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:
- Abelian. Quasi-coherent sheaves on
Spec Rhaveι-indexed coproducts and the inclusion into(Spec R).Modulescreates them (Modules/Quasicoherent/Coproducts.lean, arbitrary scheme, no hypotheses); those coproducts are exact, by the tilde equivalence withModuleCat R(Affine.lean). The second half is affine-only and is the only place an equivalence is used. - Derived.
D(QCoh(Spec R))therefore hasι-indexed coproducts,Qpreserves them, and so does everyHⁿ. These are the repository'sDerivedCategory.Coproductsinstances, now applicable because layer 1 has supplied their AB4 hypothesis. - The exact derived inclusion
D(QCoh(Spec R)) ⥤ D((Spec R).Modules)preserves them, byFunctor.mapDerivedCategory_preservesCoproductsOfShapeapplied to the exact, coproduct-preserving inclusion of layer 1. - The realization
affineQuasicoherentDerivedToDqcpreserves them: its composite with the honestDqcinclusion 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.
Every cohomology functor on the affine quasi-coherent derived category
preserves ι-indexed coproducts.
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.