13

Using PrimeQ in Mathematica 10 on integers up to $2\cdot 10^{5717}$ the function appears to work. The Documentation for Mathematica 5 says that PrimeQ is only good for integers up to $10^{16}$. Is there a definitive statement about the limit for PrimeQ implemented in Mathematica 10?

J. M.'s missing motivation
  • 124,525
  • 11
  • 401
  • 574
Frank M Jackson
  • 131
  • 3
  • 8
    As far as I know, the test used by PrimeQ has been proved correct for integers up to $2^{64}$. Also, no pseudoprime (a composite number passing the test) of any size has ever been found. – ilian Apr 27 '16 at 18:55
  • 2
    See http://math.stackexchange.com/questions/123465/do-we-really-know-the-reliability-of-primeqn-for-n1016 – Vaclav Kotesovec May 05 '16 at 15:14

1 Answers1

14

Turning my comment into an answer,

One of the tests performed by PrimeQ for machine-sized integers, namely Miller-Rabin using up to the first 12 primes as bases (as of version 10) has been proved correct for integers up to $2^{64}$ (in fact, the smallest number which that test falsely declares a prime is known to be $3186 65857 83403 11511 67461.$)

Of course,

PrimeQ[318665857834031151167461]

(* False *)

since it is rejected by a Lucas test (which is performed after a Miller-Rabin test with bases 2 and 3).

No pseudoprime (a composite number passing both tests) of any size has ever been found.

ilian
  • 25,474
  • 4
  • 117
  • 186
  • Let me ask a subsidiary question. If I tested a sequence of integers up to $2\cdot 10^{5717}$ for which PrimeQ reported (* False *) can I state that the next prime member of this sequence (if it exists) is greater than $2\cdot 10^{5717}$? – Frank M Jackson May 07 '16 at 07:36
  • 10
    Yes, the non-determinism in these tests only applies to primality: while a True result may happen to be wrong (with extremely small probability and no known examples), a False result is a guarantee of compositeness (assuming, of course, there are no bugs in the implementation, cosmic radiation doesn't flip random bits in the computer's memory etc.) – ilian May 07 '16 at 18:28
  • No, it fails already on number 9 and on number 2047. The number you mentioned is much further. https://oeis.org/A006945 – Валерий Заподовников Apr 14 '23 at 06:05
  • @ВалерийЗаподовников The number I mentioned, as explicitly stated in the answer (and in the linked paper) is the smallest strong pseudoprime to the first 12 prime bases. So when you say "it fails already on number 9 and on number 2047", I think there is some confusion as to what exactly is "it". – ilian Apr 14 '23 at 19:10
  • @ilian ok. BTW, the test uses 2020's update Baillie-PSW now, see: https://mathematica.stackexchange.com/a/283692/82985 – Валерий Заподовников Apr 15 '23 at 04:43