7

Let $\pi(x)$ be the number of primes not greater than $x$.

Wikipedia article says that $\pi(10^{23}) = 1,925,320,391,606,803,968,923$.

The question is how to calculate $\pi(x)$ for large $x$ in a reasonable time? What algorithms do exist for that?

J. M. ain't a mathematician
  • 75,051
falagar
  • 4,150
  • 3
    What's wrong with the ones given on the page? – Noldorin Jul 28 '10 at 11:34
  • 2
    There is only one algorithm there. And I don't think it's possible to calculate $pi(10^{23})$ using it. Also there are no bounds on it's complexity. – falagar Jul 28 '10 at 11:48

3 Answers3

12

The most efficient prime counting algorithms currently known are all essentially optimizations of the method developed by Meissel in 1870, e.g. see the discussion here http://primes.utm.edu/howmany.shtml

Bill Dubuque
  • 272,048
1

You can use inclusion exclusion principle to get a boost over the Eratosthenes sieve

Casebash
  • 9,211
  • I'm not sure what do you mean, but Eratosthenes sieve is way too slow for numbers like 10^23. Could you provide an asymptotic complexity of your approach? – falagar Jul 28 '10 at 11:23
1

The Sieve of Atkin is one of the fastest algorithm used to calculate $pi(x)$. The Wikipedia page says that its complexity is O(N/ log log N).

(edit)

I found a distributed computation project which was able to calculate $pi(4\times 10^{22})$, maybe it could be useful.

zar
  • 4,602
  • 4
    There is no way that $\pi(10^{23})$ could be calculated using the sieve of atkin (in any reasonable amount of time) – BlueRaja - Danny Pflughoeft Jul 28 '10 at 14:51
  • Right. This pages could be useful:

    http://www.ams.org/mcom/1996-65-213/S0025-5718-96-00674-6/S0025-5718-96-00674-6.pdf

    http://www.ieeta.pt/~tos/primes.html

     – zar Jul 28 '10 at 15:56
  • This is one of the papers linked to on Bill's page; they were able to use it calculate $\pi(10^{18})$, but more advanced refinements were needed to calculate higher than that – BlueRaja - Danny Pflughoeft Jul 28 '10 at 16:22