Recursos
Resumen
Take a single number, normalized volume at least 162 at a singular klt fourfold point, and follow how it is turned into an equation that pins the point down as an ordinary double point.
The numbers to keep in mind
- Normalized volume is the infimum of A(v)⁴ · vol(v) over valuations centered at the point. Smooth points reach the maximum n^n = 256 [0:17].
- The Spotti–Sun conjecture says the runner-up is the ordinary double point, at 2(n−1)^n. In dimension four that is 2·3⁴ = 162, and the theorem proves that bound, with equality only at the ODP.
- The difficulty: a general klt point has no equations to compute with. So the proof turns the numerical hypothesis into a local-ring constraint: embedding dimension at most five [1:07].
The chain, link by link
- Cone [1:33]. Stable degeneration gives a graded cone with the same volume, and its embedding dimension bounds the germ's. Non-isolated points are capped at 4096/27 ≈ 151.7. A degree-d index-one cover would push the volume past 256, so the cone is isolated and Gorenstein.
- Stabilizer bound [2:19]. Counting jets on a transverse slice shows that a large stabilizer would beat the minimizing valuation. So m < r/(5/2), and every primitive divisor over the vertex has A ≥ 3 (an integer, because the cone is Gorenstein).
- Hilbert series [2:47]. With the canonical weight at 5/2, Stanley duality turns the Hilbert coefficients into a signed measure. Its nonzero frequencies lie outside (−1, 1), and a band-limited test acts as a low-pass filter: Poisson summation leaves only a polynomial integral. That integral forces a function of weight in (0, 5/3).
- Threshold [3:42]. If lct ≤ 3/2, a boundary built from those functions has weight below 3/2 · 5/3 = 5/2, which leads to a contradiction on a minimal lc center. So lct > 3/2.
- Terminal section [4:15]. A general hyperplane section has discrepancy A − 1 − b along each divisor. For b = 1, A ≥ 3 is enough. For b = 2 and A = 3 the value is 0, so the threshold is needed: it forces A ≥ 4. Both bounds are needed.
- Hypersurface [5:09]. By Reid, the section is cDV, so edim ≤ 4, and the cone has edim ≤ 5. That makes the germ one equation in five-space, and Liu's lci theorem gives ≤ 162, with equality only at the ODP.
When you face a volume-gap question, don't hunt for a better valuation. Squeeze structure out of the number until an equation appears [5:36].