An Ignorant Guide to the Erdӧs-Straus Conjecture: Jason Hudson

You might also like

Download as docx, pdf, or txt
Download as docx, pdf, or txt
You are on page 1of 8

An Ignorant Guide to the Erdӧs-Straus Conjecture

Jason Hudson

This paper will explore the Erdӧs-Straus conjecture based primarily on my own experimentation instead
of through research, as I value the process of discovery. What follows are my thoughts on how the
problem is posed, some ways I found to calculate solutions, and my thoughts on how this conjecture
might be proven.
Introduction

The Erdӧs-Straus conjecture is stated simply thus:

4 1 1 1
For every integern ≥ 2, the equation = + + has positive integer solutions a, b, & c.
n a b c

4
This conjecture can synonymously be stated that for every n ≥ 2, the number can be stated as the sum
n
of three distinct unit fractions.

The conjecture has been verified up ton ≤ 108, as of the year 2000. (Hagedorn) It remains an open
problem, as the validity of the conjecture has not been conclusively proven for all n , but only for those n
that have been solved through exhaustive computer algorithms.

First Impressions

I decided to jump headfirst into the problem by finding the answers to simple questions that I hoped
would aid me in understanding the challenges the problem poses. I explored these questions from a
mostly ignorant perspective in order to allow myself to discover aspects of the problem myself.

I had several questions about the way the conjecture is posed that I decided to explore first, specifically
about the domain of n . Why does the conjecture stipulate that n ≥ 2? Why does n=1 not work? What
about the negative integers? What about n=0?

The latter question is most obvious. Integer n cannot equal zero because to do so would require
dividing 4 by zero, which is inadmissible.

To understand why negative numbers cannot be included in the domain of n I began to experiment with
possible solutions using both positive and negative integers for a, b, and c. I randomly chose several
values of n , and began to use trial and error to find solutions:

4 1 1 1
= + −
7 2 7 14

4 1 1 1
= + −
6 2 3 6

4 1 1 1
= + −
12 6 4 12

Based on the short amount of time it took me to find these solutions, and the relative ease with which I
found them, I supposed at the time that it is most likely much easier to find solutions to the problem if
you are allowed to add and subtract as you please.
So what about n=1? It turns out that this is an easy value to rule out. If n=1 we know that:

4 1 1 1
=4= + +
n a b c

1 1 1
If a, b, and c get infinitely large, + + gets infinitely small, and therefore will never come close to
a b c
reaching 4. Also, we know that the smallest positive integer is 1, which will give us the highest possible
1 1 1
value of + + . Even if we are allowed to use 1 as the value for a, b, and c, that sum at its maximum
a b c
can only equal 3. Since neither the highest possible values nor the lowest can get the equation to
balance, doing this operation for n=1 is impossible.

Developing a Method

After considering the previous question, it seemed logical to dive right in and see how easy or hard it
was to find solutions. So I began with n=5 because it seemed a reasonably small number to
experiment with. To my utter delight, this turned out to be a good first choice. I quickly found the
solution:

4 1 1 1
= + +
5 2 5 10

This confirmed what I had already assumed, that there was indeed a solution for n=5, as I would have
been utterly astonished if Erdӧs and Straus had not even tested the number 5.

At this point, I became curious as to whether it was possible for a single n to have more than one
solution. Not having much luck using guesswork and dumb luck this time, I began to devise a way that I
might be able to search for solutions in a systematic way. The following table shows the beginnings of
an exploration of solutions for n=5. My method involved systematically testing unit fractions, starting
1
with , then testing the difference between 4/n and that first fraction for solutions by means of a
2
common denominator. For instance, the method for the first row is as follows:

1 1 4 3
1.) The first unit fraction to test is , so subtract from to get .
2 2 5 10
2.) Take the denominator in the difference (10) and try to find two factors that add up to the
numerator – this ensures a unit fraction can be created
2 1
3.) 2+1=3 , so the final two unit fractions are and
10 10
4 1 1 1
4.) The final solution is = + + .
5 2 5 10

Even though using this method yielded a solution to n=5 on the first try, I continued searching for
solutions in the hope that it might be possible to find multiple solutions.
4/n for n=5 Chosen Fraction Difference Factors of Solution
(Multiplied to find 4 denominator
between and
a common den.) n
chosen fraction
4 1 5 8 5 3 2, 5, 1, 10 1 1 1
∗2 = − = + +
5 8 2 10 10 10 10 2 5 10
=
2 10

4 1 5 7 1, 3, 5, 15 none
∗3 =
5 12 3 15 15
=
3 15

4 1 5 11 1, 4, 5, 10, 20 1 1 1
∗4 = + +
5 16 4 20 20 20 2 4
=
4 20

4 1 5 15 1, 5, 25 none !!!!
∗5 =
5 20 5 25 25
=
5 25

15 1, 5, 10, 25, 50 1 1 1
∗2 + +
25 30 5 10 2
=
2 50

4 1 1 1
As the table shows, there exists at least one other solution for n=5, in which = + + , so , in
5 20 2 4
1
fact, it is possible for a single n to have multiple solutions. A slight complication arose when testing
5
1
for multiple solutions, however. From previous calculations, I knew that was part of a solution trio.
5
However, no solution could be found based on the factors of the common denominator 25. After some
15 2
fiddling, I discovered that multiplying by would allow me to use a larger denominator with more
25 2
factors, which ultimately allowed me to solve the problem in the same manner as before.

Thus, by using some sequential order of testing the values in the first column, second column, and then
the third column, I believe that it was likely that at least one solution could be found per n . One
drawback to this method is that the process only stops when a solution is found. So, if no solution is
found, one could keep doing calculations forever (This turned out to be a non-issue, as the numbers I
tested were well within the maximum number of n that solutions have been found for). A side affect
that arises from this issue is that the number of solutions cannot be counted, as there is no way of
exhaustively checking all possibilities. Despite being aware of these drawbacks, I decided the above
method could be useful to calculate at least one solution for a larger quantity of n . The following table
contains the computed values for 6 ≤ n ≤20 using the above method.

n 4 Chosen Difference Factors of Solution


n Fraction 4 denominator
between and
(Multiplied to n
find a common chosen fraction
den.)
6 4 1 6 2 1, 3, 4, 6, 12 none
∗2 =
6 8 2 12 12
=
2 12

4 1, 3, 4, 6, 12, 24 1 1 1
+ +
24 24 8 2

7 4 1 7 1 1, 2, 7, 14 none
∗2 =
7 8 2 14 14
=
2 14

2 1, 2, 7, 14, 28 none
28

4 1, 2, 7, 14, 28, none


56 56

8 1, 2, 7, 14, 28, 1 1 1
+ +
112 56, 112 112 16 2

8 4 1 8 0
∗2 =
8 8 2 16
=
2 16

4 1 8 4 1, 2, 3, 4, 6, 12, 1 1 1
∗3 = + +
8 12 3 24 24 24 24 8 3
=
3 24

9 4 1 9 3 1, 3, 9, 27 none
∗3 =
9 12 3 27 27
=
3 27

6 1, 3, 9, 27, 54 none
54
12 1, 3, 9, 27, 54, 1 1 1
+ +
108 108 36 12 3

10 4 1 10 4 1, 2, 3, 5, 6, … 1 1 1
∗3 = + +
10 12 3 30 60 60 20 3
=
3 30

11 4 1 11 4 1, 2, 3, … 1 1 1
∗3 = + +
11 12 3 33 132 132 44 3
=
3 33

12 4 1 12 4 1, 2, 3, … 1 1 1
∗4 = + +
12 16 4 48 48 48 16 4
=
4 48

13 4 1 13 3 1, 2, … 1 1 1
∗4 = + +
13 16 4 52 52 52 26 4
=
4 52

14 4 1 14 8 1, 2, 4, 7, 8, … 1 1 1
∗4 = + +
14 16 4 56 224 224 32 4
=
4 56

15 4 1 15 3 1, 2, … 1 1 1
∗4 = + +
15 16 4 60 180 180 90 4
=
4 60

16 4 1 16 20 1, 2, 4, 5, 10, 16 1 1 1
∗5 = + +
16 20 5 80 400 100 25 5
=
5 80

17 4 1 17 6 1, 5, … 1 1 1
∗5 = + +
17 20 5 85 170 170 34 5
=
5 85

18 4 1 18 4 1, 2, 3, … 1 1 1
∗5 = + +
18 20 5 90 180 180 60 5
=
5 90

19 4 1 19 6 1, 5, 19, … 1 1 1
∗5 = + +
19 20 5 95 570 570 114 5
=
5 95
20 4 1 20 4 1, 2, 3, … 1 1 1
∗6 = + +
20 24 6 120 120 120 40 6
=
6 120

Conclusions and Questions

This method is likely not the optimal method for computing solutions to this problem, however I think it
could be useful as a brute force method through which a computer program could go about verifying
the truth of the conjecture. Given enough time, I believe a program written based on this method
would be able to find solutions for every n for which a solution exists.

1
I found that was an incredibly good fraction to begin experimenting with, as doing so ensured that I
2
1
chose the largest possible fraction for each n . Moving down the table, you will notice that once n=8,
2
4
cannot will not even fit into the value . Farther down, a pattern begins to emerge in that every four
n
4
increasing values of n necessitate a change in the initial fraction. This will always be true due to the in
n
the problem.

Finding patterns in the solutions arranged via this method could be useful in proving this conjecture. For
instance, for every value of n one less than a multiple of four, there will always be a solution because
4
the difference between the initial fraction and always equals 1. This covers one quarter of the
n
possible values. If such a pattern could be found for each quarter, it could be very useful in proving the
conjecture. Perhaps with a more thorough exploration of how many iterations it takes to find an answer,
a pattern could be found that would make the results even more predictable. The results in the table
were surprising to me in that I did not expect the solutions to be so easily computed or so predictable.
Only once did I have to change the fraction I began with to find an answer, and that was before I realized
that the initial fraction would always change every 4 values of n . Results were then easily found by
doubling the difference column occasionally. I expected to have to go through several more iterations
for each n to find a solution.

Looking at the solutions from a larger table might make it possible for formulas to be created for certain
values of n . If you could find formulas could be logically proven to include all the domain of n , then it
could be proven that the conjecture is true. One possible direction to take this problem would be to
compute solutions with a computer and try to identify these types of patterns.

References
Hagedorn, Thomas R. "A Proof of a Conjecture on Egyptian Fractions." American Mathematical Monthly
107.No. 1 (2000): 62-63. Web. 3 Mar 2011. <www.jstor.org>.

You might also like