Kernel functors between triangulated categories #
A Fourier--Mukai transform is assembled from a correspondence. Geometrically,
for the projections p : X × Y ⟶ X and q : X × Y ⟶ Y and a kernel
P ∈ D(X × Y), the transform is
Φ_P(E) = Rq_*(Lp^* E ⊗^L P).
This file records the three functors of that composite and nothing else about
them: a Correspondence supplies a source functor in the role of Lp^*, a
bifunctor in the role of ⊗^L, and a target functor in the role of Rq_*.
The transform with a chosen kernel is their composite, and IsKernelFunctor
is the property of being isomorphic to such a composite.
Following StabilityCondition.Families.ExactPullback, exactness is an explicit
hypothesis rather than a conclusion. The Correspondence structure asks for
no triangulated structure at all; the section that concludes the transform is
triangulated states the hypotheses on the three constituents.
What this file does not assert #
The geometric content of Fourier--Mukai theory is exactly the part that is absent here. This file does not
- construct any of the three functors, or assume a scheme, a product, a derived category, or a derived functor anywhere;
- assume
pullandpushare adjoint, in either order, or supply a projection formula relatingtensorto either of them; - assume
tensoris monoidal, symmetric, associative, or unital, or that it is exact in either variable; - convolve two kernels or prove that a composite of two transforms is a transform;
- prove that a kernel is determined by its transform, or that any particular class of functors consists of kernel functors (Orlov's theorem);
- connect a kernel functor to a stability condition, a t-structure, or a Grothendieck group.
The last of those is deliberate placement rather than difficulty: the
Grothendieck-group statements need Triangulated.K₀, which lives under
StabilityCondition.Foundation, and importing it here would point a generic
module at the stability track.
The functorial data a Fourier--Mukai transform is built from, with no kernel chosen and no compatibility between the three functors assumed.
𝒵 plays the role of the derived category of the product, pull the role of
Lp^*, tensor the role of ⊗^L on the product, and push the role of
Rq_*. A caller supplies all three; nothing here constructs them.
- pull : Functor 𝒳 𝒵
The functor in the role of derived pullback along the first projection.
The bifunctor in the role of the derived tensor product on the product category.
tensor.obj Kis the twist by the kernelK.- push : Functor 𝒵 𝒴
The functor in the role of derived pushforward along the second projection.
Instances For
The Fourier--Mukai transform of C with kernel K: pull back, twist by
K, push forward. An abbrev so that instance search sees the composite and
derives the shift and triangulation instances from those of the three
constituents.
Instances For
A morphism of kernels induces a natural transformation of transforms. This is the categorical map needed before a cone of kernels can be transported; it uses only the functoriality already present in the abstract correspondence and makes no claim that a suitable kernel morphism or cone exists.
Equations
- C.transformMap f = C.pull.whiskerLeft (CategoryTheory.Functor.whiskerRight (C.tensor.map f) C.push)
Instances For
The Fourier--Mukai transform as a functor of its kernel.
Packaging transform and transformMap into one functor is the canonical
interface for constructions, such as dg cones, which vary the kernel. It
also makes the identity and composition laws available through ordinary
functoriality rather than requiring each consumer to restate them.
Equations
- C.kernelTransform = { obj := fun (K : 𝒵) => C.transform K, map := fun {X Y : 𝒵} (f : X ⟶ Y) => C.transformMap f, map_id := ⋯, map_comp := ⋯ }
Instances For
Evaluate the kernel-variable transform functor on one source object. This is the exact functor whose preservation of a kernel triangle says that the corresponding transforms form a pointwise distinguished triangle.
Equations
- C.kernelEvaluation E = C.kernelTransform.comp ((CategoryTheory.evaluation 𝒳 𝒴).obj E)
Instances For
Isomorphic kernels give isomorphic transforms. This is the only
comparison between two kernels that holds with no hypothesis on tensor; the
converse is the hard direction and is not asserted anywhere in this file.
Equations
- C.transformMapIso e = C.pull.isoWhiskerLeft (CategoryTheory.Functor.isoWhiskerRight (C.tensor.mapIso e) C.push)
Instances For
A functor is a kernel functor for C when it is isomorphic to the
transform of some kernel. The kernel is existentially quantified and the
witnessing isomorphism is only Nonempty: this is the property, not a choice
of kernel.
That every functor in a suitable setting has this property is the theorem Fourier--Mukai theory is aimed at. It is stated nowhere in this file.
Equations
- C.IsKernelFunctor Φ = ∃ (K : 𝒵), Nonempty (Φ ≅ C.transform K)
Instances For
A transform is a kernel functor, witnessed by its own kernel.
Being a kernel functor is invariant under natural isomorphism.
A named kernel for a kernel functor, together with its isomorphism. The choice is not canonical, and no result in this file depends on which kernel is returned.
Equations
- h.kernel = Exists.choose h
Instances For
The isomorphism witnessing that Φ is the transform of the chosen
kernel.
Instances For
The transform is triangulated as soon as its three constituents are.
Exactness of the twist tensor.obj K is a hypothesis on the chosen kernel,
not a property of C: the geometric statement it stands for is that − ⊗^L P
is exact, which for a general kernel needs the derived tensor product rather
than the underived one.
A transform of triangulated constituents is additive. Immediate from
transform_isTriangulated, and stated separately because Additive is the
hypothesis the Grothendieck-group API asks for.