In this post our main goal will be to study how exterior powers behave for locally free sheaves of finite constant rank, i.e. vector bundles. To build a slightly more general theory, we need some preliminaries to examine the interaction of exterior powers with base change. We continue from Exterior Powers 1.
Base change
Let be a ring homomorphism. Recall from algebra that we have a standard adjunction as follows: a -module is sent to , the same underlying abelian group where acts by , and an -module is sent to the base change where acts through the second factor. A few standard properties of this construction:
-
Given -modules and , we do not generally have
E.g., under the inclusion , so its restriction is , but by restricting first we get .
-
However, given -modules and , there is a canonical isomorphism
We have a map in one direction via the -linear map from to the RHS, sending . We have a map in the other direction because of the -bilinear map
which is an inverse to the map above.
Thus, base change behaves well with respect to tensor products, and so we expect it to behave well with respect to exterior algebra. We first focus on the module structure:
Theorem 1.¶ Let be a ring homomorphism and an -module. For each , there is a unique isomorphism such that
for all . Moreover, this isomorphism is natural with respect to maps of -modules, so that the diagram
commutes.
Proof. Such a map is obviously unique, and this uniqueness immediately implies that the naturality diagram commutes. To show existence, observe that the -linear unit maps induce a natural composite
sending , which is an -alternating map. Thus we have a map out of , and hence an adjunct
providing the unique map in the theorem statement. For the inverse, consider the -linear map
where the first map is the isomorphism coming from the remark above this proof, and the second is the base change of the canonical surjection. Then this satisfies
which is alternating in the tensor factors, so we obtain an induced map satisfying the above equation after replacing the outer tensors with wedges. This is a two-sided inverse as we can observe by plugging in . □
Corollary 2.¶ Let be an -module.
-
If is an ideal, then there is a canonical isomorphism
of -modules. For example, if is a maximal ideal, then
as -vector spaces.
-
If is a multiplicative subset, then there is a canonical isomorphism
of -modules.
Let us briefly remark that everything here works well with respect to the multiplicative structure on . Note that if is a ring homomorphism and is a graded-commutative -algebra, then we can form the base change as a graded-commutative -algebra by first forming it as a -module, and then declaring
as usual. Since preserves direct sums, this inherits a natural grading. It is clear from Theorem 1 that becomes an isomorphism of graded-commutative -algebras.
For ringed spaces
Now let be a ringed space. We will apply our constructions from the previous post to sheaves of -modules to construct a graded-commutative -algebra. We begin on the level of presheaves.
Let be a pre--module, i.e. an object of . For each , define a new premodule by
If , then induces a natural restriction map of -modules given on standard wedges by
This makes a pre--module. We claim it has the universal property that maps correspond to maps which are alternating, in the sense that they are alternating over all sections (note that here is just as a pre--module). Indeed, the canonical surjection is compatibly defined sectionwise, and from this the property is clear.
Note also that if is a morphism of premodules, then we get an induced map because we get one sectionwise which is compatible with restriction. This defines a functor .
Definition 3.¶ The functor is defined on the level of -modules by sheafifying, i.e. we have a commutative diagram
Exterior powers of sheaves are obtained by including sheaves among presheaves, taking the presheaf exterior power, and sheafifying. The induced sheaf functor is dashed.
Diagram TeX source
\begin{tikzcd}[sep=scriptsize]
{\Mod(\O_X)} & {\Mod(\O_X)} \\
{\PMod(\O_X)} & {\PMod(\O_X)}
\arrow["{\LA^k}", dashed, from=1-1, to=1-2]
\arrow[hook, from=1-1, to=2-1]
\arrow["{\LA^k}", from=2-1, to=2-2]
\arrow["L"', from=2-2, to=1-2]
\end{tikzcd}of functors where denotes sheafification.
The fundamental properties of sheafification tell us firstly that if is open, then , since this property holds on the level of premodules by definition. Secondly, to compute the stalks recall that we just need to compute the stalks of the corresponding premodule.
For a premodule, the maps induce maps of -modules given by , and since these are compatible with our restriction maps there is an induced map
of -modules for every containing , and hence of -modules. To see that this is an isomorphism, observe that the canonical surjection induces stalk maps
which are given by
for sections defined in a neighborhood of . If any two classes on the LHS are equal, then the sections agree in a small enough neighborhood, showing that the RHS is zero. Thus we get a map which is a two-sided inverse to the natural one defined above.
Proposition 4.¶ Let be a ringed space and be an -module.
-
For every open set , we identify .
-
For each , the canonical map induced by restriction is an isomorphism of -modules.
-
There is a unique surjection of -modules which on stalks corresponds to the canonical surjection under (2).
-
has the universal property that composition with the map of (3) induces a correspondence between maps and alternating maps , i.e. maps so that the induced map
on stalks is an alternating map of -modules.
Proof. We already proved (1) and (2). The surjection (3) comes from the fact that we have a surjection of premodules, where is the underlying presheaf of , and then sheafifying both sides gives us the desired map. This has the desired property by construction of the isomorphism in (2).
To prove the universal property, note that a morphism of sheaves is the same as a morphism of presheaves where is the underlying presheaf of . This is the same as a sectionwise alternating map , which obviously implies that it is stalkwise alternating. On the other hand, if is stalkwise alternating, then for each and ,
(under the identification ) is zero if two of the sections are equal. Thus in that case, because is a sheaf. □
Let us note that just as for tensor products, we have the canonical sheafification unit which is a morphism of premodules inducing an isomorphism on stalks (where again is the underlying presheaf). In particular, for each there is a map , so given sections , we can always form their wedge product (which is alternating) and get a section over of . Moreover, every section restricts to a finite linear combination of such wedges on an open cover. Two sections of the presheaf with the same image in its sheafification agree locally.
Now let us construct the graded sheaf . If is a premodule, then we have a bilinear multiplication map
because it exists sectionwise and is natural with respect to the restriction maps. Sheafifying, we obtain a multiplication map
Using the sectionwise wedge product defined above, and the sectionwise tensor product, this is the unique map satisfying
on all sections. This recovers the wedge product if we take stalks. Thus, defining
this induces a multiplication map
which on stalks recovers the wedge product . This makes a graded-commutative sheaf of -algebras, because we can check all properties of this multiplication stalkwise. Usually, we will focus on properties of the th powers themselves. As a corollary of our earlier theory, we do have the following important result:
Proposition 5.¶ Let be -modules. For each there is a canonical isomorphism
of -modules.
Proof. We could simply use the fact that it's obvious for premodules, and then sheafify that result. But since it's nice to have an explicit description of the map in one direction, it is simply the one induced by
which on stalks is the same corresponding map of -modules. The resulting direct-sum map is an isomorphism on stalks, so the one above is an isomorphism. □
Before specializing to the locally free case, we will consider the behavior under base change.
Proposition 6.¶ Let and be ringed spaces and a morphism, and let .
-
If are sheaves of rings on , is a morphism, and is an -module, then there is a canonical isomorphism
of -modules.
-
If is an -module, then there is a canonical map
of -modules.
-
If is an -module, then there is a canonical isomorphism
of -modules. In particular, there is a canonical isomorphism
of -modules which induces the isomorphism
of Theorem 1 on all stalks.
Proof. (1) We get a map in one direction by supplying , a map of -modules, which comes from taking tensor powers and precomposing with the -module map . Now we may check on stalks and conclude by Theorem 1.
(2) If we take the th tensor power of the counit map , we obtain a map
which is the left adjunct to a map . By applying to the standard surjection and composing, we get
which clearly has the alternating property, so it induces the map we wanted.
(3) The unit induces a map , so composing with the map of (2), applied to on the ringed space , we obtain . This is the right adjunct of a map . Taking stalks, we obtain the isomorphisms
from earlier. The second statement is simply applying (1) to this with the morphism of sheaves of rings. □
Now, keep as a ringed space and let be a free -module, so that there is a set of global sections of which induce an isomorphism . Assuming is ordered, we know that is a global section of , so we may form the corresponding map
This is an isomorphism because we know it's an isomorphism on stalks, by Theorem 3. Thus, more generally, if is a locally free sheaf, then is also locally free, trivialized on any open cover that is trivialized by, with local basis given as above.
We reserve the term vector bundle for a locally free sheaf of finite constant rank. If is a vector bundle of rank , we write for its top exterior power. Since taking exterior powers is compatible with restriction, we know that is a line bundle.
Proposition 7.¶ Let be a ringed space.
-
If is a morphism of vector bundles of the same rank, then is an isomorphism iff is surjective iff is an isomorphism of line bundles.
-
Suppose
is a short exact sequence of vector bundles on . Then there is a canonical isomorphism
In particular, induces a group homomorphism from the Grothendieck group of the exact category of vector bundles on .
Before we prove this, let us consider some general phenomenon first. But we are building towards (2), because for (1), the properties are checked pointwise, so we are immediately done because of the case of finite free modules of the same rank over a ring. Also, a simple consequence of (2) is as follows: since is a group homomorphism, we deduce the more general fact that if we have an exact sequence
of vector bundles, then
Let us now consider some generalities; we momentarily return to the case of -modules, so consider an exact sequence
of -modules. If we make the hypothesis that is generated by elements, then in fact there is a canonical map
which is slightly counterintuitive because this would seem to involve lifting elements along . To see this, let be some elements in and choose lifts of these along to . Then we have a linear wedging map
sending . Observe that if some is of the form for some , then this map is : on elementary wedges as above, up to a sign we can move this term to be adjacent to the terms, and then write the image as
but is zero because is generated by elements. So, if was another lift of , then for some , and then the above argument shows that
are equal as maps . Since this is still alternating, this defines a map
as desired. We can summarize this concisely as follows:
Lemma 8.¶ Suppose is a short exact sequence of -modules, and assume is generated by elements. Then there is a unique map
making the diagram
The top exterior power of M prime wedged with a lifted exterior power of M double prime factors uniquely through the dashed map to the exterior power of M.
Diagram TeX source
\begin{tikzcd}[sep=scriptsize]
{\LA^k M'\otimes_A\LA^l M} & {\LA^k M\otimes_A\LA^l M} \\
{\LA^k M'\otimes_A\LA^l M''} & {\LA^{k+l}M}
\arrow["{\LA^ki\otimes 1}", from=1-1, to=1-2]
\arrow["{1\otimes\LA^lp}", from=1-1, to=2-1]
\arrow["\wedge", from=1-2, to=2-2]
\arrow[dashed, from=2-1, to=2-2]
\end{tikzcd}commute for each . Moreover, suppose the sequence splits and is generated by elements. Then is generated by elements and the map is the isomorphism of Example 21.
Proof. All that remains is the second statement, and clearly if the sequence splits then is generated by elements. That the map is identified with the isomorphism of Example 21 follows from the fact that this map was induced by including the direct summands and taking the wedge product, which is exactly what providing a section of would do with the above diagram. □
Now we carry this over to sheaves. We say that an -module is generated pointwise by sections if is generated over by elements for all . We also say an exact sequence of -modules is pointwise split if at each point the corresponding exact sequence of -modules is split.
Corollary 9.¶ Suppose is a short exact sequence of -modules, and assume is pointwise generated by sections. Then for each , there is a unique map
making the diagram
The corresponding diagram for sheaves: the wedge map factors uniquely through the dashed map from the tensor product of exterior powers of F prime and F double prime.
Diagram TeX source
\begin{tikzcd}[sep=scriptsize]
{\LA^k \F'\otimes_{\O_X}\LA^l \F} & {\LA^k \F\otimes_{\O_X}\LA^l \F} \\
{\LA^k \F'\otimes_{\O_X}\LA^l \F''} & {\LA^{k+l}\F}
\arrow["{\LA^ki\otimes 1}", from=1-1, to=1-2]
\arrow["{1\otimes\LA^lp}", from=1-1, to=2-1]
\arrow["\wedge", from=1-2, to=2-2]
\arrow[dashed, from=2-1, to=2-2]
\end{tikzcd}commute. Suppose in addition the sequence is pointwise split, and that is pointwise generated by sections. Then is pointwise generated by sections, and the map is an isomorphism, given on stalks by the isomorphism of Example 21.
Proof. We have a collection of maps which satisfy this property by the previous result, so we just want to check that these glue together. We know that all sections of the domain are locally finite linear combinations of sections of the form
where and for some open . Now for each , there is a smaller neighborhood of on which all of the 's admit lifts to sections of , and thus we have a local section
such that for each . Since was arbitrary, these local sections agree on overlaps and we may glue and assume that is defined over . Now if instead of we started with a section over the tensor product which was only locally a finite linear combination of sections of the form above, then we would take the same linear combination of their images and obtain local images which agree on overlaps. Thus we can find such that for all . This is of course unique, so we have found our map. Note that the diagram commutes because it commutes stalkwise.
For the second statement, our hypotheses put us exactly in the case of the previous result at each point, yet now that we have the morphism of sheaves we are done. □
Note that this finishes the proof of Proposition 7 because a vector bundle is pointwise a free and hence projective module. Note that in fact our multiplicativity of determinants applies slightly more generally, to the case where the sheaves aren't actually vector bundles of finite rank, but they are just pointwise free modules of constant finite rank. But, e.g. for coherent sheaves on schemes, such a thing is a vector bundle anyway.
For schemes
Now let us study the special case of quasicoherent sheaves on a scheme, where we can give some applications of the theory developed above. The fundamental observation is that for a quasicoherent -module, the exterior powers remain quasicoherent, and they are given on affine opens by exterior power modules, not just pointwise.
Theorem 10.¶ Let be an -module. There is a natural isomorphism
of -modules for each . In particular, if is a scheme, is a quasicoherent -module, and is an affine open, then there is a canonical isomorphism
for all .
Proof. The second statement follows immediately from the first, and the fact that taking exterior powers commutes with restriction. For the first statement, we use the corresponding fact for tensor products to obtain a map
which is locally alternating and thus induces . At each prime , the stalk of this map is
which is an isomorphism of -modules by Corollary 2(2). Naturality is clear because of the universal properties defining this map. □
One often computes exterior powers by embedding a scheme into a larger space and computing the module there. To do this, we want to have good behavior with respect to restriction. We first state the very classical projection result. Let be a morphism of ringed spaces, be an -module and be an -module. Then the canonical map
(where the second map is tensoring with the counit) induces an adjunct map
called the projection map, clearly natural in both and .
Proposition 11.¶ Let be a morphism of ringed spaces, be an -module and be an -module. Suppose we are in one of the following circumstances:
- is a closed embedding on underlying spaces; or
- is locally free of finite rank; or
- is an affine morphism of schemes, and and are quasicoherent.
Then the projection map Equation (1) is an isomorphism.
Proof. (1) Using the explicit description of stalks of a closed embedding, both sides are supported in , and at the point the map on stalks is
which is an isomorphism by the commutativity and cancellation properties of tensor products.
(2) Let be an open set on which . Restricting Equation (1) to , we obtain the map
after applying naturality, which is an isomorphism because preserves finite direct sums.
(3) By restricting to an affine open, we reduce to showing that if is a ring homomorphism, is an -module and is a -module, then the natural map
is an isomorphism of -modules. This is of course the case because the cancellation property identifies the RHS with where is a -module through the action on . Since the tensor product is computed using only its -module structure, this is equal to . □
Of course, this is quite a bit more generality than we usually need, since in fact we normally apply this to closed immersions with and a line bundle. In any case, the corresponding result doesn't hold for exterior powers, but we do at least have a formula for twisting with line bundles:
Lemma 12.¶ Let be a ringed space and a line bundle on . For any -module , there is a canonical isomorphism
of -modules for each .