0:005:52

Recursos

Resumen

Suppose an slc log Calabi–Yau pair has trivializations on its normalization components that refuse to glue. You can find where the obstruction lives, turn it into a single scalar, and see why one exponent a(d, Φ) = m·L kills it, however many components the pair has.

Where the gluing fails

  • The normal index theorem gives one even m for every component. The trivialization on component i is a rational m-canonical form θᵢ [0:28].
  • At a generic node, the residues from the two branches differ by a scalar r_e. Draw a graph with components as vertices and nodes as edges [1:02].
  • Rescaling θᵢ changes the ratios at vertex i. A tree can always be fixed this way. A cycle cannot, because the product of ratios around a closed walk does not change under rescaling. Think of it as currency arbitrage [1:24].
  • Practice: ratios 2, ½ and i have product i. Rescaling can't remove it, but fourth powers do [2:08].

From loop to self-map to Hodge rank

  • Go through the minimal (klt) strata of dlt models. Kollár linking inside a component has multiplier 1, and the comparison across an edge has multiplier r_e. A closed walk becomes a crepant birational self-map f of one klt stratum A, with f*θ_A = (∏ r_e) θ_A and dim A ≤ d − 1 [2:24].
  • On the Calabi–Yau-type and abelian factors, the scalar is an eigenvalue of an integral operator that preserves a positive-definite form on the lattice of T(U). By Kronecker's theorem it is a root of unity, with φ(e) ≤ rank T(U) [3:17].
  • Practice: if the rank is at most 4, then e ∈ {1,2,3,4,5,6,8,10,12}, so lcm = 120 [4:25].

The uniform exponent

Theorem 6.1 gives L(n, m). Take L = lcm L(q, m) for q ≤ d − 1 and set a = m·L. Propagate constants along a spanning tree. Every extra edge then closes up automatically, and the S₂ extension makes a(K_X + B) Cartier and trivial [4:45]. The rule: find the loops, turn each one into a self-map of one klt stratum, and bound its scalar by the rank of the top-form Hodge structure.

Uniform index bounds for slc Calabi–Yau pairs

OpenAI Math Preprints

Clase 81 / 100

Ver clases