Alphabeta Math
LemmaStatement: AI-adaptedProof: AI-generatedPipeline-generatedprecheck passjudge pass (gpt-6-luna)verified 2026-11-04 (gpt-6.1-sol)
How statement and proof provenance work

The first chip identifies the source of the statement or construction; the second identifies the source of its local proof or verification.

  • Literature-sourced: the exact statement appears in a cited source; only wording and notation differ.
  • AI-adapted: a semantically identical restatement of literature-sourced material, modulo indexing, notation, and boundary cases adopted by the library.
  • AI-generated: a genuinely novel statement formulated by AI, with no source for the claim itself.

These labels describe origin, not correctness: citations and verification chips remain separate evidence.

Unique lc-place residue pushforward and subadjoint discriminant effectivity

Statement

Let X be a smooth integral projective complex variety, or a normal integral projective Gorenstein terminal complex threefold with isolated singularities, and let μ:Y→X be an actual proper smooth log resolution and G an effective rational divisor. In the terminal branch retain the supplied effective exceptional divisor with μ-ample negative required by the isolated-centre vanishing supplier; no arbitrary small-resolution claim is included. Let W⊂X be a reduced integral closed subvariety of codimension at least one. Suppose

K_Y+E+F=μ^*(K_X+G),

where E is ONE smooth prime divisor mapping onto W, all coefficients of F are strictly below one, E is not in Supp F, and E∪Supp F∪Exc(μ) is SNC. Negative coefficients of F are assumed exceptional; this holds for the displayed log pullback because G is effective, with exceptional discrepancies handled by the actual resolution in either branch. Put P=ceil(−F). Thus P is effective integral exceptional and E is not a component of P.

Assume the already constructed isolated-centre ideal/relative-vanishing interface

μ_*O_Y(P−E)=I_W,     R¹μ_*O_Y(P−E)=0.

The conclusion is the canonical quotient identification

μ_*O_E(P)=O_W,     (μ|E)_*O_E=O_W.

In particular W is normal and the geometric generic fibre of μ|E is connected. The construction does not assume a klt replacement on W or a residue-pushforward theorem from the literature.

Suppose also μ|E factors as E --f→ V --ν→ W with V smooth and ν birational and an isomorphism at the generic point of each nonexceptional divisor. Write FE=F∣E and assume the SNC restriction permits coefficientwise rounding P∣E=⌈−FE⌉. This equality follows from the above SNC condition when components restrict as distinct prime divisors; the condition is explicitly retained on any subsequently modified model.

For a prime divisor Q on V not exceptional over W, write f∗Q=Σj wj Pj, including all components and setting dj=coeffPj(FE), with dj=0 for omitted boundary components. Then some j satisfies

d_j≥1−w_j.

Consequently the subadjoint discriminant coefficient δQ=maxj (dj+wj−1)/wj is nonnegative, and the pushforward to W of the discriminant is effective. No claim about signs of ν-exceptional discriminant coefficients is made.

Assume the Axiom of Choice through the supplied coherent-cohomology interfaces. A unique lc PLACE is an explicit hypothesis, stronger than a unique lc centre. No connectedness assertion about all special fibres is included.

Facts & Assumptions

[A1]

The actual unique-place resolution and the rounding compatibility on the consumed model are explicit hypotheses, and are not inferred from uniqueness of the non-klt centre. The Axiom of Choice is inherited through the cohomology interfaces.

[F1]

The exact suppliers are Isolated lc-centre SNC vanishing and polynomial nonvanishing, A normal Noetherian domain is the intersection of its height-one localizations, Regular varieties are normal, Cech cohomology computes quasi-coherent cohomology on a separated scheme, The Axiom of Choice. The first supplies the actual ideal equality and relative vanishing without imposing the later strict numerical nonvanishing bound. The normal codimension-one criterion handles the exceptional allowance; regular-local normality handles E; flat field base change for H⁰ is constructed by the finite affine Čech kernel argument.

Proof

Given: Exactly the Statement hypotheses.

1.1givenA1F1algebra

Normality of X gives μ∗OY(P)=OX. A rational function regular away from the exceptional locus with poles bounded by the effective exceptional P has no pole at any codimension-one point of X and hence is regular on X. Conversely regular functions pull back regularly. This proves the equality as sheaves, not just as global sections.

2.1givenA1F1step 1.1algebra

Because E is a Cartier divisor on smooth Y, there is an exact sequence 0→OY(P−E)→OY(P)→OE(P)→0.

3.1givenA1F1step 2.1algebra

Pushforward and the assumed R¹ vanishing give 0→IW→OX→μ∗OE(P)→0.

4.1givenA1F1step 3.1algebra

The quotient is OW. The image of 1 under the quotient identification is precisely the restriction to E of the canonical section of OY(P). Thus this identifies the relevant map as well as the abstract quotient sheaf.

5.1givenA1F1step 4.1algebra

Since P is effective and does not contain E, multiplication by its restricted canonical section gives an injection OE→OE(P). Write g=μ∣E. Pullback of regular functions gives an injection OW→g∗OE, because W and E are integral and g is dominant. The composite OW→g∗OE→g∗OE(P)=OW

6.1givenA1F1step 5.1algebra

is the identity by the explicit image-of-1 description in Section 1. Since the second arrow is injective, its image lies in OW and contains all of OW, so g∗OE=OW canonically.

7.1givenA1F1step 6.1algebra

Here is the direct normality argument. Let U⊂W be affine and h∈K(W) integral over OW(U). Pull h back to the function field of E. At every point of g−1U, the same monic equation makes it integral over the local normal ring of smooth E; hence it is regular there. Therefore it is a section of g∗OE(U)=OW(U). Every such affine coordinate ring is integrally closed, so W is normal. No normalization or connectedness theorem is required for this deduction.

8.1givenA1F1step 7.1algebra

Let K=K(W) and η its generic point. Localization of the direct-image equality gives H0(Eη,O)=K. This localization can be checked on a finite affine Čech cover of the proper separated generic fibre: localization commutes with the finite kernel computing H⁰. The same flat Čech argument after a field extension gives H0(Eη×KKbar,O)=Kbar.

9.1givenA1F1step 8.1algebra

A disconnected scheme has a nontrivial idempotent global function, obtained by setting 0 and 1 on the two clopen parts. The displayed field has none, proving geometric generic connectedness. The generic fibre is regular, since its affine rings are localizations of the regular rings of Y restricted to smooth E; over the characteristic-zero field K, a regular finite-type scheme is geometrically regular. Thus it is a smooth geometrically connected generic fibre. In the codimension-two case it is a curve; smooth connected components of a smooth curve are disjoint, so it is geometrically integral.

10.1givenA1F1step 9.1algebra

This is the precise connectedness used to reconstruct the generic marked rational curve. Connectedness of EVERY special fibre would require a separate Stein-factorization/formal-functions bridge; it is not claimed here or used in this component. The stable rational family, once separately constructed, has connected special fibres by its own definition and construction.

11.1givenA1F1step 10.1algebra

Fix Q as in the Statement. Near its generic image in the now normal W, ν is an isomorphism and Q has a uniformizing rational function t. Suppose for contradiction that dj<1−wj for every component over Q. Since wj is a positive integer, ⌈−dj⌉≥wj

12.1givenA1F1step 11.1algebra

for every j. The rounding compatibility gives P∣E≥f∗Q over that generic point. After removing the images of the finitely many other boundary divisors, the rational function 1/g∗t has divisor at least −P|E and so is a local section of OE(P). By g∗OE(P)=OW it must be regular at the generic point of Q downstairs. But 1/t has a simple pole in that discrete valuation ring, a contradiction.

13.1givenA1F1step 12.1algebra∎

Hence some dj≥1−wj. For that j, (dj+wj−1)/wj≥0, so the maximum δQ is nonnegative. This proves exactly the needed effectivity after pushing the discriminant to W; exceptional Q do not contribute to that pushforward and are not constrained by this argument. Coefficients dj=0 are included, so a reduced unmarked branch wj=1 already satisfies the required inequality.

Remarks

The exact source passage consulted is the retained extracted proof of Kawamata Theorem 1 on page 6. No source-provided residue-pushforward theorem is imported; the short exact sequence and ideal vanishing give it here. Construction of a genuine unique lc place and the marked rational model remains with the geometry lane. Full special-fibre connectedness is not needed by this component and is not asserted.

Depends on

Dependency tree · 2 levels

Sources