The spread of low k´s (1 up to 13) of Mersennes
 

I have made a statistic of the spread of low k`s.

I see that the low k´s are getting rare, when the exponents are getting bigger.

And every k=12*c is more common than others:

[url]https://mersenneforum.org/showpost.php?p=382174&postcount=3[/url]

[URL="http://www.hoegge.dk/mersenne/mersennefactors1.png"]mersenne/mersennefactors1.png[/URL]

I'd been meaning to look at this for a while. I finally got around to it.

In the present situation, we want p and q = 2*k*p + 1 both to be prime. The Bateman-Horn Conjecture says there is a constant C = C(k) such that the number of such p up to X is asymptotically

C*x/log[sup]2[/sup](X)

We also require q == 1 or 7 (mod 8). This introduces a "fudge" factor, giving

C*x/log[sup]2[/sup](X), for k == 0 (mod 4),

(1/2)*C*x/log[sup]2[/sup](X), for k == 1 (mod 4),

0, for k == 2 (mod 4), and

(1/2)*C*x/log[sup]2[/sup](X), for k == 3 (mod 4).

The restriction q == 1 or 7 (mod 8) means that 2 is a quadratic residue (mod q). Also, since k|(q-1), the finite field [b]F[/b][sub]q[/sub] contains q q-th roots of unity. We would expect that, among primes q having 2 as a quadratic residue, and q == 1 (mod k), 2 would be a 2k-th power residue of 1/k of them.

If we [i]assume[/i] this proportion holds under the restriction that (q-1)/(2*k) is [i]prime[/i], we obtain a conjectural formula for the number of q up to X for which q = 2*k*p + 1 is a prime which divides M[sub]p[/sub]. Obviously, the factor 1/k reduces the constant multiplier if k > 1.

The formula for C is a bit complicated to go into here, but there is one aspect that may pertain to the value k = 12 being "favored."

If l is a prime not dividing 2*k, about 1/(l-1) of the primes p will be in the residue class -(2*k)[sup]-1[/sup] (mod l); l divides q = 2*k*p + 1 for these p. Thus, if 3 does not divide k, we can expect about half the values of 2*k*p + 1 to be divisible by 3. If 3 divides k, however, this possibility is eliminated.

The value k = 12 is thus thrice blessed -- it is divisible by 4, making the "fudge factor" as large as possible; the factor 1/k is not terribly small; and, the factor 3 eliminates the possibility of q being divisible by 3, which helps make the constant C bigger than it otherwise would be.

Someone more ambitious than I am might like to compute values of C for some given k's, and perhaps check how well the assumed factor 1/k fits the data.