Bounded Gaps Between Primes
This presentation explores a breakthrough proof that consecutive primes are sometimes separated by at most 240 numbers. The work improves on the previous best bound of 246 by introducing a hybrid method that combines classical distribution theorems with newer techniques requiring special factorization structure. The key innovation lies not in the numerical improvement itself, but in a systematic relaxation of equidistribution conditions that enlarges the admissible support for sieve optimization, achieved through exact recursive computation of high-dimensional integrals and polynomial optimization in 49 variables.Script
For over a century, mathematicians have known that prime numbers thin out as they grow larger, yet no one could prove that arbitrarily large primes are ever separated by a bounded distance. In 2013, Zhang shattered that barrier, and the Polymath project pushed the bound down to 246. This paper takes it one step further, to 240.
The challenge is about moduli: Bombieri–Vinogradov gives you equidistribution up to the square root of x, while Zhang-type estimates reach higher but only for moduli with special smooth factors. The author's key idea is to stop choosing between them and instead combine both, creating a support that is part simplex and part smooth-structured.
Earlier proofs required the entire modulus to be smooth or densely divisible. This work observes that the underlying exponential-sum arguments only need divisors in certain explicit ranges. By extracting just the factorization properties that matter, the paper relaxes global smoothness into local divisor conditions, and that subtle shift opens up a wider parameter space.
The final proof comes down to integrating polynomial functions over a 49-dimensional region and checking that a certain quadratic form ratio exceeds one. The computation is exact, not floating-point. The author recursively decomposes the integrals into monomial-simplex pieces, builds coefficient matrices indexed by powers of the threshold delta, and then performs exact rational arithmetic to certify the result.
The paper uses a polynomial basis limited to degree 19, smaller than the degree 27 used in the previous proof, yet still achieves a tighter bound. The author explicitly notes that computational resources, not the method itself, imposed this limit. The same enlarged support could support higher-degree polynomials and likely push the bound even lower, but verifying that claim will require substantial computational effort.
The bound of 240 is the smallest possible improvement over 246 within the admissible-tuple framework, because it corresponds to the shortest tuple with 49 elements. The real contribution is not the six-number reduction, but a systematic way to enlarge the support by combining different equidistribution mechanisms without forcing every modulus into the narrowest class. If you want to explore how this method might extend further or build your own videos explaining research like this, visit EmergentMind.com.
