Section 15.1 · PDF pages 846–849
Nef numerics in every characteristic
The notation transcription is provisional. Check it against the printed mathematics when citing or reusing a formula.
15.1 Nef numerics in every characteristic
The nef numerical package has a broader scope than the generic characteristic-zero limit in Proposition 12.4. Here the field can have any characteristic, and the projective scheme need not be reduced or equidimensional. We keep these hypotheses.
Lemma 15.5 (Cohomology and backward differences). Let X be a projective scheme over a field k and let ℱ be coherent, with support of dimension at most r. For invertible sheaves L_{1}, . . . , L_{s} put
Selectable PDF notation • transcription unchecked
P_{ℱ}(m_{1}, . . . , m_{s}) = χ(X, ℱ ⊗L^{m}_{1}^{1}⊗· · · ⊗L^{m}_{s}^{s}).This is a polynomial with rational coefficients of total degree at most r, for all integer arguments. For s = d = dim X, the source intersection number is its mixed backward difference
Selectable PDF notation • transcription unchecked
∑
I_{X}(L_{1}, . . . , L_{d}) = (−1)^{|S|}P_{𝒪X}(−1_{S}). (NN.1)
Selectable PDF notation • transcription unchecked
S⊆{1,...,d}It is additive in the tensor-product class of each line bundle. In particular,
Selectable PDF notation • transcription unchecked
I_{X}(L, . . . , L)χ(X, L^{m}) = m^{d}+ O(m^{d−1}) (d ≥1). (NN.2)
Selectable PDF notation • transcription unchecked
d!
The source’s lbH0 is h^{0}(X, L) and its IsNef is nonnegative line-bundle degree on every integral complete curve in X.
Source: Development/AbelianVarietyTheory/Definitions/QuasicoherentCohomology/HdimEqZeroOfSubsingleton.lean:26--29,33--56,577--578; Development/AbelianVarietyTheory/Definitions/QuasicoherentCohomology/FiniteCohPushforwardIffOfIsClosedImmersion.lean:26--33,323--332,574--585; Development/AbelianVarietyTheory/Definitions/Divisors/LineBundleGlobalSectionsDimCongrPicardClass.lean:32--55.
Proof. The source coh is the cohomology of the underlying abelian sheaf on the Zariski site; its scalar ac-tion is induced by multiplication through k → Γ(X, 𝒪_{X}). Thus it is ordinary k-linear sheaf cohomology. Concretely, on a finite affine open cover, all finite intersections are affine because X is separated. Quasi-coherent sheaves are acyclic on these intersections, so the cover’s Cech complex computes both descriptions, with the same scalar action. The definition of hdim consequently gives the usual dimension. Coherent coho-mology on a proper scheme over k is finite dimensional and vanishes in sufficiently high degree [25, III_{1},
Corollary 3.2.3, p. 117]. The source finite-support sum chi is therefore the ordinary finite alternating Euler sum. Its twist operation glues tensor products using the transition functions of the given line bundles.
We give the polynomial argument, including the dimension bound. Let G_{r} be the subgroup of the Grothendieck group of coherent sheaves generated by sheaves with support of dimension at most r. Set G_{−1} = 0. Twisting by L^{−1} induces an operator T_{L}−1; put δ_{L} = 1 −T_{L}−1. Then
δ_{L}G_{r} ⊆G_{r−1}. (NN.3)
To prove this for a generator ℱ, write L = A⊗B^{−1} with A, B sufficiently ample. Over an infinite field, each has a section avoiding the finitely many associated points of ℱ. Multiplication by that section is injective on ℱ, and its cokernel has support of dimension at most r −1. This proves (NN.3) for A and B. The operator identity
Selectable PDF notation • transcription unchecked
δ_{L} = δ_{A} −T_{A}−1T_{B}δ_{B}proves it for L. Over a finite field one can first extend to an infinite field; affine Cech cohomology com-mutes with a flat field extension, preserving its dimension and all Euler characteristics. We will use only algebraically closed fields below.
All twisting operators commute. Consequently every mixed product of r + 1 such differences has zero Euler characteristic. The resulting integer function on ℤ^{s} has all differences of total order r+1 zero. Iterating( ) the one-variable Newton difference formula expresses it as a rational linear combination of∑ ∏_{i}^{m}_{ai}^{i} with
_{i}a_{i} ≤r. This proves the polynomial assertion at all integer arguments, including the negative arguments in (NN.1).
The transition units in lbInvProd are exactly ∏_{i∈S}g_{i}^{−1}, so its twist is ⊗_{i∈S}L_{i}^{−1}. This proves (NN.1), with its sign. Furthermore
Selectable PDF notation • transcription unchecked
δ_{L⊗M} = δ_{L} + δ_{M} −δ_{L}δ_{M}.After the other d−1 differences, the final term is a product of d+1 differences and vanishes by (NN.3). This gives additivity in every argument; commutativity gives symmetry. For a single variable the dth backward difference of a degree-d polynomial is d! times its leading coefficient. Applying (NN.1) gives (NN.2).
Finally a source IsCompleteCurve is an irreducible closed subset of dimension one with its reduced struc-ture and proper structural map. It is exactly an integral complete curve. On such a curve, χ(L) −χ(𝒪) is
its usual degree; equivalently the source χ(𝒪) −χ(L^{−1}) is that degree. One can check this using a rational section: normalization gives a sum of orders at the closed points, weighted by residue degrees; the kernel and cokernel contributed by the finite-length normalization quotient have equal length. Thus the source interNumRed tests precisely these curve degrees.
Background fact 15.6 (Characteristic-free nef inputs from the original papers). For a coherent sheaf ℱ on a projective scheme and a nef line bundle L over an algebraically closed field,
h^{q}(X, ℱ ⊗L^{m}) = O(m^{r−q}) (q ≥1), r = dim Supp ℱ, (NN.4)
where negative exponents mean eventual vanishing. This is Fujita’s Theorem 6.2, pp. 370–371 [47]; the possibly stronger exponent involving his index of positivity is not needed. His curve criterion Theorem 2.7, p. 361, identifies numerical semipositivity with nefness on integral varieties. Applied to the reduced compo-nents of the support, it gives the hypothesis in (NN.4). The proof explicitly treats positive characteristic in
Lemma 5.8 and nilpotent schemes in 5.10; there is no characteristic-zero restriction.
We also use: if H is ample and L is nef on a projective scheme over a field, then H ⊗L^{m} is ample for every m ≥0 [48, Proposition 2.13]. The cited author version includes the 2018 correction; that correction leaves this proposition unchanged.
Proposition 15.7 (All three source nef numerical assertions). Over every algebraically closed field, the source assertions NefProductNonneg, NefRiemannRoch and NefBigIff hold with all their stated scheme hy-potheses.
Source: Development/AbelianVarietyTheory/PublishedTheorems/Divisors2/ExistsAlgClosedBaseChange.lean:15--32.
Proof. First suppose d = dim X ≥ 1 and L_{1}, . . . , L_{d} are nef. Choose an ample H. By Fact 15.6, H ⊗ L_{i}^{n} is ample for every positive integer n. Mixed top intersection of ample line bundles is nonnegative, including on nonreduced schemes, for the following reason. Replace each bundle by a positive very ample power. Successively choose sections avoiding the associated points of the coherent quotient produced at the preceding step. This is possible over the infinite field k. Each multiplication is injective, and each cokernel has support of one smaller dimension. After d steps, it is supported in dimension zero (or is zero). By additivity of Euler characteristics the intersection is its Euler characteristic, hence its nonnegative section dimension. Undoing the positive powers divides by their positive product, using the multilinearity proved in Lemma 15.5.
Apply this to H ⊗L^{n}_{i} and divide the resulting inequality by n^{d}. Expanding by multilinearity, its limit as n →∞ is I_{X}(L_{1}, . . . , L_{d}); every term containing an H has a smaller power of n and disappears. The limit is nonnegative. For d = 0, the empty intersection is χ(𝒪_{X}) = h^{0}(𝒪_{X}) ≥0. This proves NefProductNonneg. In particular no reduction or equidimensionality assumption has entered the argument: the sections avoided all associated points, including embedded ones, and the final zero-dimensional sheaf retains their lengths.
For NefRiemannRoch, let L be nef and D = dim X ≥1. By (NN.4), each nonzero higher cohomology term is O(m^{D−1}); there are only finitely many such terms. Consequently
Selectable PDF notation • transcription unchecked
h^{0}(X, L^{m}) −χ(X, L^{m}) = O(m^{D−1}).Combine this with (NN.2), and identify the source topSelfIntersection with I_{X}(L, . . . , L):
Selectable PDF notation • transcription unchecked
I_{X}(L, . . . , L)h^{0}(X, L^{m}) = m^{D}+ O(m^{D−1}). (NN.5)
Selectable PDF notation • transcription unchecked
D!
This is precisely the absolute big-O statement required, since the norm of a real number is its absolute value. For D = 0, X is a finite scheme over k, a disjoint union of spectra of artinian local algebras. An invertible module over each of these local algebras is free of rank one. Thus h^{0}(X, L^{m}) = h^{0}(X, 𝒪_{X}) for every m, which is the empty source intersection. The error is identically zero, and is in particular O(m^{−1}), as demanded by the source’s integer exponent D −1.
Finally assume the integral hypothesis in NefBigIff. Properness gives all the finite-dimensional section spaces in the source definition of IsBig. Put a = I_{X}(L, . . . , L)/D!. Formula (NN.5), with the exact zero-dimensional case just proved, gives
Selectable PDF notation • transcription unchecked
h^{0}(X, L^{m})_{D} −→a. (NN.6)
Selectable PDF notation • transcription unchecked
m
For m ≥1 the denominator is positive. If the source’s frequent-growth condition holds with c > 0, its arbi-trarily large m satisfying h^{0}(X, L^{m}) ≥cm^{D} force a ≥c: otherwise convergence would put all sufficiently large ratios strictly below c. Hence the top self-intersection is positive. Conversely, if that intersection is positive, then a > 0. Convergence makes the ratio at least a/2 for all sufficiently large m. Taking c = a/2 proves the required frequently quantified lower bound, indeed an eventual one. The source dimension wit-ness can be chosen to be D; any witness already in IsBig equals it, since both equal the same topological Krull dimension. This proves both directions, with the factorial, the quantifier on m and the dimension-zero case unchanged.