## Erdösproblem 3, part II

I checked my dictionary, but apparently there is no english word for the dutch word ‘tangconstructie’. Anyway, remember how I, rhetorically, asked ‘how else could it be?’ yesterday? Well, I am more than delighted to say that I am able to give a counterexample to the metaconjecture that every proof of a counterexample to the question of Erdös and Graham I talked about in the previous post, works via a case-by-case-argument.

Consider the set $X = \{2,3,4,5,6,8,9,10,12,15,16\}$ and let $lcm\{S\}$ be the least common multiple of all members of $S$, for a set $S$ of positive integers. Then $\displaystyle \sum_{x_i \in X} \dfrac{1}{x_i} = 2 - \dfrac{1}{lcm\{X\}}$. Now, assume we have a partition of $X$ in two sets $A$ and $B$ with $\displaystyle \sum_{a_i \in A} \dfrac{1}{a_i} < 1$ and $\displaystyle \sum_{b_i \in B} \dfrac{1}{b_i} < 1$. Then in particular we have:

$2 - \dfrac{1}{lcm\{X\}}$
$= \displaystyle \sum_{x_i \in X} \dfrac{1}{x_i}$
$= \displaystyle \sum_{a_i \in A} \dfrac{1}{a_i}$ $+$ $\displaystyle \sum_{b_i \in B} \dfrac{1}{b_i}$
$\le 1 - \dfrac{1}{lcm\{A\}} + 1 - \dfrac{1}{lcm\{B\}}$
$\le 2 - \dfrac{2}{lcm\{X\}}$

And boom goes the dynamite.