MathLabs

Worked solution: Ngô's proof via the Hitchin fibration and geometric support theorem (2008)

Step 7 of 9: Match GG and HH via purity and the product formula
In plain words

Once the cohomology of the Hitchin fibration is pinned down everywhere, the same argument, run for the smaller group HH's own Hitchin fibration, gives a matching description there. A further estimate called purity, controlling the size of the numbers involved, shows the two descriptions must agree exactly wherever HH's parameter space sits inside GG's.

Because point counts on Hitchin fibers factor, place by place, into the local counts from Step 3, this global equality forces the corresponding local orbital integrals to agree at every place — which is exactly the fundamental lemma for Lie algebras.

#MG,a(Fq) = #MH,a(Fq)for generic a\#\mathcal{M}_{G,a}(\mathbb{F}_q) \ = \ \#\mathcal{M}_{H,a}(\mathbb{F}_q) \qquad \text{for generic } a
Detailed analysis

Ngô carries out the same decomposition-and-support analysis for the endoscopic group HH, whose parameter space AHA_H maps naturally into AA via the embedding of HH's root data into GG's. Deligne's purity theorem, from his proof of the Weil conjectures, bounds the Frobenius weights appearing in the cohomology of both fibrations, and combined with the support theorem this forces the point counts of the GG-fiber and the transported HH-fiber over generic aa to coincide exactly, not merely up to a bounded error (Ngô 2010, §7–8; Hales 2011, §6).

By the product formula for masses — the global count factors as a product of local affine Springer fiber counts, one per place of the curve, almost all trivially equal because γ\gamma is unramified there — this global equality forces the individual local terms to match at every remaining place, precisely the κ\kappa-weighted orbital integral identity conjectured by Langlands and Shelstad, now established for Lie algebras of any reductive group over a local field of positive characteristic.

This is the climax of Ngô's paper: a global, geometric statement matching cohomology of two families of varieties has been translated back into the local, arithmetic statement from Step 1.

Terms in this step
purity (Weil conjectures)
A property, proved by Pierre Deligne in 1974, that the eigenvalues of the Frobenius map on the cohomology of an algebraic variety over Fq\mathbb{F}_q have a precisely controlled absolute value; it rules out cancellation and pins down exact point counts.
Knowledge used in this step