MathLabs
Step 3 of 5: Major arcs: explicit LL-function zeros, verified by computer
In plain words

Near a rational a/qa/q, S(α)S(\alpha) is governed by how primes distribute in arithmetic progressions mod qq, which classically ties to the zeros of Dirichlet LL-functions L(s,χ)L(s,\chi): the fewer zeros near the line Re(s)=1\mathrm{Re}(s)=1, the more evenly primes are spread and the sharper the major-arc estimate. Rather than rely on classical zero-free regions (which give correct qualitative behavior but hide large, impractical constants), Helfgott used rigorous, computer-verified tables of LL-function zeros (work of David Platt) to get fully explicit, unconditional bounds — with no need to assume the Generalized Riemann Hypothesis, even though GRH would make the argument shorter.

S(α)≈μ(q)ϕ(q)∑χ mod qχˉ(a) ψ(x,χ),controlled via explicit zero-free regions for L(s,χ)S(\alpha) \approx \frac{\mu(q)}{\phi(q)}\sum_{\chi \bmod q} \bar\chi(a)\, \psi(x,\chi), \qquad \text{controlled via explicit zero-free regions for } L(s,\chi)
Detailed analysis

A second technical innovation on the major arcs is Helfgott's use of Gaussian smoothing weights η(t)=e−t2/2\eta(t)=e^{-t^2/2} in place of the sharp cutoffs used in classical treatments; the Mellin transform of a Gaussian-weighted exponential sum brings in parabolic cylinder functions, for which Helfgott had to derive new fully explicit bounds via the saddle-point method, since none were available in the literature. Combining the explicit LL-function zero data with these smoothed estimates yields a major-arc contribution matching the expected singular series S(n)⋅n2/2\mathfrak{S}(n)\cdot n^2/2 to within an error small enough to be dominated once the minor arcs are controlled — provided n≥1027n\ge 10^{27}.