cp's OEIS Frontend

Subsequence of A024702 which considers all primes rather than only twins.

This sequence seems to play an important role in studying the twin prime conjecture; see also A057767, A273257, and related.

Dinculescu calls the numbers M(j) = (prime(j)^2 - 1)/6 "basic numbers", and [M(j), M(j+1)] a "twin interval" when j is the index of a twin prime. He notes that the length of such an interval equals four times the corresponding twin rank k(j) = (prime(j) + prime(j+1))/6, see near eq.(3.3) in the 2018 paper.

Also, a(n) is the number of unordered partitions of n into four terms of A024702.

a(n) > 0 for 4 <= n <= 2*10^4. Conjecture: a(n) > 0 for all n >= 4. A stronger conjecture: lim inf a(n) = +oo.

This is a quadratic analog of Goldbach's conjecture, asking for the smallest k such that any sufficiently large number congruent to k modulo 24 can be written as the sum of k squares of primes. k = 1 is trivially false. Let n = 49*t + 2 (t > 0), then 24*n + 2 = 24*49*t + 98, which is a multiple of 7^2. If p^2 + q^2 = 24*n + 2, since the squares of primes other than 7 are congruent to 1, 2, 4 modulo 7, we must have p = q = 7, but p^2 + q^2 < n. So k = 2 is false. Let n = 245*t + 103, then 24*n + 3 = 24*245*t + 2475, which is a multiple of 5. If p^2 + q^2 + r^2 = 24*n + 3, since the squares of primes other than 5 are congruent to 1, 4 modulo 5, we must have that at least one of p, q, r is 5. Suppose that p = 5, then q^2 + r^2 = 24*245*t + 2450, which is a multiple of 7^2. As is shown above, q = r = 7, but p^2 + q^2 + r^2 < n. So k = 3 is also false. On the other hand, if the case k = 4 is true, then all the cases k >= 4 are trivially true, because we can add as many 5^2 as needed. So the case k = 4 is the most interesting.

Numbers k such that 4*k+1 is a prime or the square of a prime congruent to 3 modulo 4.

If p is a Gaussian prime of norm 4*a(n)+1 (there are two up to association if a(n) is a prime, one if a(n) is the square of a prime), then for any Gaussian integer x, we have x^a(n) == 0, 1, i, -1 or -i (mod p) where i is a primitive fourth root of unity.

Numbers k such that 6*k+1 is a prime or the square of a prime congruent to 5 modulo 6.

If p is an Eisenstein prime of norm 6*a(n)+1 (there are two up to association if a(n) is a prime, one if a(n) is the square of a prime), then for any Eisenstein integer x, we have x^a(n) == 0, 1, w, w^2, -1, -w or -w^2 (mod p), where w = (1+sqrt(-3))/2 is a primitive sixth root of unity.

Results for p = q are given in A024702, which is complementary.

All integer results when viewed in the triangle occur in loosely diagonal, interrupted "bands" roughly (or exactly) parallel to main diagonal, such that q - p = 24m, where m = 1 for the first band closest to the main diagonal, m = 2 for the second band, m = 3 for the third band, etc. The main diagonal p = q can be considered as fitting in this pattern where m = 0.

A general "rule" can be stated: If q-p = 24m for any m >= 0 and primes p < q, then p*q-1 is divisible by 24. This follows algebraically from the known "rule" that p^2 - 1 is divisible by 24 for any prime p > 3 as given in A024702.

No result will occur twice, even when including A024702, because the product of any two primes is unique within the set.

Integer results have a density of about 12% to 13% for all possible p,q pairs among the first few hundred primes.

Sequence motivated by a comment in A024702: "The set of prime factors of a(n), however, appears to include all primes".

a(n) = 0 for n: 44, 63, 71, 80, 89, 95, 97, 108, 118, 122, 132, 141, 150, etc. Robert G. Wilson v, Feb 15 2017

The main entry is A323015, which is the unordered version.

Also, a(n) is the number of ordered partitions of n into four terms of A024702.

a(n) > 0 for 4 <= n <= 2*10^4. Conjecture: a(n) > 0 for all n >= 4. A stronger conjecture: lim inf a(n) = +oo.

Chowla conjectures that all numbers are the sum of no more than four terms of the form (p^2-1)/24 where p is a prime > 4.

Define "density" of a(n) as D(n) = sum(i=1 to n, a(n))/n. Define "average size of a hit" as H(n) = sum(i=1 to n, a(n))/sum(i=1 to n, b(n)) where b(n) = 0 if a(n) = 0, and b(n) = 1 if a(n) > 0. While D(n) declines from a maximum of around 88% at n = 17, to 46.6% at n= 1742, and down to 27.8% at n =~ 40000. Whereas H(n) increases from 1.0 to a maximum of 1.3904 at n = 1742 and then declines slowly to about 1.356 at n =~40000. This shows a strong increase in the "clustering tendency" of these sums onto particular values of n up through n = 1741, and strong persistence of that tendency even as the density declines significantly at large n.

The "hit density" of a(n), defined as sum(i=1 to n, b(n))/n, reaches its maximum of 80% at n = 10 and declines to 20.51% at n = 40,000 as it continues to fall almost steadily in that range and likely to continue.

a(n) reaches values of 8 at n = 3407, 15392, 18282, 32817, 37337 (for n<=40000), which is the highest value of a(n) in this range.

A persistent tendency for more frequent large values of a(n) for n> 40000 is conjectured, with the likelihood that 8 is NOT the maximum value, and the possibility that ever larger values can always be found at higher n.