# How to prove that the subsets of $\mathbb{N}$ that don't contain arithmetic progressions of some length form closed sets of a topology?

I have exactly the same problem as this person, which I will rewrite below:Topology and Arithmetic Progressions. The reason I'm posting this is that I'm stuck at a later stage than the OP of that post. So here's the problem:

Consider the following property of a subset $F$ of the set $ℕ$ of positive integers: there is $n ∈ ℕ$ such that $F$ contains no arithmetic progressions (APs) of length $n$. Prove that subsets with this property together with the whole $ℕ$ form a collection of closed subsets in some topology on $ℕ$.

You may need van der Waerden's theorem: For any $n \in \mathbb{N}$ there exists $N \in \mathbb{N}$ such that for all $A \subset \{1, \cdots, N\}$ either $A$ contains an arithmetic progression of length $n$ or $\{1, \cdots, N \} \setminus A$ contains an arithmetic progression of length $n$.

Here is my attempt:

It suffices to show that the subsets with the given property are closed under arbitrary intersection and finite union. For finite intersection, suppose that $\cap_i A_i$ is an intersection of sets with the given property. For every $i$, the set $A_i$ has an associated $n_i$ such that $A_i$ contains no APs of length $n_i$. Consider $n = min(n_i)$, we show that $\cap_i A_i$ does not contain APs of the length $n$. If it did, any such AP would be an AP for the set $A_i$ that corresponds to $n$, a contradiction. For finite union, by induction it suffices to show that the subsets are closed under binary unions.

For this suppose that $f_1$ and $f_2$ are subsets of $\mathbb{N}$ with the aforementioned property. Then we have associated $n_1, n_2 \in \mathbb{N}$ such that $f_1$ does not contain APs of length $n_1$ and ditto for $f_2$ and $n_2$. We choose $n = max(n_1, n_2)$. Without loss of generality we can assume $max(n_1,n_2) = n_1$, since the other case can be argued symetricallly. Suppose that $f_1 \cup f_2$ contained an AP of length $n$. This AP would have the form $\{a + (i - 1)d \ | \ i \in \{1, \cdots, n\} \}$ for some $a, d \in \mathbb{N}$. The indicies $i$ can be put into two sets $f_1'$ and $f_2'$ with the following two properties: Firstly, both $f_1'$ and $f_2'$ are subsets of $\{1, \cdots, n\}$, and we have $f_1' \cup f_2' = \{1, \cdots n\}$. Secondly, $f_1'$ is in bijection with $f_1$, where the forward map sends $i$ to its corresponding element of the AP namely $a + (n - 1)d$, and the same for $f_2'$ and $f_2'$. By van der Waerden's applied to $n$, there exists a $N \geq n$ such that either $f_1'$ contains an AP of length $n$, or $\{1, \cdots, N\}$ contains an arithmetic progression of length $n$. However the former case is impossible since $n = n_1$, so the AP in $f_1'$ would map to an AP in $f_1$ of length $n_1$ and we are given that this is not the case. So it suffices to consider the case where $\{1, \cdots, N\} \setminus f_1'$ contains an arithmetic progression of length $n$.

Here is where I'm stuck. How do I complete the proof? Hints are preferred to answers.

HINT: I would approach the union a bit differently. Let $f=f_1\cup f_2$, and suppose that $f$ is not closed; then it contains arbitrarily long arithmetic progressions. Suppose that $$\{a+kd:k=0,\ldots,\ell\}$$ is an arithmetic progession in $f$, where $\ell$ has yet to be determined (and remember that we can take $\ell$ to be as large as we want. Let $$I_1=\big\{k\in\{0,\ldots,\ell\}:a+kd\in f_1\big\}\;,$$ and let $I_2=\{0,\ldots,\ell\}\setminus I_1$. Show that if $I_1$ (I_2, respectively) contains an arithmetic progression of length $m$, then so does $f_1$ ($f_2$, respectively). Then use van der Waerden’s theorem to choose $\ell$ appropriately.

• Dear @Brian M. Scott, could you please check if I finished the proof correctly: mathb.in/42134. Sep 1, 2015 at 12:53
• I fixed a typo in the statement of van der Waerden's lemma: mathb.in/42136. Sep 1, 2015 at 13:24
• Clarification of an assumption I made, mathb.in/42137. Sep 1, 2015 at 13:37

If $F, G$ both have this property then take $n = n_F + n_G$, then $F\cup G$ has no arithmetic progressions of length $n$. Proof: if it did then it would need an arithmetic progression of length $m \geq n_G$ in $G$ or one of length $m \geq n_F$ in $F$, which is impossible. Think of gluing together finite arithmetic progressions to see this. Thus the subsets with this property are closed under finite union.

Now if $F = \bigcap_i F_i$, let $F_i$ be a subset with no arithmetic progressions of length $n_i$. Then take $n = \min_i\{n_i\}$ which exists for any set of natural numbers. $F$ has no arithmetic progressions of length $n$ since it's the subset of such a set. QED.

Clearly $\varnothing$ is such a set. Add in the whole space $\Bbb{N}$ artificially, and you then have a topology.