Documentation

DerivedAlgGeo.AlgebraicGeometry.DerivedCategory.Dqc.AffineCoproducts

Coproducts in the affine quasi-coherent realization of Dqc #

Dqc.lean gives the honest locus Dqc(X) its ι-indexed coproducts on an arbitrary scheme (#721), and the inclusion into the all-sheaf derived category preserves them. Nothing there says anything about the affine realization functor of AffineRealization.lean, and that is what this file supplies.

Four layers, none of them transported along an unproved equivalence:

  1. Abelian. Quasi-coherent sheaves on Spec R have ι-indexed coproducts and the inclusion into (Spec R).Modules creates them (Modules/Quasicoherent/Coproducts.lean, arbitrary scheme, no hypotheses); those coproducts are exact, by the tilde equivalence with ModuleCat R (Affine.lean). The second half is affine-only and is the only place an equivalence is used.
  2. Derived. D(QCoh(Spec R)) therefore has ι-indexed coproducts, Q preserves them, and so does every Hⁿ. These are the repository's DerivedCategory.Coproducts instances, now applicable because layer 1 has supplied their AB4 hypothesis.
  3. The exact derived inclusion D(QCoh(Spec R)) ⥤ D((Spec R).Modules) preserves them, by Functor.mapDerivedCategory_preservesCoproductsOfShape applied to the exact, coproduct-preserving inclusion of layer 1.
  4. The realization affineQuasicoherentDerivedToDqc preserves them: its composite with the honest Dqc inclusion is layer 3, and that inclusion reflects coproducts, because closure makes it create them.

What this does not prove #

Preservation of coproducts is one of the two standing hypotheses of every compact-generation argument; it is not the argument. In particular this file still claims no unbounded essential surjectivity of affineQuasicoherentDerivedToDqc — AffineQuasicoherentDqcIdentification remains uninhabited — no compact-equals-perfect comparison, no generator closure, and no S-linearity. The bounded identification, which is proved, comes from AffineQuasicoherentIdentification.lean by a cone argument that needs finite amplitude and does not extend to the unbounded case.

Coproducts are indexed by a type in the universe of the coordinate ring, one indexing type at a time, matching the repository's universe-explicit colimit idiom rather than claiming closure under all coproducts.

The abelian layer #

The inclusion of affine quasi-coherent sheaves into all module sheaves preserves ι-indexed coproducts: the affine case of the arbitrary-scheme statement, recorded here as the moduli lane's entry point.

With affineQuasicoherentSheavesInclusion_preservesFiniteLimits and its colimit counterpart this makes it an exact, coproduct-preserving functor, which is exactly the input the derived layer below consumes.

The derived layer #

The affine quasi-coherent derived category has ι-indexed coproducts.

This is DerivedCategory.hasCoproductsOfShape at the abelian category AffineQuasicoherentSheaves R, and it is available exactly because affineQuasicoherentSheavesHasExactCoproducts supplies the AB4 hypothesis that instance asks for. Coproducts alone would not do: the construction needs quasi-isomorphisms to be stable under coproducts, which is exactness.

The localization functor from cochain complexes of affine quasi-coherent sheaves preserves ι-indexed coproducts, so the coproduct of a family in D(QCoh(Spec R)) is represented by their degreewise coproduct.

Derived tilde preserves ι-indexed coproducts. It is an equivalence (affineQuasicoherentDerivedEquivalence), so this is its left adjoint half and needs no further input; recording it identifies the coproducts of layer 2 with the ones D(ModuleCat R) already had.

Derived global sections preserves ι-indexed coproducts, being the quasi-inverse equivalence. This is a genuine statement about the affine quasi-coherent category and not about RΓ on an arbitrary scheme, where commuting with coproducts is the quasi-compactness-and-separatedness input to compact generation and is not proved anywhere in this repository.

The realization #

The exact affine derived inclusion preserves ι-indexed coproducts.

Functor.mapDerivedCategory_preservesCoproductsOfShape applied to the exact, coproduct-preserving inclusion of affine quasi-coherent sheaves into all module sheaves. Both AB4 hypotheses that theorem needs are present and neither is assumed here: on the source by affineQuasicoherentSheavesHasExactCoproducts, on the target by Scheme.Modules.ab4.

The affine realization into Dqc(Spec R) preserves ι-indexed coproducts.

The composite of the realization with the honest Dqc inclusion is the exact derived inclusion above, and that inclusion reflects coproducts (SchemeQuasicoherentDerivedCategory.ι_reflectsCoproductsOfShape), so preservation transfers from the composite to the realization itself.

The Dqc side of that argument is where the honesty lives: reflection holds because the locus creates its coproducts, so the coproduct of a family of realized complexes is computed in the all-sheaf derived category and then observed to stay in the locus. Nothing here identifies the two categories.