0:005:11

Recursos

Resumen

You can trace why, for every fixed C, a positive proportion of the first N prime gaps exceed C log p_n. You follow the paper's chain from a weight on two adjacent intervals to an actual count of consecutive gaps, and you see why that answers the Erdős–Prachar question about p_n/n.

The question and the corollary

Record-gap results (Westzynthius, Erdős–Rankin, Ford–Green–Konyagin–Maynard–Tao) make single gaps enormous. They say nothing about how often gaps are large. The theorem gives #{n ≤ N : p_{n+1} − p_n > C log p_n} ≥ c(C)N for every large N. [0:49] Erdős–Prachar: p_n/n < p_{n+1}/(n+1) holds exactly when p_{n+1} − p_n > p_n/n. Since p_n/n ~ log p_n, the case C = 2 is enough once finitely many indices are dropped.

Why counting empty intervals is not enough

[1:25] With h = 6, the single gap from 113 to 127 contains 8 starts of empty windows. A longer gap contains even more, so a sparse set of very long gaps can account for many empty windows. [1:55] The fix is to ask for a prime just before the empty stretch. Take h = ⌊λ log X⌋, a block J = (m, m+h] and the next block I = (m+h, m+2h]. The weight W(m) ≥ 0 has mass tending to 1, a prime count in J tending to λ (detects), a prime count in I of at most ε (suppresses), and a second-moment constant K_λ that does not depend on ε.

[2:32] W is a normalized square Z², where Z is a signed sum of smooth divisor sums over shift sets S ⊆ I. A prime in J lies outside every S, so the detector stays intact. A prime at m + a in I forces that divisor to equal 1. Deleting a then couples level j with level j + 1. With alternating signs, neighboring levels nearly cancel, so w → 1 and v → 0.

From weight to gaps

[3:25] Two Cauchy–Schwarz steps turn this into δX ordinary starts with a prime in J and none in I. For each such start, take p(m), the last prime in J. It begins a gap longer than h. [3:57] A prime p can be chosen only by starts with p − h ≤ m ≤ p − 1, so at most h starts choose it, however long its gap is. That gives at least δX/h ≥ δX/(λ log X) distinct primes. Choose λ > C and X = ⌊p_N/3⌋, and you get c(C) = δ/(4λ).

Rule to reuse: to get a gap count from an interval statement, pick one canonical prime per start and bound how many starts can choose it.

Positive density of large prime gaps

OpenAI Math Preprints

Clase 52 / 100

Ver clases