MathLabs

Worked solution: Maynard's multidimensional sieve: bounded gaps between primes (2013)

Step 5 of 8: A single number MkM_k decides everything: the variational problem
In plain words

Since S2S_2 is roughly proportional to ∑mJk(m)(F)\sum_m J_k^{(m)}(F) and S1S_1 to Ik(F)I_k(F), the ratio between them — maximized over all reasonable choices of FF — becomes the single number that decides how many primes among n+h1,…,n+hkn+h_1,\dots,n+h_k can be forced to appear. Call this best possible ratio MkM_k.

The bigger MkM_k is, the more primes Maynard's method can guarantee inside a bounded-length window of size kk. Everything now hinges on how large MkM_k can be shown to be for suitable kk — a calculus-of-variations question with no primes in sight anymore.

Mk=sup⁡F∈Sk∑m=1kJk(m)(F)Ik(F),rk=⌈θMk2⌉M_k=\sup_{F\in\mathcal{S}_k}\frac{\sum_{m=1}^{k} J_k^{(m)}(F)}{I_k(F)},\qquad r_k=\Big\lceil\frac{\theta M_k}{2}\Big\rceil
Detailed analysis

For a fixed level of distribution θ>0\theta>0 and small δ>0\delta>0, Maynard defines Mk=sup⁡F∈Sk(∑m=1kJk(m)(F))/Ik(F)M_k=\sup_{F\in\mathcal{S}_k}\big(\sum_{m=1}^k J_k^{(m)}(F)\big)/I_k(F), the supremum taken over the class Sk\mathcal{S}_k of suitable smooth functions FF (Maynard 2013, §4, Proposition 4.2). Setting ρ=θMk/2−ϵ\rho=\theta M_k/2-\epsilon in the counting sum S=S2−ρS1S=S_2-\rho S_1 from Step 2, a short computation with the asymptotics from Step 4 shows S>0S>0 for all large NN, so — following the logic of Step 2 — infinitely many nn have at least rk=⌈θMk/2⌉r_k=\lceil \theta M_k/2\rceil of n+h1,…,n+hkn+h_1,\dots,n+h_k prime.

This is Maynard's Proposition 4.2, and it is the crux of the whole method: bounded prime gaps of size rkr_k inside an admissible H\mathcal{H} of length kk follow automatically once one exhibits, for that kk, a smooth FF making the ratio (∑mJk(m)(F))/Ik(F)\big(\sum_m J_k^{(m)}(F)\big)/I_k(F) large enough. In particular rk≥2r_k\ge2 (two or more primes among the kk shifts) already gives a bound on lim inf⁡n(pn+1−pn)\liminf_n(p_{n+1}-p_n), namely the diameter of any admissible kk-tuple realizing it.

So the entire analytic-number-theory problem has been converted into: find, for a suitable kk, a good enough FF. The next step supplies exactly that, via explicit test functions and numerical optimization.

Terms in this step
Variational problem
A problem of finding the function (here, FF of kk variables) that maximizes or minimizes some quantity built from it (here, the ratio ∑mJk(m)(F)/Ik(F)\sum_m J_k^{(m)}(F)/I_k(F)), rather than finding a single optimal number.