Patrick Lei

All notes

Hodge theory

Talks by Kevin Chang and Caleb Ji, Columbia, Fall 2021.

Seminar on Mixed Hodge Structures
Fall 2021
Notes by Patrick Lei
Lectures by Various

Columbia University

Disclaimer

These notes were taken during the seminar using the vimtex package of the editor neovim. Any errors are mine and not the speakers’. In addition, my notes are picture-free (but will include commutative diagrams) and are a mix of my mathematical style and that of the lecturers. If you find any errors, please contact me at plei@math.columbia.edu.

1 Kevin (Oct 28): Intro to mixed Hodge structures; Hodge theory for smooth varieties

Throughout the seminar, we work over ℂ like reasonable people.

1.1 Review of Hodge theory

Let X be a smooth projective variety. Then we have the Hodge decomposition

Here, Hp,q contains forms of type (p,q). Deligne constructs a mixed Hodge structure on smooth varieties in Théorie de Hodge II and on all varieties in Théorie de Hodge III.

Definition 1.1.1.

Let A be one of ℤ,ℚ,ℝ. A Hodge structure over A can be defined in several ways:

  1. 1.

    A Hodge structure of weight n is a finitely-generated A-module H with a decomposition

    satisfying Hp,q=H¯q,p.

  2. 2.

    A Hodge structure of weight n is a finitely-generated A-module H with a decreasing filtration F∗∙Hℂ that is n-opposite in the sense that if p+q=n+1, then Fp∩F¯q=0 and Fp⊕F¯q=Hℂ.

  3. 3.

    A Hodge structure of weight n is a finitely-generated A-module H with an action of the Deligne torus 𝕊≔Resℝℂ⁡ℂ× on Hℝ of weight n.

Remark 1.1.2.

In our definition, p,q,n can all be negative.

To see that these definitions are equivalent, from a decomposition ⨁p+q=nHp,q we take the filtration

In the other direction, for a filtration F, we take

From now on, when we say Hodge structure we will take A=ℤ.

Definition 1.1.3.

A morphism of Hodge structures of weight n (H,F)→(H′,F′) is a morphism of modules that respects the filtration after tensoring with ℂ.

Tensor products of Hodge structures are Hodge structures after adding the weights, and the dual of a Hodge structure is still a Hodge structure after taking the negative of the weight.

Examples 1.1.4.
  1. 1.

    Let X be a smooth projective variety. Then by the Hodge theorem, Hn⁢(X,ℤ) is a Hodge structure of weight n.

  2. 2.

    Let X be a smooth proper variety. Then Hn⁢(X,ℤ) is a Hodge structure of weight n by a result of Deligne.

  3. 3.

    (Tate-Hodge structure) We will define a Hodge structure ℤ⁡(1) of weight −2 where the underlying ℤ-module is 2⁢π⁢i⁢ℤ and ℤ⁡(1)⊗ℤℂ=H−1,−1. We will denote by ℤ⁡(n) the n-th tensor power of ℤ⁡(1). We may twist Hodge structure by ℤ⁡(n) to change the weight.

    If Z↪X is a subvariety of a smooth projective variety of codimension r, there are Gysin maps

    This is not a map of Hodge structures by weight reasons, but after twisting by ℤ⁡(−r) we obtain a morphism

    that is a morphism of Hodge structures.

    Now set U≔X∖Z. Then we have a long exact seqeunce

    We will introduce a new increasing filtration W on Hn⁢(U) where

    Taking the associated graded module

    Now W has allowed us to isolate the weight n part and the weight n+1 part, and now we want to consider mixed Hodge structures.

1.2 Mixed Hodge structures

Definition 1.2.1.

A mixed Hodge structure is a finitely-generated ℤ-module H together with two filtrations:

  • •

    The weight filtraation W, which is an increasing filtration on Hℚ;

  • •

    The Hodge filtration F, which is a decreasing filtration on Hℂ.

These are required to be compatible in the sense that for all n, the associated graded piece

is a Hodge structure over ℚ.

Example 1.2.2.

Let H be an ordinary Hodge strucure of weight n. This is a mixed Hodge structure, where we take Wn−1⁢Hℚ=0 and Wn⁢Hℚ=Hℚ.

Definition 1.2.3.

A morphism of mixed Hodge structures f:(H,W,F)→(H′,W′,F′) is a morphism of ℤ-modules H→H′ that is compatible with both filtrations.

Theorem 1.2.4.

Any morphism of mixed Hodge structures is automatically strictly compatible with both filtrations, which means that if h′∈Im(f)∩Wn′⁢(H′), there exists h∈Wn⁢(H) such that f⁡(h)=h′. The same result is true for F,F′.

Remark 1.2.5.

The category of mixed Hodge structures is an abelian category.

Theorem 1.2.6 (Main theorem of Hodge II).

Let U be a smooth variety. Then Hn⁢(U,ℤ) is functorially a mixed Hodge structure.

1.3 Hypercohomology spectral sequences

Before we discuss the proof of the main theorem, we will discuss this important tool. Let X be a smooth projective variety. Then recall that

This implies that the Hodge-de Rham spectral sequence degenerates. Recall1 that the Hodge-de Rham spectral sequence is given by

This is a hypercohomoology spectral sequence.

Definition 1.3.1.

Let (K,F) be a filtered complex of abelian sheaves with F decreasing and finite on each component. Then the hypercohomology spectral sequence for (K,F) is given by

In our case, the Hodge-de Rham spectral sequence is the hypercohomology spectral sequence for (ΩX∙,F), where F is the filtration bête and is defined by

This gives us E1p,q=𝕓⁢ℍp+q⁢(X,ΩXp⁢[−p]). Using the holomorphic Poincaré lemma, the map C→ΩX∙ is a quasi-isomorphism, and this gives us the Hodge-de Rham spectral sequence.

We will now say some things about filtered complexes.

  1. 1.

    If (K,F)→(K′,F′) is a morphism of filtered complexes, there is an induced map of hypercohomology spectral sequences.

  2. 2.

    If (K,F)→(K′,F′) is a filtered quasi-isomorphism (which means it induces quasi-isomorphisms on all associated graded complexes), then the map of hypercohomology spectral sequences is an isomorphism.

  3. 3.

    F induces a filtration on Hk⁢(X,K) given by

  4. 4.

    Every complex K carries a canonical filtration τ, which is an increasing filtration and is given by

    In this case, the associated graded pieces are given by Grpτ⁡K=Hp⁢(K)⁢[−p]. If K→K′ is a quasi-isomorphism, then (K,τK)→(K′,τK′) is a filtered quasi-isomorphism. Thus we can discuss canonical filtrations of items in the derived category.

1.4 Hodge theory for smooth varieties

Let U be a smooth variety. We will find a good compactification of U:

  1. 1.

    First we consider the Nagata compactification X′.

  2. 2.

    Then resolve the Nagata compactification using resolution of singularities, and now we have a good compactification X.

Definition 1.4.1.

A compactification X of U is good if D=X∖U is a simple normal crossings divisor. This means that D locally analytically looks like a union of hyperplanes.

Now that we have a good compactification, we know D is a union of smooth irreducible hypersurfaces

For all I⊂[s], write DI=⋂i∈IDi. If I is empty, then D∅=X. Now write D(i)≔⨆|I|=iDi. Also denote the inclusion j:U↪X.

We want to study H∗⁢(U,ℂU)=𝕓⁢ℍ∗⁢(U,ΩU∙)=𝕓⁢ℍ∗⁢(X,Rj⁣∗⁢ΩU∙). But now the higher pushforwards vanish, and so we obtain 𝕓⁢ℍ∗⁢(X,j∗⁢ΩU∙). Our strategy will be to study the logarithmic de Rham complex

which are defined as follows:

  • •

    Logarithmic 1-forms are locally spanned by dzizi for 1≤i≤t and dzi for t+1≤i≤n. Here, we choose coordinates such that D={z1⁢⋯⁢zt=0}.

  • •

    Logarithmic k-forms are given by ΩXk⁢(log⁡D)=⋀kΩX1⁢(log⁡D).

In fact, the inclusion ΩX∙⁢(log⁡D)↪j∗⁢ΩU∙ is a quasi-isomorphism, so we can instead study the hypercohomology 𝕓⁢ℍ⁢(X,ΩX∙⁢(log⁡D)).

2 Kevin (Nov 04): Hodge theory for smooth varieties II

Recall that last time, we considered a smooth compactification X⊇U of a smooth variety, and the boundary is a simple normal crossings divisor. We have this chain of quasi-isomorphisms

Therefore it suffices to study 𝕓⁢ℍ∗⁢(X,ΩX∙⁢(log⁡D)). We are still atempting to prove Theorem 1.2.6.

2.1 Setting everything up

Definition 2.1.1.

The weight filtration W on ΩX∙⁢(log⁡D) is defined by

where α is a holomorphic form on X and z1,… are the local equations for D.

Recall the definitions of DI,D(i) from last time. We will denote the natural maps aI:DI→X and ai:D(i)→X.

Definition 2.1.2.

Let I⊂[s] with |I|=m. The residue map resI:Wm⁢ΩX∙⁢(log⁡D)→aI⁣∗⁢ΩDI∙⁢[−m] is defined by

Here, α is holomorphic, DI={z1⁢⋯⁢zm=0}, and α′ has fewer poles.

The important fact is that resI vanishes on Wm−1 and therefore descends to a map

Lemma 2.1.3.

Define

This map is an isomorphism.

Corollary 2.1.4.
  1. 1.

    We have the identity

  2. 2.

    The map (ΩX∙⁢(log⁡D),W)→id(ΩX∙⁢(log⁡D),τ) is a filtered quasi-isomorphism.

Now we want to upgrade our known quasi-isomorphisms to filtered quasi-isomorphisms. By the previous discussion, we now have filtered quasi-isomorphisms

This identifies all of the hypercohomology spectral sequences. The final geometric fact that we need is the following:

Lemma 2.1.5.

The identification Rm⁢j∗⁢ℂU≅am⁣∗⁢ℂD(m) is in fact defined over ℤ. This means that the diagram

commutes.

2.2 An exercise in homological algebra

The first thing we will do is consider the weight spectral sequence, which is just the hypercohomology spectral sequence for (ΩX∙⁢(log⁡D),W). Here, we have

But we can identify the E1-page with

But now we see that everything is actually defined over ℚ, so we now have a spectral sequence

Everything here actually has a Hodge structure. But now the bête filtration F induces filtrations on the weight spectral sequence:

  1. 1.

    (first direct filtration) (Fd);

  2. 2.

    (second direct filtration) (Fd∗);

  3. 3.

    (inductive filtration) (Fr).

Some facts about these filtrations are:

  1. 1.

    Fd⊂Fr⊂Fd∗;

  2. 2.

    The differentials dr are compactible with the direct filtrations;

  3. 3.

    If dr is strictly compatible with Fr for r∈{0,…,r0−3}, then Fd=Fr=Fd∗ for all r∈{0,…,r0}. Here, f:(H1,F1)→(H2,F2) is strict if for all a∈F2p∩Imf, then f∈ImF1p. The important thing is that all morphisms of mixed Hodge structures are strict.

  4. 4.

    This is basically just the previous fact for r0=∞.1

We are now ready to state the main theorem:

Theorem 2.2.1.
  1. 1.

    Fd=Fr=Fd∗ for all terms of the weight spectral sequence.

  2. 2.

    The filtration W on Hk⁢(U,ℂ) comes from a filtration on Hk⁢(U,ℚ). Moreover, neither W nor F depends on the choice of compactification X.

  3. 3.

    The filtrations W⁡[k] defined by W⁢[k]m=Wm−k and F make Hk⁢(U,ℤ) a functorial mixed Hodge structure.

Lemma 2.2.2.

The hypercohomology spectral sequence for (GrmW⁡ΩX∙⁢(log⁡D),F) degenerates at the E1-page. Moreover, the induced filtration on E1−p,q≅H−2⁢p+q⁢(D(p),ℂ) is q-opposite.

Proof.

We know that (Grmw⁡ΩX∙⁢(log⁡D),F)≅am⁣∗⁢(ΩD(m)∙,F)⁢[−m]. By the Hodge decomposition for D(m), we obtain degeneration at E1, and by the Hodge structure on H−2⁢p+q⁢(D(p),ℂ) we obtain q-oppositeness. □

Lemma 2.2.3.

The differential d1 is strictly compatible with F.

Proof.

Because d1 is a morphism of Hodge structures, we are done. □

Lemma 2.2.4.

The filtration Fr on E2−p,q is q-opposite.

Proof.

We know that E2−p,q=ℋ⁡(E1−p−1,q→E1−p,q→E1−p+1,q). Because Fr is the induced filtration, we are done by the previous lemma. □

Lemma 2.2.5.

For all r≥0, dr is strictly compatible with Fr. Moreover, for all r≥2, dr=0.

Proof.

The r=0 case follows from the first lemma. For r=1, this is the second lemma. Finally, to prove degeneration, we can simply proceed by induction. We know that Fd=Fr=Fd∗ on Er. But now by the second fact, we know that dr is compatible with Fr. Thus we have

By the final fact, we see that

is a subquotient of H−2⁢p+q⁢(D(p),ℚ)⁢(−r). Therefore, (Hk⁢(U,ℚ),W⁡[k],F) is a mixed Hodge structure.

Finally, we discuss the functoriality of the mixed Hodge structure. To do this, we want the filtrations

to be functorial. For U→V, we want to prove that Hk⁢(U,ℂ) is functorial. We also compactify the morphism to X→Y. But now we know that Hk⁢(V,ℂ)→Hk⁢(U,ℂ) comes from

All we need to do is to show that there is a compactification that works, and this is given by resolving X and taking Y to be the closure of the graph.

To prove independence of the compactification, suppose X1,X2 are compactifications. Then let X′ resolve X1,X2 simultaneously, and then X′→U¯⊂X1×X2 is a resolution of singularities. By functoriality, the isomorphism

is an isomorphism of mixed Hodge structures.

3 Caleb (Nov 11): Cohomological descent

Definition 3.0.1.

A simplicial object in a category C is a functor Δop→C, where Δ is the category with objects [n] for n≥0 and morphisms are order-preserving set maps. There are distinguished functions di:[n−1]→[n] skipping i and si:[n+1]→[n] repeating i.

The maps di,si induce maps di:Xn→Xn−1 and Si:Xn→Xn+1 on a simplicial object. We will refer to simplicial objects in C by S⁡(C) and denote Sn⁢(C)≔Δ≤nop→C and S+⁢(C)≔Δ≥−1op→C given by adding [−1]=∅ to Δ.

For various reasons, we will assume that C has all finite products and fiber products. If this is not sufficient, add assumptions until what is written here makes sense.

3.1 Coskeleta

Given a simplicial object X∙∈S⁡(C), let skn⁢(X)∈Sn⁢(C) be the n-th truncation of X∙.

Definition 3.1.1.

Let Y∙∈Sn⁢(C). Then the coskeleton functor is defined by

Example 3.1.2.

Let n=0 and Y0∈C. Given X0→Y0, consider

We obtain maps X0→Y0×Y0 and X2→Y0×Y0×Y0, and so on. Then we define

and give it the natural simplicial structure.

For an example of a simplicial object, if Y−1 is a topological space and Y0=⨆Ui, where Ui form an open cover of Y−1, then the Yi are the i-intersections.

Now choose Y∙∈Sm⁢(C). Set

Given a map skn⁢(X∙)→Y∙, we need to define a map Xn→lim[k]→[n]Yk. For every φ:[k]→[n] and k≤m, we have a morphism

for k≤n.

Proposition 3.1.3.

The map X∙←skm⁢(coskm⁢(X∙)) is an isomorphism.

The proof essentially follows from the construction. For n≥m, the single copy of Yn corresponding to [n]→[n] determines everything.

Remark 3.1.4.

The skeleton skm also has a left adjoint, which is the inclusion Sn⁢(C)→S⁡(C).

The adjunction id=coskm⁢skm is not generally an isomorphism, but is an isomorphism for n-coskeleta when 0≤n≤m.

3.2 Hypercoverings

Definition 3.2.1.

Let P be a class of morphisms containing all isomorphisms and stable under base change and composition. Then a simplicial object X∙ of C is a P-hypercovering if for all n≥0 (or n≥−1), the natural adjunction X∙→coskn⁢(skn⁢(X∙)) induces Xn+1→coskn⁢(skn⁢(X∙))n+1 is in P.

Example 3.2.2.

The map cosk0⁢(S′/S)→S is a P-hypercovering if and only if S′→S is in P.

For example, if P is a surjective condition for topological spaces, then S′=⨆Ui↠S obtains the Čech construction.

We may compute cohomology using hypercoverings. Define Hˇi⁢(S,ℱ)=Hi⁢(Moore⁡(ℱ⁡(S))), where the Moore complex was defined in a different seminar by Caleb. If we take the limit over all hypercoverings, we recover the derived functor cohomology.

We are most interested in hypercoverings of singular schemes, which we will be used to construct Hodge structures.

Definition 3.2.3.

A simplicial object X∙∈S⁡(C) is split if there exist subobjects N⁢Xj with isomorphisms

Remark 3.2.4.

In the case of simplicial abelian groups, we have

Under nice conditions, split objects have unique splittings. A nice fact is that given split n-truncated objects skn⁢(X) and N⁢Xn+1→(coskn⁢(skn⁢(X)))n+1, we can recover Xn+1. This is because we know that

and the map to the coskeleton tells us what the maps out of Xn+1 are.

Theorem 3.2.5.

If skn⁡X is split, then there exists f:X′→X with skn⁡(f) an isomorphism and X′ split. Furthermore, if X is an augmented P-hypercover, so is X′.

From now on, to be safe, we will assume that we are in one of the following cases:

  1. 1.

    C=𝖳𝗈𝗉/B with P the proper surjective morphisms;

  2. 2.

    C is the category of schemes étale over some base with P the surjective étale morphisms;

  3. 3.

    C is a topos with P the epimorphisms.

Corollary 3.2.6.

For m≥0 and an augmented m-truncated P-hypercovering Z in one of the first two situations, the face and degeneracy maps for Z are proper or étale. If X∙→S is a proper or étale hypercovering, then all structure maps are proper or étale.

Theorem 3.2.7.

Let S be a separated scheme over a field k. Then there exists a dense open immersion S↪S¯ into a proper k-scheme and an augmented proper hypercovering X¯∙→S¯ such that X¯n is a projective regular k-scheme and the part of X¯n lying over S¯∖S is a (strict) normal crossings divisor.

In de Jong’s alterations paper, he mentions that this result as an application. Of course, in characteristic 0, we simply need resolution of singularities, but in positive characteristic, de Jong’s alterations theorem is needed.

Proof.

First, to construct the open immersion, we use Nagata’s compactification theorem. Of course, if we are interested in quasiprojective varieties, then this is unnecessary. The upshot is that now we have a dense open immersion S↪S¯ into a proper k-scheme.

By Johan, we obtain a regular X¯0→S¯ wich is a proper surjection, where the preimage of S¯∖S is a strict normal crossings divisor. Now given X≤m, we know that coskm⁡Xm is a proper hypercovering of S. We know that each term is S-proper. By Johan applied to (coskn⁡X≤m)m+1, we obtain X¯′ proper over (coskm⁡X≤m)m+1 such that the part over S¯∖S is a normal crossings divisor.

Now set N⁢Xm+1=X¯′ and take the construction

and here we obtain an (m+1)-truncated solution. By induction, we are done. □

3.3 Cohomological descent

Let C be a site with the topology generated by E-morphisms for some class of morphisms E. Let X∙ be a simplicial object of C. Set X~∙ to be the category sheaves on the following site:

  • •

    The objects are E-morphisms U→Xn;

  • •

    The morphisms are commutative squares

    where f is any morphism in C;

  • •

    A covering of [Ui→Xn] is given by any covering of Ui.

This is meant to formalize the notion of sheaves on X∙. A map U∙:X∙→Y∙ induces maps U∙⁣∗:X~∙→Y~∙ and U∙∗:Y~∙→X~∙.

Now let a:X∙→S. We have a∗:S~∙→X~∙. Here, we have (a∗⁢ℱ)n=an∗⁢ℱ. Also, we have a∗:X~∙→S~∙, and a∗⁢ℱ∙ is the equalizer of

On the level of derived categories, we obtain a∗:D+⁢(S)→D+⁢(X∙) and R⁢a∗:D+⁢(X∙)→D+⁢(S).

Definition 3.3.1.

A morphism a:X∙→S is a morphism of cohomological descent if the natural transformation id→R⁢a∗∘a∗ on D+⁢(S) is an isomorphism.

Note that the adjunction of (a∗,R⁢a∗) means that equivalently, a∗:D+⁢(S)→D+⁢(S∙) is fully faithful.

Definition 3.3.2.

A morphism a is universally of cohomological descent if every base change aS′:X∙×SS′→S′ is of cohomological descent.

Definition 3.3.3.

A map of spaces a0:X0→S is a map of cohomological descent if cosk0⁡(a0):cosk0⁡(X∙/S)→S is of cohomological descent.

Theorem 3.3.4.

Morphisms universally of cohomological descent form a Grothendieck topology.

We now discuss some properties and applications. Unfortunately, because of time reasons, we cannot review any descent theory.

First, the condition that id→R⁢a∗∘a∗ is an isomorphism on D+⁢(S) is equivalent to the isomorphism

and Ri⁢a∗⁢(a∗⁢ℱ)=0. Note that this is a derived version of ordinary descent theory.

Theorem 3.3.5.

Let X∙ be a simplicial space. For any complex K′ in D+⁢(X∙), there exists a spectral sequence

On the 0-skeleton, we have

Theorem 3.3.6.

Let X→S be a proper surjective of either topological spaces or schemes with the étale topology. Then X→S is univerally of cohomological descent.

Theorem 3.3.7.

Let X∙→S be a proper hypercovering. Then it is universally of cohomological descent.

All of these theorems have extremely long proofs.

4 Kevin (Nov 18): Applications and examples of mixed Hodge structures; Hodge theory for all varieties

4.1 End of Hodge II

The main theorem is as follows:

Theorem.

The weight spectral sequence

degenerates at the E2 page. The Hodge structures on E2 induce a mixed Hodge structure on H−p+q⁢(U,ℚ).

Corollary 4.1.1.

If Hk⁢(U,ℂ) has nonzero weight (p,q), then 0≤p,q≤k and k≤p+q≤2⁢n.

Proof.

We consider the spectral sequence. All terms that do appear must lie above a line of slope −2.1 The lines of slope −1 are the cohomology, and so by staring at this for long enough, we obtain the desired result.2 □

Corollary 4.1.2.
  1. 1.

    If X is any smooth compactification of U, then Im(Hk⁢(X,ℚ)→Hk⁢(U,ℚ)) is the bottom weight part Wk⁢(Hk⁢(U,ℚ)).

  2. 2.

    If Y→U↪X is any morphism with Y smooth and proper, then

Proof.
  1. 1.

    We will use the rising sea (Grothendieck) fact that Im(Hk⁢(X,ℚ)→Hk⁢(U,ℚ)) does not depend on the choice of X. Therefore we may assume that X∖U is a simple normal crossings divisor. Now the weight spectral sequence for X maps to the weight spectral sequence for U. But recall that the weight spectral sequence for X only consists of Hk⁢(X) in the rightmost column, while the weight spectral sequence for U has nontrivial things to the left. Therefore,

  2. 2.

    Recall that maps of mixed Hodge structures are strict. This implies that

    By the previous part of this corollary, we are done. □

Theorem 4.1.3 (Global invariant cycles theorem).

If f:U→S is smooth and proper with S smooth and separated, and X is a smooth compactification of U, then Im(Hk⁢(X,ℚ)→Hk⁢(Us,ℚ))=Hk⁢(Us,ℚ)π1⁢(S,s).

Proof.

We will only prove the projective case. First, we use the fact that Hk⁢(U,ℚ)↠Hk⁢(Us,ℚ)π1⁢(S,s). This follows from the degeneration of the Leray spectral sequence for smooth projective morphisms, which gives

Using the second corollary for Us↪U↪X, we obtain

using the previous corollary. In the general case, we dominate U by something projective. □

Remark 4.1.4.

Using the theorem, we can prove the degeneration of the Leray spectral sequence for smooth proper morphisms.

Now we will do some computations.

  1. 1.

    Consider U=ℂ×↪X=ℙ1. Then D=pt⊔pt. The only nonzero differential is

    and by the cohomology of ℂ×, we obtain a ℚ in the (0,0) position and a ℚ in the (−1,2) position. Thus H0 is a pure Hodge structure, but the first cohomology H1⁢(ℂ×,ℚ)=Gr2W⁡H1⁢(U,ℚ) actually lives in weight 2.

  2. 2.

    Consider a punctured Riemann surface3 U=Rg∖D, where D is a disjoint union of p points. There is only one nontrivial differential, which is surjective, so

4.2 Hodge III

Kevin claims that despite the paper being 70 pages, the actual content is only two pages. Today, we will discuss the other 68 pages.

Our goal is to construct a mixed Hodge structure on Hk⁢(X,ℚ) for any complex variety X. Our strategy is to take a proper hypercovering Y∙→X such that each Yn is smooth. We will use cohomological descent, which will tell us that

The right hand side is approximated by Hk⁢(Yn,ℚ) (in the sense that there is some spectral sequence converging to Hk⁢(Y∙,ℚ)), and each of the terms has a mixed Hodge structure, which will induce a mixed Hodge structure on Hk⁢(Y∙,ℚ).

Now recall the following from Caleb’s lecture:

  • •

    A simplicial variety is a sequence of varieties Y∙ with face and degeneracy maps (a simplicial object in the category of varieties).

  • •

    An augmented simplicial variety Y∙→X is simply a simplicial variety Y∙ with a map Y0→X.

  • •

    A sheaf ℱ∙ on a simplicial variety is a collection of sheaves ℱn on each Yn equipped with maps ℱ∙⁢(p):ℱn→p∗⁢ℱm, where p:[n]→[m] is a map in the simplex category. Of course, these maps are required to be compatible.

Remark 4.2.1.

If S∙ is a constant simplicial variety, then a sheaf on S∙ is a cosimplicial sheaf on S.

Given a map f:Y∙1→Y∙2, we can define f∗,f∗ pointwise. If we have an augmentation a:Y∙→X, then we define (a∗⁢ℱ)n=an∗⁢ℱ, while we define

When X is a point, then a∗=Γ is the global sections functor. We may then define the right derived functors

Definition 4.2.2.

A proper hypercovering Y∙→X is a hypercovering such that

is proper and surjective for all n≥−1. For n=−1, this says that Y0→X is a proper surjection.

Theorem 4.2.3 (SGA4).

Proper hypercoverings satisfy cohomological descent. This means that ℱ→R⁢a∗⁢a∗⁢ℱ is an isomorphism.

Corollary 4.2.4.

If Y∙→X is a proper hypercovering, then

Thus we can replace X by some simplicial resolution (proper hypercovering) Y∙→X and then do all of our computations on Y∙. Then whatever structure we have on Y∙ will descends to X.

4.3 Mixed Hodge complexes

Apparently all of the homological algebra we did in Hodge II is very general, so we can rip it off.4 This is exactly what we will do now.

Definition 4.3.1.

The filtered derived category D⁢F⁢(C) of an abelian category C is constructed by taking the category of filtered complexes in C, identifying homotopic maps (that respect the filtrations), and inverting filtered quasi-isomorphisms.

The bifiltered derived category D⁢F2⁢(C) is defined in the same way, except we have two filtrations.

Definition 4.3.2.

For A∈{ℤ,ℚ,ℝ}, an A-Hodge complex of weight n is the following data:

  1. 1.

    A complex KA∈D+⁢(A) such that Hk⁢(KA) is finitely generated;

  2. 2.

    A filtered complex (Kℂ,F)∈D+⁢F⁢(ℂ) (where F is a decreasing filtration) and an isomorphism α:KA⊗Aℂ→Kℂ in D+⁢(ℂ).

This data must satisfy the following conditions:

  1. 1.

    The differential d on Kℂ is strict with respect to F.

  2. 2.

    For all k, (Hk⁢(KA),F) is a Hodge structure of weight n+k.

Definition 4.3.3.

Let X be a space. An A-Hodge complex of sheaves of weight n is the following data:

  1. 1.

    A complex KA∈D+⁢(X,A);

  2. 2.

    A filtered complex (Kℂ,F)∈D+⁢F⁢(X,ℂ) and an isomrphism α:KA⊗Aℂ→Kℂ.

We also require that (R⁢Γ⁢(KA),R⁢Γ⁢((Kℂ,F)),R⁢Γ⁢(α)) is a Hodge complex of weight n.

Example 4.3.4.

If X is smooth and proper, then (ℤX,(ΩX∙,F),α) is a Hodge complex of sheaves of weight 0. Here, α is the map ℂX≃ΩX∙.

Definition 4.3.5.

A mixed Hodge complex is the following data:

  1. 1.

    A complex K∈D+⁢(ℤ) such that Hk⁢(K) is finitely generated;

  2. 2.

    A filtered complex (Kℚ,W)∈D+⁢F⁢(ℚ) (with W an increasing filtration) and an isomorphism α:K⊗ℤℚ≅Kℚ;

  3. 3.

    A bifiltered complex (Kℂ,W,F)∈D+⁢F2⁢(ℂ) and an isomorphism of filtered complexes β:(Kℚ,W)⊗ℚℂ→(Kℂ,W).

We require that for all n, (GrnW⁡Kℚ,(GrnW⁡Kℂ,F),β) is a ℚ-Hodge structure of weight n.

Definition 4.3.6.

A mixed Hodge complex of sheaves is the following data:

  1. 1.

    A complex K∈D+⁢(X,ℤ) such that 𝕓⁢ℍk⁢(X,K) is finitely generated;

  2. 2.

    A filtered complex (Kℚ,W)∈D+⁢F⁢(X,ℚ) and an isomorphism K⊗ℤℚ≃Kℚ;

  3. 3.

    A bifiltered complex (Kℂ,W,F)∈D+⁢F2⁢(X,ℂ) and an isomorphism of filtered complexes β:(Kℚ,W)⊗ℚℂ→(Kℂ,W).

We require that for all n, (GrnW⁡Kℚ,(GrnW⁡Kℂ,F),β) is a ℚ-Hodge complex of sheaves of weight n.

Here are some facts about these things:

  1. 1.

    If (K,(Kℚ,W),α,(Kℂ,W,F),β) is a mixed Hodge complex of sheaves, then

    is a mixed Hodge complex.

  2. 2.

    For a smooth variety U, the data (R⁢j∗⁢ℤU,(R⁢j∗⁢ℚU,τ),α,(ΩX∙⁢(log⁡D),W,F),β) is a mixed Hodge complex of sheaves on X. Here, β is defined as the zigzag

  3. 3.

    The main theorem of Hodge II says that (Hk⁢(MHC),W⁡[k],F) is a mixed Hodge structure.

5 Kevin (Dec 02): Hodge theory for all varieties II

Recall our geometric setup, where we have a proper hypercovering Y∙→X of X by smooth varieties. We may assume that Y∙↪Y¯∙, where Y¯n=Yn⊔Dn and Dn is a simple normal crossings divisor. Because proper hypercoverings satisfy cohomological descent, we can work on Y∙. Thus we need to do Hodge theory on simplicial varieties.

5.1 Existence of mixed Hodge structures

Definition 5.1.1.

A mixed Hodge complex of sheaves on a simplicial space Y∙ is a tuple

of sheaves on Y∙ such that we have a mixed Hodge complex of sheaves on each Yn. Here, the first grading is the degree in the complex and the second is the (co)simplicial degree.

Example 5.1.2.

An example of a mixed Hodge complex of sheaves on Y¯∙ is given by

Taking the derived functor of global sections on each Y¯n, we obtain a cosimplicial mixed Hodge complex.

Now we wamt to obtain a complex (in the usual sense) with the same cohomology as a given cosimplicial complex. Suppose we have a sheaf ℱ∙ on Y¯∙. We find some injective resolution ℱ∙↪𝒦∙⁣∙. Next, we consider the cosimplicial complex

and this gives us a double complex K∙⁣∙ using the Moore complex. Finally, from a double complex we obtain an ordinary complex

with differential

Also, we have

and also R⁢Γ⁢(Y¯∙,ℱ∙)≃𝐬⁢R⁢Γ∙⁢(Y¯∙,ℱ∙).

Remark 5.1.3.

There exists a filtration L on 𝐬⁢K∙ given by Lr⁢(𝐬⁢K)∙=⨁q≥rKp⁢q. This gives us a spectral sequence

Somehow all of the homological algebra we did applies to bounded below complexes of sheaves. The good thing about this is that now 𝐬⁢R⁢Γ∙⁢R⁢j∗⁢ℤY∙ computes the cohomology of X. All we need to do is to prove that this is a mixed Hodge complex, and this means that we need to become Deligne.

Theorem 5.1.4 (Main theorem).

Let K∙⁣∙ be a cosimplicial mixed Hodge complex. Then there exists a natural filtration δ⁡(W,L) on 𝐬⁢K∙ such that (s⁢K∙,δ⁡(W,L),F) is a mixed Hodge complex.

Proof.

Define

Then we can compute

But now each Kℚ∙p is a mixed Hodge complex, and thus each of the associated graded pieces is a pure Hodge complex of weight n. This implies that 𝐬⁢K∙ is a mixed Hodge complex. □

Applying the main theorem to 𝐬⁢R⁢Γ∙⁢R⁢j∗⁢ℤY∙, we see that Hn⁢(X,ℤ) is a mixed Hodge structure. This is functorial because it is possible to construct compatible proper hyperresolutions, and this implies that it is unique.

If X is smooth, we can choose the proper hyperresolution to simply be X itself, so we have the same mixed Hodge structure as before. From the mixed Hodge complex, we have a weight spectral sequence

This spectral sequence degenerates at E2, and the differentials on E1 are sums of Gysin maps.

Proposition 5.1.5.

All of the Hodge numbers hs⁢t that appear in Hn⁢(X,ℚ) satisfy 0≤s,t≤n.

Proof.

First, all of the E1−a,b have nonnegative Hodge components. To prove the upper bound, if Hp⁢(Dq(r),ℚ)⁢(−r) contributes to Hn⁢(X,ℚ), then p+q+r=n. In particular, p+r≤n and any hs′⁢t′ appearing in Hp⁢(Dq(r),ℚ) has 0≤s′,t′≤p, so when we twist by (−r), we get s,t≤p+r≤n. □

Proposition 5.1.6.

If X is proper, then all of the weights of Hn⁢(X,ℚ) are at most n.

Proof.

If X is proper, then each Yn is already proper, so each Dn=∅. Therefore, considering the spectral sequences, we only have nonzero terms for r=0, so the spectral sequence becomes

If Hb⁢(Y−a,ℚ) contributes to Hn⁢(X,ℚ), then we must have b≤n. □

5.2 Hodge characteristic

Given a variety X=U⊔Z, then

For complex varieties, χ=χc, but here we use compactly supported cohomology. For example, if X is compact, consider the long exact sequence

and then use the fact that Hcn⁢(U)=Hn⁢(X,Z). Also note that

This tells us that the compactly supported Euler characteristic is motivic and defines a ring homomorphism

Here, K0⁢(𝖵𝖺𝗋) is the Grothendieck ring of varieties with relations [X]=[U]+[Z] and [X×Y]=[X]⁢[Y]. This loses a lot of information, however, so we construct something better.

Theorem 5.2.1.

Let Y⊂X be a locally closed subvariety. There is a natural mixed Hodge structure on each Hn⁢(X,Y,ℤ) respecting the long exact sequence of the pair (X,Y).

Theorem 5.2.2.

The (relative) Künneth theorem respects mixed Hodge structures.

Example 5.2.3.

We would like to construct a mixed Hodge structure on Hcn⁢(𝔸1). We know this is ℚ in degree 2 and vanishes everywhere else. There is an isomorphism of mixed Hodge structures Hc2⁢(𝔸1)≅H2⁢(ℙ1), and thus Hc2⁢(𝔸1) is pure of type (1,1). By the Künneth formula, Hc2⁢n⁢(𝔸n) is pure of type (n,n).

Recall that any Hodge structure V defines an element [Vℚ]∈K0⁢(𝖧𝖲) by taking graded pieces. Now define

Definition 5.2.4.

The Hodge characteristic with compact support is the ring morphism

Definition 5.2.5.

The Hodge-Euler polynomial with compact support is the ring morphism

Remark 5.2.6.

If X is smooth and proper, then the Hodge-Euler polynomial is a generating function for the Hodge numbers of X.

Example 5.2.7.

Let X be a torix variety and let sk be the number of torus orbits of dimension k. Therefore, we know that

By the Künneth formula and the computation for 𝔸1, we have

In the case when X is smooth and proper, the Poincaré polynomial is ∑ksk⁢(t2−1)k.

Example 5.2.8.

Let X be proper irreducible nodal curve with geometric genus g. Assume that X has p nodes. From topology, we know that H0⁢(X,ℚ)=ℚ is pure of type (0,0). Also, H2⁢(X,ℚ)≅ℚ is also pure of type (1,1). It remains to compute the mixed Hodge structure on H1⁢(X,ℚ). This tells us that

and thus we have h00=p,h10=g,h01=g.

6 Caleb (Dec 09): The Ax-Schanuel conjecture for period maps

Note: this lecture was given over the course of two seminars on two consecutive days.

6.1 Polarized Hodge structures

One motivating question is the one of when a complex torus ℂg/Λ is an abelian variety, a Jacobian, etc. When g=1, this is always true because ℂ/Λ is always an elliptic curve by the Weierstrass ℘-function (and is its own Jacobian). When g>1, of course complex tori are not always algebraic.

Let C be a curve of genus g. Then the Jacobian of C is

If δ1,…,δ2⁢g is an integral homology basis, we can send each δi to the map ∫δi−. Now let ω1,…,ωg be a basis of H0⁢(C,ΩC1). Then H1⁢(C,ℤ) embeds as a period matrix

Now normalize ωi so that that the period matrix has the form (InΩ). We want conditions on Ω such such that our complex torus is the Jacobian of a curve. Now on

consider the alternating bilinar form

Given two 1-forms ω,η with period vectors u=(u1,…,u2⁢g) and v=(v1,…,v2⁢g), we can compute

The proof is not completely trivial, but we can take the fundamental polygon of the curve and then use Stokes to write the integral as a sum of products of periods. If ω1,…,ωg is a basis for H0⁢(C,ΩC1), then ωi∧ωj=0 because h2,0=0. If we write P=(InΩ), we see that

This implies that Ω=ΩT. We also see that i⁢Q⁢(ω,ω¯)>0. Therefore the imaginary part of Ω is positive-definite. The two conditions we have obtained are called the Riemann bilinear relations. These are the conditions for having a principal polarization, but of course not all principally polarized abelian varieties are Jacobians.

Given a g×2⁢g matrix B, a polarization is a skew-symmetric form E with B⁢E⁢BT=0 and B⁢E⁢B∗>0.By linear algebra, there exists a diagonal matrix D such that

If D=In, then this is a principal polarization.

Remark 6.1.1.

A polarization on H1⁢(A,ℂ) for A a complex torus gives us an element of H2⁢(A,ℤ). By the Lefschetz (1,1)-theorem, this is c1⁢(L) for some line bundle L. By the positive-definite property of the polarization, L is ample.

6.2 Lefschetz decomposition

Recall that the Hodge decomposition is

By Dolbeault, we know that Hp,q⁢(X)=Hq⁢(X,Ωp). Also, Hp,q=Hq,p¯. If we consider a partition of unity (considering X as a real manifold), we obtain a Hermitian metric h on any compact complex manifold X. Then we know h=g−i⁢ω, where ω is a nondegenerate (1,1)-form. We know X is Kähler if and only if ω is closed.

We know that g induces an L2-metric on forms, so we obtain the Hodge star operator

Now assume that X is compact Kähler.

Definition 6.2.1.

Define the operator L:Hk⁢(X,ℝ)→Hk+2⁢(X,ℝ) by L⁡(α)=α∧[ω].

Theorem 6.2.2 (Hard Lefschetz).

For k≤n, Ln−k:Hk⁢(X,ℝ)→H2⁢n−k⁢(X,ℝ) is an isomorphism.

Definition 6.2.3.

The primitive cohomology of X is defined to be

This terminology is justified by the fact that

This implies that

Remark 6.2.4.

This decomposition works also for complex coefficients because L has degree (1,1) and is thus compatible with the Hodge decomposition.

6.3 Polarizations

Define an integral form on Hk⁢(X,ℝ) for k≤n by

Q takes real values, and if [ω] is integral, it takes integral values on Hk⁢(X,ℤ). Also, Q is nondegenerate, bilinear, and (−1)n-symmetric. Also, Q⁡(Hp,q,Hp′,q′)=0 unless (p,q)=(q′,p′). Extending to Hk⁢(X,ℂ), we have ip−q⁢Q⁢(α,α)>0 for nonzero α∈Hprp,q⁢(X). We know [ω] is integral if and only if X is a smooth projective variety over ℂ by the Kodaira embedding theorem.

Definition 6.3.1.

An integral polarized Hodge structure of weight k is given by a Hodge structure (Vℤ,Fp⁢Vℂ) of weight k and an integer-valued (−1)k-symmetric bilinear form Q on Vℤ such that

  1. 1.

    Q⁡(Hp,q,Hp′,q′)=0 unless (p,q)=(q′,p′);

  2. 2.

    ip−q⁢Q⁢(α,α¯)>0 for all nonzero α of type (p,q).

These are called the Hodge-Riemann bilinear relations. To see why they are true, consider the following:

  1. 1.

    Integrating over X is the same as taking the cap product with [X], and of course [X] has type (n,n). Because L has type (1,1), we need p+p′=q+q′.

  2. 2.

    The Hodge star is given by

    This implies that

6.4 Unpolarized period domains

Let π:X→B be a smooth proper map or projective varieties (or a proper holomorphic submersion).

Theorem 6.4.1 (Ehresmann).

In this case, π is a locally trivial fibration as smooth manifolds. Thus X≅X0×B locally, and the fibers of X→X0 are complex submanifolds. Thus the family of complex structures varies holomorphically.

Consider the sheaf Rk⁢π∗⁢ℂ on B. This is a local system because in the diagram,

X×BU≅X0×U, so Hk⁢(X0,ℂ)≃Hk⁢(Xb,ℂ) for all b∈U. If X0 is Kähler, we may assume U=B and then Hk⁢(X0,ℂ)≃Hk⁢(Xb,ℂ) and the Hodge numbers are the same. However, the Hodge filtration is not the same. If Ak is the k-differential forms, recall that

Then the Hodge filtration is given by

If we write bp,k=dimFp⁢Hk⁢(Xb,ℂ), the period map

is defined by b↦Fp⁢Hk⁢(Xb,ℂ)⊂Hk⁢(Xb,ℂ)≃Hk⁢(X0,ℂ). By a theorem of Griffiths, this map is holomorphic. If we take all p for a fixed k, the period mappings send b to a complete flag of Hk⁢(X,ℂ). Because Xb is Kähler, Hodge symmetry implies that

The open set D⊂Fl⁡(Hk⁢(X,ℂ)) satisfying this condition is the period domain.

6.5 Polarized period domains

Consider X→B. Then a polarized Hodge structure on Hprk⁢(Xb) is given by a bilinear form Q such that

  1. 1.

    Fp⁢Hk⁢(Xb,ℂ)=Fk−p+1⁢Hk⁢(Xb,ℂ)⟂;

  2. 2.

    Hk⁢(Xb,ℂ)=Fp⁢Hk⁢(Xb,ℂ)⊕Fk−p+1⁢Hk⁢(Xb,ℂ)¯;

  3. 3.

    (−1)k⁢(k−1)2⁢ip−q⁢Q⁢(α,α¯) for nonzero α∈Fprp⁢Hk⁢(Xb)∩Fq⁢Hprk⁢(Xb)¯.

Flags satisfying the first condition are the domain Dˇ⊂Fl⁡(Hprk⁢(Xb)) called the compact dual. The flags satisfying all three conditions form an open subset D⊂Dˇ.

Definition 6.5.1.

Define G⁡(R)=Aut⁡(HR,QR), where R=ℤ,ℚ,ℝ,ℂ.

Definition 6.5.2.

A variation of Hodge structures (Vℤ,F∙) of weight k over B consists of a local system Vℤ of finite rank and a finite filtration of Vℂ⊗𝒪B by holomorphic subbundles Fp such that

  • •

    The filtration Fb∙ of (Vℂ)b is a Hodge filtration;

  • •

    (Griffiths transversality) If ∇ is the Gauss-Manin connection of Vℂ⊗𝒪B, then ∇(Fp)⊆Fp−1.

We now have a global period map φ:B→G⁡(ℤ)\D. In the case of X→B, we have b↦[Hprk⁢(Xb)]. If Γ contains the monodromy representation on Hprk⁢(Xb), then φ can be lifted to Γ\D.

Definition 6.5.3.

Let H be a Hodge structure. Then the special Mumford-Tate group M⁢TH is the algebraic ℚ-subgroup of End⁡(Hℚ) such that for any tensor power H′=H⊗k⊗(H∨)⊗ℓ, the rational Hodge classes of H′ are precisely the rational vectors fixed by M⁢TH.

Definition 6.5.4.

The Deligne torus 𝕊 is the group ℂ× considered as a subgroup of G⁢L2⁢(ℝ).

Now let z act on Hp,q by zp⁢z¯q. This is a real representation, so ℝ×⊂𝕊 acts on Hℝ. We also have S1⊂ℂ×, and these give different definitions of the Mumford-Tate group.

Definition 6.5.5.

The Mumford-Tate group is the smallest algebraic group over ℚ whose real points contain the image of S1.

Definition 6.5.6.

The big Mumford-Tate group is the same thing but with R× in place of S1.

Example 6.5.7.

Consider ℚ⁡(n)=(2⁢π⁢i)n⊂ℂ of type (−n,−n). Then the big Mumford-Tate group is 𝔾m if n≠0 and {1} if n=0, and the special Mumford-Tate group is {1}.

Example 6.5.8.

Consider X=Eτ and write V=H1⁢(Eτ,ℚ). Then the special Mumford-Tate group is ℚ⁡(τ) if Eτ has complex multiplication and S⁢L2⁢(ℚ) otherwise.

Definition 6.5.9.

Let D be the polarized period domain and suppose x∈D.

  1. 1.

    The orbit M⁢Tx⁢(ℝ)⁢x is called a Mumford-Tate subdomain.

  2. 2.

    If M is a normal subgroup of M⁢Tx, then M⁡(ℝ)⁢x is called a weak Mumford-Tate subdomain.

  3. 3.

    Let π:D→Γ\D be a quotient map. For D′⊂D a weak Mumford-Tate subdomain, then π⁡(D′)⊂Γ\D is a complex analytic subvariety, called a weak Mumford-Tate subvariety. Also, given a period map φ:X→Γ\D, we call φ−1⁢π⁢(D′) also a weak Mumford-Tate subvariety.

Now let X/ℂ be smooth. Write M⁢THℤ=M⁢TX. Let Γ be the image of the monodromy π1⁢(X)→M⁢THℤ⁢(ℚ). Let G be the identity component of the ℚ-Zariski closure of Γ and D=D⁡(G) be the associated weak Mumford-Tate subdomain, φ:X→Γ\D is the period map and Dˇ be the compact dual of D. Now consider

Theorem 6.5.10 (Ax-Schanuel).

Let V⊂X×Dˇ be an algebraic subvariety and U be an irreducible analytic component of U∩XD such that

Then the projection of U→X is contained in a proper weak Mumford-Tate subvariety.

6.6 o-minimality

Definition 6.6.1.

A structure S is a collection of subsets {Sn}n∈ℕ of ℝn such that

  1. 1.

    Sn is closed under finite intersections, unions, and complements.

  2. 2.

    The collection (Sn) is closed under finite cartesian products and projection.

  3. 3.

    For every polynomial P∈ℝ⁡[x1,…,xn], the set of roots of P is an element of Sn.

Note that if P⁡(x1,…,xn) is a polynomial, the vanishing sets of P⁡(x1,…,xn)−xn+12 are in Sn+1, and now projecting down, we see that {P⁡(x1,…,xn)≥0}∈Sn. We call the elements U∈Sn the S-definable subsets of ℝn. For U∈Sn,V∈Sm, a map f:U→V is definable if its graph is.

Example 6.6.2.

Let ℝalg be the collection of real semialgebraic subsets of ℝn (the Boolean algebra generated by {P≥0}).

Definition 6.6.3.

A structure S is o-minimal if S1=(ℝalg)1 is the set of finite unions of intervals.

Example 6.6.4.

The o-minimal structure ℝexp is the structure generated by the graph of exp:ℝ→ℝ.

Example 6.6.5.

The structure ℝsin is not o-minimal.

Example 6.6.6.

The structure ℝan generated by graphs of restrictions of real analytic functions to open balls B⁡(R) is o-minimal.

Example 6.6.7.

The structure ℝan,exp generated by ℝan,ℝexp is o-minimal.

Definition 6.6.8.

An S-definable topological space M i a topological space M with a finite covering {Vi} of M and homeomorphisms φi:V−i→Ui⊂ℝn such that

  1. 1.

    All Ui and all intersections Ui⁢j are S-definable.

  2. 2.

    The transition functions φi⁢j:Ui⁢j→Uj⁢i are S-definable.

We can also define S-definable continuous maps, but it will not be helpful to write down this definition.

Example 6.6.9.

Consider ℂ∗⊂ℂ and let e:ℂ→ℂ∗ be the covering map. Then we want to put an ℝalg-definable structure on ℂ∗. We can do it in more than one way:

  1. 1.

    Simply take ℂ∗⊂ℂ=ℝ2 and induce the structure from ℂ∗.

  2. 2.

    Define Fa={z∈ℂ∣a⋅Imz⊂Rez⊂(1+ε)+a⁢Imz}. Under the map e, this covers ℂ∗. Taking thinner strips, we obtain an open covering of ℂ∗, and we obtain a different structure for each a.

Example 6.6.10.

Let X be a real variety. By covering X with affines, the induced ℝalg-structures on affines give us an ℝalg-structure on X.

Definition 6.6.11.

Let X be a complex variety. Let Xdef be the S-definable topological space with underlying set X⁡(ℂ) and then use the previous example.

We will now discuss applications of o-minimality. The first is a counting theorem of Pila-Wilkie. We will begin by recalling the elementary theory of heights that we should have learned as undergrads (but didn’t).

Definition 6.6.12.

For r∈ℚ, define H⁡(r)=max⁡{|a|,|b|} where r=ab is reduced. Then for α∈ℚn, write H⁡(α)=max⁡H⁡(αi).

Note that there are only finitely many points of ℚn with bounded height. Now let U⊂ℝn be a subset. Define N⁡(U,t)=#⁢{α∈U∩ℚn∣H⁡(α)≤t}. Then let

and Utr=U∖Ualg.

Example 6.6.13.

Consider Z=ex⁢y. Then Ualg is given by y=z=0.

Theorem 6.6.14.

Let U⊂ℝn be definable for some o-minimal structure. Then for any ε>0, N⁡(Utr,t)=O⁡(tε).

This result shows that most ℚ-rational points are algebraic. The next application is the Definable Chow theorem of Peterzil-Starchenko. Recall Chow’s theorem, which says that if X is a proper complex variety and Y⊂Xan is a closed complex analytic subvariety, then Y is an algebraic subvariety.

Theorem 6.6.15 (Definable Chow).

Fix an o-minimal structure and let X be any complex variety. Let Y be a closed complex analytic subvariety whose underlying set is definable. Then Y is algebraic.