Have not been able to get an answer to this on http://math.stackexchange.com, so trying here too...

Given the following two sets:

  • $P^-(n) = \{p \leq n : p \equiv -1\pmod 6\}$
  • $P^+(n) = \{p \leq n : p \equiv +1\pmod 6\}$

For example:

  • $P^-(40) = \{5,11,17,23,29\}$
  • $P^+(40) = \{7,13,19,31,37\}$

Given the following two functions:

  • $C^-(n)=|P^-(n)|$
  • $C^+(n)=|P^+(n)|$

For example:

  • $C^-(40) = 5 $
  • $C^+(40) = 5 $


  1. Has it been proved that $\forall k \exists n : k=|C^-(n)-C^+(n)|$?

  2. Has any bound been proved for $|C^-(n)-C^+(n)|$ relatively to $n$ (e.g., $\ln \ln n$)?

  3. What is the largest known value of $|C^-(n)-C^+(n)|$, and for what value of $n$ does it hold?

  • 5
    $\begingroup$ You might find this paper interesting dms.umontreal.ca/~andrew/PDF/PrimeRace.pdf $\endgroup$ May 12, 2014 at 7:28
  • $\begingroup$ @Jonah Sinick: Thanks. I have already looked into the Prime Races subject, and found out that it's mostly about prime numbers of the form $4n\pm1$ (not $6n\pm1$ as in my question). Have I missed anything? $\endgroup$ May 12, 2014 at 7:50
  • $\begingroup$ I was going to link to the same paper. It talks about primes which are $3n+1$ vs $3n+2$ as well. $\endgroup$ May 12, 2014 at 8:08
  • 6
    $\begingroup$ You got a very good answer on m.se, you just don't know one when you see it. $\endgroup$ May 12, 2014 at 9:20
  • $\begingroup$ The m.se question: math.stackexchange.com/questions/787641/… $\endgroup$ May 12, 2014 at 10:17

1 Answer 1


Assume that the Riemann hypothesis for the non-principal $L$-series $\pmod{3}$ is false, say, this series has a zero $\rho=\sigma+i\gamma$ with $\sigma>1/2$. Then Turan and Knapowski have shown that both $C^-(n)-C^+(n)>n^{\sigma-\epsilon}$ and $C^-(n)-C^+(n)<-n^{\sigma-\epsilon}$ happen infinitely often.

If the Riemann hypothesis holds true, then $\frac{C^-(e^t)-C^+(e^t)}{e^{t/2}}$ is (up to a small correction) almost periodic, hence if there exists a sufficiently large value for $n$ such that $C^-(n)-C^+(n)>0.01\sqrt{n}$, say, then there are infinitely many $n$ such that $C^-(n)-C^+(n)>0.0099\sqrt{n}$, and similarly in the other direction.

Hence to prove that $C^-(n)-C^+(n)$ is unbounded in both directions one has to compute the corrections to make the difference almost periodic, and look for $n$ which make the difference large. You can model the proof on Kaczoroswki's article "On the Shanks-Rényi Race Problem mod 5", J. Number Theory 50, 106-118.

If you assume the Riemann hypothesis together with linear independence of the imaginary parts of the zeros of the $L$-series, then the various summands in the explicit formula behave like independent random variables, and you immediately get that $\limsup\frac{C^-(n)-C^+(n)}{\sqrt{n}}=\infty$, $\liminf\frac{C^-(n)-C^+(n)}{\sqrt{n}}=-\infty$.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.