In this section, we study the relation between the geometry and
algebra of perfect stacks. We begin
with some basic properties of tensor
products of -categories. Then we prove (Theorem 4.7)
that the -category of sheaves on a product of perfect stacks,
and more generally a derived fiber product,
is given by the tensor product of the
-categories of sheaves on the factors. This implies that the
-category of sheaves on a perfect stack is self-dual, which in turn allows us to describe
-categories of functors by -categories of integral kernels (Corollary
4.10). Finally,
we extend this last result to
a relative setting where the base is an arbitrary derived stack with affine diagonal.
4.1. Tensor products of -categories
In this section, we consider some properties of tensor products of -categories
that will be used in what follows. First we prove
(Proposition 4.1) that -categories of modules
over associative algebra objects are dualizable, with duals given by modules for the
opposite algebra, and that tensor product of algebras induces tensor products on module categories.
We then discuss the tensor product of small stable categories, and its compatibility
with passing to the corresponding presentable stable -categories of -objects.
4.1.1. Algebras and modules
Recall the tensor product of presentable stable -categories
developed in [L4], [L5]. Namely, the -category
of presentable -categories (with
morphisms given by left adjoints) carries a natural symmetric monoidal
tensor product that preserves stable objects.
Let be a monoidal -category. For two presentable
-categories and left tensored over , we denote
by the -category of left adjoints
from to that preserve the tensor over .
Similarly, the opposite category of is the
-category of presentable
-categories with morphisms given by right adjoints. For two
presentable -categories and left cotensored over
, we denote by the -category
of right adjoints from to that preserve the cotensor over
.
Lemma 4.2.Let and be stable presentable -categories that are
left tensored and cotensored over .
Let be a right adjoint that is tensored and cotensored over . Assume further that is colimit preserving.
Then is conservative if the induced functor is conservative
for any .
Proof.Suppose is not conservative. Then to prove the lemma, it suffices to exhibit a
presentable -category also
cotensored over and a nontrivial right adjoint
cotensored over such that is trivial.
Define to be the full -subcategory of of -acyclic objects,
that is,
objects such that is trivial. Our first
task is to show that is indeed presentable.
Observe that is equivalent to the fiber product , where the limit is computed in the -category
of -categories. Recall by [L2, Proposition 5.5.3.13], the natural functor
preserves limits. Furthermore,
the forgetful functor also preserves limits since it has a left adjoint (given by induction).
Since the functor preserves colimits and is -linear, we may regard it as a morphism in . Thus can be computed as a limit in ,
and so can be regarded as an object of .
In other words,
is presentable and furthermore tensored over .
Finally, since is tensored over , it is automatically cotensored as well.
Now it remains to show that the inclusion is indeed a right adjoint
and cotensored over .
Since preserves all limits and colimits (and in particular -filtered colimits),
the adjoint functor theorem applies. Finally, since
is cotensored over , is as well.
∎
Lemma 4.3.Let be a stable presentable -category which is left tensored and cotensored over , and let be an associative algebra in . Then the forgetful functor
is conservative.
Proof.Observe that for any tensored over , the pullback
induced by the induction is
conservative. In other words, if a functor out of
(which preserves colimits in each variable) is trivial when restricted
to , then it is necessarily trivial.
Consequently, switching to opposite categories, we have that the corresponding functor
induced by the forgetful functor
is conservative.
Now we can apply Lemma 4.2 with
to obtain that is conservative.
∎
Proof of Proposition 4.1.We first prove that is equivalent to
by the natural evaluation functor. Consider the
adjunction
where is the induction, and is the forgetful functor.
The above adjunction induces an adjunction
and thus a functor to modules over the monad acting on . The functor underlying
is given by tensoring with , so we also have an equivalence
.
By its universal characterization, the functor
is colimit preserving.
Note as well that and hence is also -linear
(or in other words, the adjunction satisfies an analogue of the projection formula).
Thus it follows from Lemma 4.3 that is also conservative.
Thus satisfies the monadic Barr-Beck conditions, and we obtain the
desired equivalence .
Next, we can apply this to the instance where is the
-category of left modules over another associative algebra
to conclude that there is a natural equivalence . We now have a chain of
adjunctions
in which the composite is colimit preserving and
conservative, and hence satisfies the monadic Barr-Beck conditions.
Furthermore, the above adjunction naturally extends to a diagram in which the cycle of left adjoints (denoted
by bowed arrows), and hence
also the cycle of right adjoints (denoted by straight arrows), commute
Here is the induction, is the forgetful functor,
is the natural functor factoring through ,
and is its right adjoint. From this diagram, we obtain a morphism of monads
Now the underlying functors of the monads and are both equivalent to the tensor , so
the above morphism of monads is an equivalence. Thus we obtain the promised equivalence
.
Finally, we show that the -category of left -modules
is a dualizable -module by directly exhibiting the
-category of right -modules as its
dual. The trace map is given by the two-sided bar construction
The unit map is given by the induction
where we regard as an -bimodule.
One can verify directly that the composition
is equivalent to the identity. First, is equivalent to
regarded as an -module, and
second, is equivalent to .
∎
4.1.2. Small stable categories
We have been working with the symmetric monoidal structure on the -category
of presentable
-categories as developed in [L4],
[L5].
We will also need the tensor product of small stable idempotent complete
-categories,
in particular, the -categories of compact objects in presentable stable -categories.
Let be the full -subcategory of the -category
of stable categories (with morphisms exact functors) consisting of
those -categories that are idempotent complete. Recall that an
-category is idempotent complete if the essential image of
the Yoneda embedding is closed under retracts.
Proposition 4.4.The -category carries a symmetric monoidal structure characterized by the property
that for , the -category of exact functors
is equivalent to the full -subcategory
of all functors that preserve finite colimits in
and separately. Furthermore, passing to the corresponding
stable presentable -categories of -objects
is naturally a symmetric monoidal functor.
where the tensor product of the right hand side is calculated in the -category
of presentable -categories (with morphisms left adjoints), and
the superscript c denotes the full -subcategory of compact objects of a presentable -category.
Since is idempotent complete
and retracts of compact objects are compact,
is idempotent complete as well.
Thus the tensor product is indeed an object of .
For , let be the full -subcategory of
functors that preserve finite colimits in
and separately.
We claim that For , the tensor product
corepresents the functor
in the sense that for any , there is a canonical equivalence
As a consequence, the associativity and symmetry of the tensor product
will immediately follow from the analogous properties of .
For , let be
the full -subcategory of
functors that preserve colimits in
and separately.
To prove the claim, observe that the inclusion induces a fully faithful functor
Its essential image consists of functors that preserve compact objects.
By definition of the monoidal structure on the -category
of presentable -categories, we have a further equivalence
Since the compact objects of are generated by finite colimits of external products of compacts objects, we obtain an equivalence between
and the full -subcategory of consisting of functors that preserve compact objects.
In other words,
we have the asserted equivalence that characterizes the tensor product
Finally, the assertion that the functor
is symmetric monoidal is immediate from the constructions
and the natural equivalence , for .
∎
Remark 4.5. Given small stable idempotent complete
-categories ,
by construction
their tensor product is again
a small stable idempotent complete
-category. Though it is possible to consider other versions of a tensor
product on small stable -categories that need not preserve idempotent complete
-categories,
our approach builds it in from the beginning.
4.2. Sheaves on fiber products
In this section, we study the -category of sheaves on the derived
fiber product of perfect stacks. The main technical result is
that it is equivalent to the tensor product of the -categories of sheaves
on the factors
(Theorem 4.7). The proof involves first showing that an analogous
assertion holds for -categories of perfect complexes.
The remainder of the section is devoted to collecting corollaries of the main techinical result.
Recall that for a perfect stack ,
compact (equivalently, perfect or dualizable) objects in the -category
form a small stable idempotent complete -category , and there is canonical
equivalence
.
Recall as well the tensor product of small stable idempotent complete -categories,
and that the functor is symmetric monoidal.
Proposition 4.6.Let be perfect stacks. Then external tensor product
defines an equivalence
In other words, the -category of perfect complexes on the product is the
(small stable idempotent complete) tensor product of the
-categories of perfect complexes on the factors.
Proof.Set .
By Proposition
3.24, we know that the external product takes compact objects
to compact objects, and is generated by external products.
Thus it suffices to verify that for we have an
equivalence
Using the fact that each is dualizable
and satisfies the projection formula (since it is perfect), we calculate
∎
We now prove our main theorem which identifies -categories of sheaves
on fiber products algebraically. The proof relies on the following
consequence [L2, Corollary 5.5.3.4] of the -categorical
adjoint functor theorem: there is a canonical equivalence
between the opposite of the
-category of presentable -categories with morphisms
left adjoints and the -category of presentable
-categories with morphisms right adjoints. In other words, we can
reverse diagrams of presentable -categories, in which the
functors are all left adjoints, by passing to the corresponding right
adjoints.
We will also use the fact that the calculation of small limits of
presentable -categories is independent of context. Namely,
by [L2, Proposition 5.5.3.13, Theorem 5.5.3.18], the forgetful
functors from , to all
-categories preserves small limits. In particular, given a small
diagram of both left and right adjoints, the universal maps from the
limit to the terms of the diagram are also both left and right
adjoints.
Proof.To begin, consider the case , so .
By Proposition 4.6 and
the fact that is symmetric monoidal,
the external product functor provides an equivalence
.
To begin the case of a general perfect stack ,
consider the augmented cosimplicial diagram
with the obvious maps constructed from the given maps .
Applying the (contravariant) functor , we obtain an augmented simplicial -category
with structure maps given by pullbacks. By the absolute
case of the theorem when , if we forget the augmentation,
we obtain the simplicial -category with simplices
and structure maps given by tensor contractions.
This is precisely
the two-sided bar construction
[L4, 4.5] whose geometric realization, by definition [L5, 5],
calculates the tensor product
of -modules
.
Furthermore, the augmentation provides the natural map
which we will prove is an equivalence.
The above geometric realization is a colimit in , and hence (as
observed prior to the statement of the theorem) may be evaluated as a limit in the opposite category
. Thus we find that is also the
totalization of the cosimplicial -category
with structure maps given by pushforwards. (We note for
future reference that these structure maps are pushforwards along
affine morphisms, hence are also colimit preserving, i.e., left
adjoints.)
In
particular, pushforward along the augmentation provides a natural
functor
which is an equivalence if and only if is an equivalence.
To summarize some of the above structure, we have a diagram of commuting left (lower arrows) and right (upper arrows) adjoints
where is the universal map from the totalization to the zero cosimplices,
and likewise, is the universal map from the zero simplices to the geometric realization.
Thus we obtain a map of monads
acting on .
The geometric pushforward is conservative and preserves
colimits since is affine. Hence by the Barr-Beck theorem, we
have a canonical equivalence
We also claim that the universal map is conservative and
preserves colimits. For the first assertion, recall that is
nothing more than the forgetful map from the totalization to the zero
cosimplices. Since the -categories involved are all stable,
evaluating conservatism of a functor is equivalent to determining if
nonzero objects are sent to zero. But an object sent to zero in the
zeroth cosimplices is sent to zero in all cosimplices, and hence is
equivalent to the zero object in the limit -category.
To see preserves colimits, recall that the structure maps of
our cosimplicial diagram are both right and left adjoints. It then
follows (as observed prior to the statement of the theorem) that the
totalization may be evaluated equivalently back in the category . In particular, the universal functor is a
morphism in , and hence a left adjoint and so
preserves colimits.
We may now apply the Barr-Beck theorem, giving a canonical equivalence
Thus it remains to show that the above morphism of monads is an equivalence.
It is a straightforward diagram chase to check that the monad
is nothing more than the composition of the geometric functors associated
to the initial cosimplicial maps
Thus by base change, it is equivalent to the monad
.
This concludes the proof of the theorem.
∎
Proof.By Theorem 4.7, we have a canonical factorization
. Using this
identification, we can define the unit and trace by the
correspondences and
, where is the relative diagonal. We need to check that
the following composition is the identity:
The argument is a chase in the following diagram (with Cartesian square):
Applying base change and identities for compositions, we have equivalences of functors
Remark 4.9. The above argument also shows that is dualizable
over perfect -modules when is smooth and proper: smoothness is required for the unit in to be
perfect and properness is required for the trace to land in perfect -modules.
For a derived stack ,
let be stable presentable -modules.
To reduce notation, we write
for the stable presentable -category of -linear colimit preserving
functors .
Proof.The statement is an immediate consequence of Theorem 4.7 and the fact that
is self-dual (Corollary 4.8). It implies that
internal hom of -modules out of is calculated by
tensoring with .
∎
Remark 4.11. The equivalence of Corollary 4.10 is naturally
monoidal in the following sense. For perfect stacks
mapping to a perfect stack , there is a convolution map
given by pulling back and pushing forward with respect to the triple
product (see Section 5.2).
On the other hand, we have a composition map
and the equivalence of the theorem intertwines these composition maps.
Finally, for a finite simplicial set , we would like to compare the formation of mapping stacks
(this can be viewed
as the
cotensoring of stacks over simplicial sets) with the formation of
tensor products of -categories (the tensoring of
symmetric monoidal -categories over simplicial sets). By the notation
, we mean the geometric realization of the simplicial
-category given by the constant assignment of to each simplex
of the simpicial set .
Proof.We calculate by induction on the simplices as an
iterated fiber product of copies of over perfect stacks (by
Proposition 3.24), and apply Theorem 4.7 at each stage.
∎
4.3. General base stacks
In this section, we extend Corollary 4.10
on integral transforms to a relative setting where the base is allowed to be an arbitrary derived stack
with affine diagonal (not necessarily perfect).
In order to describe integral transforms relative to such a base, we
will utilize the simple behavior of -categories of sheaves under affine base change.
Proposition 4.13.Let be a derived stack with affine diagonal, and let be
an affine over . Then is a self-dual -module.
In particular, for any there is a canonical
equivalence
Proof.Since has affine diagonal, the map is a
relative affine, which implies that is colimit preserving and
conservative. Hence satisfies the monadic Barr-Beck criteria
[L4, Theorem 3.4.5], implying that the natural map is an equivalence. By the projection formula, the monad
is equivalent to the functor , with
monad structure given by the algebra structure on . As a
consequence, we see that is equivalent to
. Now Proposition 4.1 gives
that is self-dual as a -module.
Since the -category is a dualizable -module it
follows that the functor
commutes with
limits of -module categories. Thus we have equivalences
Another application of Proposition 4.1 implies the following equivalence
Using that the map is affine and that , we obtain that the above is further equivalent to
Here we have used that sends all colimits to limits, and that commutes with colimits (since it is the left adjoint to the mapping stack over ).
∎
We now show that in the
general setting where the base is an arbitrary derived stack
with affine diagonal,
functors continue to be given by integral kernels.
Theorem 4.14.Let be a perfect map of derived stacks with affine diagonal, and let be an arbitrary map of derived stacks. Then there is a natural
map that is an
equivalence of -categories.
We define a functor by sending a quasi-coherent sheaf to the functor . This is
a colimit preserving functor, since is perfect. Furthermore,
using the projection
formula for the map , this functor naturally admits the
extra structure of -linearity as follows
for any . Therefore, we in fact obtain a functor
, and the rest of this proof
will be devoted to showing it is an equivalence.
Recall that the -category of quasi-coherent sheaves on is given by the limit
By working locally in the target,
this provides a description of the -category of -linear functors with values in as
the limit
Likewise,
we have a description of the -category of quasi-coherent sheaves on
the fiber product as a limit
This follows from the fact that the functor takes colimits
to limits, and the fiber product functor commutes
with all colimits (because it has a right adjoint).
Now one can analyze the functor
by considering the terms in the above two limits. That is, to prove the theorem,
it suffices to prove it locally in the target : for any , we must show that the functor
is an equivalence.
We will prove this in two steps. First, we will deal with case that the base is affine.
Afterward, we will use this case to deal with a general base .
So assume for the time being that . Then is a perfect stack
over , and by Corollary 4.8,
is a self-dual -module. Thus we have equivalences
By Proposition 4.13 we know that the
functor takes affine base change to tensor product of
-categories. Therefore we have an equivalence
Putting together the above equivalences, we conclude that we have equivalences
This proves the theorem
when is affine.
Working locally in the base , we will now use the above
discussion to prove the theorem in general. As above, since
and fiber products behave well with respect to colimits we can
calculate the -category of quasi-coherent sheaves on the fiber product
as the limit
To calculate the -category of functors, we use the following: by
Proposition 4.13, the -category is
a dualizable -module and hence the functor commutes with all limits. Therefore we have an
equivalence
and so in particular we obtain equivalences
By the adjunction between induction and restriction, and a repeated application
of Proposition 4.13, we also have equivalences
Finally, by the above discussion and the affine case of the theorem with base ,
we obtain the following chain of equivalences
Since we previously reduced the theorem to the case when ,
this completes the proof.
∎