saying that everybody--be it a pedantic professor or a crazy computer hacker--is welcome to provide computer computations and/or info from the literature, encyclopedias, etc. (You never know, your reputation may gain dramatically :) ).

EDIT:

$$ 1\ \ =\ \ \frac 13\ +\ \frac 14\ +\ \frac 15\ + \frac 16\ +\ \frac 1{20} $$

is the unique shortest of its type, $\ F_3=5.\ $ Thus,

$$ |M_3|=1 $$

and the requested minimal $\ n\ $ with $\ |M_n|> 2\ $ is GREATER than 3 (against a comment below).

$\endgroup$
14
  • $\begingroup$ Anything wrong with the mathematics of my question? $\endgroup$
    – Wlod AA
    May 26, 2018 at 5:29
  • $\begingroup$ I should think n=3 is minimal. I imagine the growth is exponential, with inequality occurring more often than not. Gerhard "Consider Bin Packing Harmonic Series" Paseman, 2018.05.25. $\endgroup$ May 26, 2018 at 5:29
  • $\begingroup$ @GerhardPaseman Gerhard, that $\ n\ $ is greater than 3. ***** Gerhard, you didn't downvote my question, did you? $\endgroup$
    – Wlod AA
    May 26, 2018 at 5:39
  • $\begingroup$ @GerhardPaseman and others, sure, I will remove my Q (or you guys can do it as some love to do it), no big deal--it's the same old tired story. $\endgroup$
    – Wlod AA
    May 26, 2018 at 5:41
  • 1
    $\begingroup$ I am finding 2 candidates for $M_4$, both with 8 terms. How about you? Gerhard "Not Much Room To Maneuver" Paseman, 2018.05.27. $\endgroup$ May 27, 2018 at 23:03

2 Answers 2

1
$\begingroup$

This time I got some numerical data by writing a simple Perl program (I am willing to add its code below if asked). Within their scope, the provided answers are complete.


$F_2=\mathbf 3\ $ as illustrated by unique $\ 1\ =\ \frac 12+\frac 13+\frac 16$. There are also additional $6$ expressions of $4$ summands starting with $\frac 12$.
$F_3=\mathbf 5\ $ as llustrated by unique $\ 1\ =\ \frac 13+\frac 14+\frac 15+\frac 16+\frac 1{20}.\ $ There are also $27$ expressions of $6$ summands (starting with $\frac 13$).
$F_4=\mathbf 8,\ $ where there are $77$ expressions of $8$ summnads. One could say that there are so many of them because an expected shorter expression is missing--it feels like these minimal expressions are not minimal. Can one turn this intuition into a theorem?
$F_5=\mathbf{10},\ $ where there are $161$ minimal length expressions. Again $\ \frac{F_n}n = 2\ $ (which seems to be high).
$F_6=\mathbf{11}.\ $ This time $\ \frac{F_n}n < 2,\ $ and the number of different minimal expressions is only $4$.
$F_7=\mathbf{13},\ $ and there are $\ 4\ $ minimal expressions. The situation is similar as for $\ F_6$.
$F_8=\mathbf{15}\ $ is served by $\ 19\ $ minimal expressions. It's an inbetween case--indeed, $\ \frac{F_8}8<2\ $ but by now, it's quite close to $\ 2.$

Etc. Have fun. I can copy in more specific results like some actual expressions, etc.

$\endgroup$
1
  • $\begingroup$ One can show with simple inequalities that a 7 term sum adding up to at least 1 and starting with 1/4 must have 1/5, at least one of 1/6 and 1/7, and if not both 1/6 and 1/7, at least one term strictly larger than 1/10. If all of 1/4 through 1/7 are included, one needs a fraction at least 1/11 in size. This gives 4 cases of four fractions to extend, and 4 cases of five fractions to extend. Gerhard "Can You Extend To Completion?" Paseman, 2018.05.30. $\endgroup$ May 30, 2018 at 15:26
0
$\begingroup$

I'll prove, step by step, $$ |M_n|>1\quad\Rightarrow\quad n>3 $$

-

THEOREM 1 $$ F_3 > 4 $$

PROOF   Let $\ 3<a<b<c$ be integers. Then

$$ \frac 13+\frac 1a+\frac 1b+\frac 1c\,\ \le \,\ \frac 13+\frac 14+\frac 15+\frac 16\,\ = \,\ \frac{19}{20}\,\ <\,\ 1 $$ END of Proof


THEOREM 2 $$ F_3=5 $$

PROOF $$ \frac 13+\frac 14+\frac 15+\frac 16+\frac 1{20}\,\ =\,\ 1 $$ END of Proof


THEOREM 3   If integers $\,\ 3<a<b<c<d\,\ $ are such that $$ \frac 13+\frac 1a+\frac 1b+\frac 1c+\frac 1d\ =\ 1 $$

then $$\ (3\,\ a\,\ b\,\ c\,\ d)\ \ =\ \ (3\,\ 4\,\ 5\,\ 6\,\ 20)$$.

PROOF   Let $\ (a\ b\ c\ d)\ $ be as above except for $$ (a\ b\ c\ d)\ \ne\ (4\ 5\ 6\,\ 20).\ $$

Since the respective Egyptian sum is still $\ 1,\ $ we get $\ d<20.\ $

Call integers $\ 3\ a\ b\ c\ d,\ $ to be seds (sed = a selected denominator). No sed can be a prime $\ p\ $ such that $\ p>\frac {19}2)\ $ (because there cannot be only one sed which divides $p$).

Next, $\ 7\ $ is not among seds because the only other possible multiple of $\ 7\ $ among seds would be $\ 14.\ $ However the denominator of the sum

$$ \frac 17+\frac 1{14}\ =\ \frac 3{14} $$

is divisible by $\ 7.\ $ Thus, $\ 7\ $ is out. Then also $\ 14\ $ is out because it'd be the only sed divisible by $\ 7.$

Also, the only potential seds divisible by $\ 5\ $ are $\ 5\,\ 10\,\ 15,\ $ while there can be no sed divisible by $\ 5\ $ without no other seds divisible by $\ 5.\ $ More generally, every but one Egyptian sum of inverses of seds divisible by $\ 5\ $ has its (reduced) denominator divisible by $\ 5:$

  • $\ \frac 15+\frac 1{10}\ =\ \frac 3{10} $
  • $\ \frac 15+\frac 1{15}\ =\ \frac 4{15} $
  • $\ \frac 1{10}+\frac 1{15}\ =\ \frac 16\qquad $ (the only exception!)
  • $\ \frac 15+\frac 1{10}+\frac 1{15}\ =\ \frac {11}{30} $

In the above exceptional case we have:

$$ \frac 13+\frac 1a+\frac 1b+\frac 1c+\frac 1d\,\ \le\,\ \frac 13+\frac 14+\frac 16+\frac 1{10}+\frac 1{15}\,\ = \,\ \frac{11}{12}\,\ <\,\ 1 $$

Thus, after all, $\ 5\ $ and its multiples are out as seds. This leaves potential seds to by divisible only by $2$ and $3$, and by no other primes. Hence

$$ \frac 13+\frac 1a+\frac 1b+\frac 1c+\frac 1d\,\ \le\,\ \frac 13+\frac 14+\frac 16+\frac 18+\frac 1d\,\ = \,\ \frac 78+\frac 1d\,\ <\,\ 1 $$

(because $\ d>8$).   END of Proof

Brute force? Can you do it simpler? (... so that question $\ |M_4|\ =\ 1\ $ would be a snap?)

$\endgroup$

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.