MathLabs
Step 4 of 5: Minor arcs: sharper bounds via Vaughan's identity and the large sieve
In plain words

On the minor arcs, S(α)S(\alpha) should behave like a 'random' sum and hence be small, but proving this rigorously with strong enough exponents was the main obstruction to lowering the threshold below 10130010^{1300}. Helfgott developed sharper, 'log-free' bounds — meaning the bound on S(α)/nS(\alpha)/n depends only on qq (the denominator of the nearby rational) without extra factors of log⁡n\log n creeping in from the usual decomposition of Λ\Lambda via Vaughan's identity, combined with large-sieve-type inequalities and a technique adapted from the Barban–Vehov–Graham smoothing sieve. These sharper bounds are precisely what let the minor-arc error stay small enough for nn as low as 102710^{27}, rather than requiring astronomically larger nn.

∣S(α)∣≪nq1/2(log⁡n)c(“log-free” in the exponent of q), for α∈m|S(\alpha)| \ll \frac{n}{q^{1/2}}(\log n)^{c} \quad \text{(``log-free'' in the exponent of } q\text{), for } \alpha \in \mathfrak{m}
Detailed analysis

Because major and minor arcs behave differently at the boundary between them, Helfgott further introduces 'intermediate' arcs — tails where the standard major/minor treatment loses efficiency — with their own tailored estimates. Putting the major-arc main term (Step 3) together with these minor- and intermediate-arc error bounds (all fully explicit, with every implied constant tracked numerically) yields the unconditional statement: R(n)>0R(n)>0, and hence nn is a sum of three primes, for every odd n≥1027n \ge 10^{27}.