# Covering the primes by arithmetic progressions

Define the length of a set of arithmetic progressions of natural numbers $A=\lbrace A_1, A_2, \ldots \rbrace$ to be $\min_i | A_i |$: the length of the shortest sequence among all the progressions. Say that $A$ exactly covers a set $S$ if $\bigcup_i A_i = S$. Let $P'$ be the primes excluding 2.

What is the longest set of arithmetic progressions that exactly covers the primes $P'$?

In other words, I want to maximize the length of a set of such arithmetic progressions. Call this maximum $L_{\max}$.

$L_{\max} \ge 2$ because $$P' \;=\; \lbrace 3,5 \rbrace \cup \lbrace 7,11 \rbrace \cup \cdots \cup \lbrace 521,523 \rbrace \cup \cdots$$ Perhaps it is possible that $$P' \;=\; \lbrace 3, 11, 19 \rbrace \cup \lbrace 5, 17, 29,41,53 \rbrace \cup \lbrace 7,19,31,43 \rbrace \cup \cdots \;,$$ but I cannot get far with sequences of length $\ge 3$. (I know Green-Tao establishes that there are arbitrarily long arithmetic progressions in $P$, but I don't know if that helps with my question.)

I am number-theoretically naïve, and apologize if this question is nonsensical or trivial. In any case, I appreciate the tutoring!

Addendum. Although my question should be revised (as Idoneal suggests) in light of George Lowther's proof that 3 cannot be in a progression of length 4, George has shown that it is likely that $L_{\max}=3$ but certification requires resolving an open problem. So I've added the open-problem tag. Thanks for everyone's interest!

-
some general heuristics and explicit hard conjectures suggest that you may cover primes by (even disjoint) arithmetic progressions of arbitrary length $n$. Indeed, if you have already covered all primes up to $p$ (but not $p$), then there must exist an arithmetic progression $\{\p,p+d,p+2d,\dots,p+(n-1)d}$ for infinitely many $d$'s. So cover $p$ and proceed. –  Fedor Petrov Nov 20 '10 at 20:02
@Fedor: Your comment seems to definitively answer my question: $L_{\max}$ has no upper bound! I did not know the existence result you use. Thanks! –  Joseph O'Rourke Nov 20 '10 at 20:15
it is not a result, but rather widely open conjecture... –  Fedor Petrov Nov 20 '10 at 20:49
As Fedor suggests, the question "does every prime belong to an arithmetic progression of primes of length $d$," to which this question is equivalent, is very, very open for large $d$. So this question should probably be re-tagged "open-question"; that said this statistic is probably an interesting one for many other subsets of the integers. –  Daniel Litt Nov 20 '10 at 21:21
@Fedor, Daniel: Oh, I see, sorry for misunderstanding. –  Joseph O'Rourke Nov 20 '10 at 21:29

Despite the comments to the question (including mine), this is a bit easier than it seems at first sight. We can show that $L_{\max}=2$ or $3$. Almost certainly we have $L_\max=3$. However, determining which of these is actually the case seems to be beyond current technology, according to this MO answer "Are all primes in a PAP-3?".
Showing that, $L_{\max} < 4$ is easy. That is, not every odd prime is contained in an arithmetic progression of primes of length 4. More specifically, 3 is not contained in an arithmetic progression of length 4. Suppose that $\lbrace x, x+d, x+2d, x+3d \rbrace$ was such a progression for $d > 0$. Then $x\not=2$, otherwise we would have $x=2,d=1$, but $x+2d=4$ is not prime. So, $x=3$. But, then, $x+3d=3(1+d)$ is not prime.
Edit: Looking at $\tilde L_\max \equiv \max_A\liminf_i \vert A_i\vert$ might be more interesting. I expect that this is infinite but, again, showing that $\tilde L_\max > 2$ appears to be beyond current means.
No no, I meant what I said, but I've been mistaken, sorry. The natural conjecture is that $p$ lies in arithmetic progression of length $p$, but of course not more, by trivial reasons (modulo $p$). So, most probably we may cover all primes but finite number of them by arithmetic progressiond of arbitrary length. –  Fedor Petrov Nov 21 '10 at 6:58